review a bevy of new desktop processors arrives

A bevy of new desktop processors arrives

So far, 2009 has been a year of renewed competition in desktop processors, as a resurgent AMD has rolled out portions of its 45nm CPU lineup with reasonable success. For its part, Intel has responded by dropping prices on its own Core 2 processors to remain competitive, while riding high on the Core i7’s undisputed performance leadership at the top of the market.

Tighter competition means better choices for consumers, but in technology, it almost always means more choices, as well. That principle was on ample display last week as AMD introduced several new products that essentially complete its transition to 45nm technology. Meanwhile, Intel has countered by quietly freshening up its low-end and mid-range offerings a bit and trumpeting the release of a new flagship CPU, the Core i7-975 Extreme.

With these moves, you have, uh, five.. no, four… wait, maybe it’s six different strata of desktop CPUs from which to choose. Not counting the low-power versions. We have new processors to test that fit into… uh, I think four of those categories.

Marketing is hard, folks. Especially when you’re on the receiving end of it.

Anyhow, there’s much ground to cover. Given the scope of the new releases, we decided to compare June’s crop of CPUs against, well, everything we could swing. The result is an enormous roundup of 26 different types of processors, including five new ones. We’ve poked, prodded, tested for performance, measured power efficiency, overclocked, and considered the value propositions. The only question now is whether I’ll pass out before I’m able to finish writing this thing. Should make for interesting times, no?

The new entries, from bottom to top
One of the least expensive chips on our agenda is the most novel. AMD has, at last, produced a native dual-core processor based on its latest technology, and the Athlon II X2 250 is the first incarnation of it. This chip features the same execution core and feature set as the Phenom II, but unlike other recent Athlon X2-branded products, it is not based on a gimpy quad-core chip. Instead, this is a true dual-core, 45nm processor with 1MB of L2 cache per core—and no L3 cache at all. The chip itself is made up of “only” 234 million transistors and fits into a die area of 117.5 mm²—well under half the size of the a Phenom II, by both measurements. Yet this is a true Socket AM3 processor, with support for dual channels of DDR3 memory, HyperTransport 3 speeds of up to 4 GT/s, and backward compatibility with Socket AM2+ motherboards and DDR2 memory.

One would, of course, expect this silicon to become a very popular choice for some high-profile missions beyond the desktop, including a range of laptops and sub-notebooks of various designations.

On the desktop, the Athlon II X2 250 looks like a decent “value dual-core” CPU option, with a core clock speed of 3GHz, a relatively tame 65W TDP rating, and a price tag of just $87. Officially, the Athlon II X2 250 supports DDR2 memory at up to 800MHz and DDR3 memory at up to 1066MHz, but it worked just fine for us with 1333MHz DDR3 memory. Since its new-generation CPU cores should achieve higher clock-for-clock performance than older Athlon processors, and since it starts with a relatively high clock frequency, the X2 250 could be a very nice value in this part of the market.

Gigabyte’s MA770T-UD3P

To help illustrate that point, Gigabyte sent us a motherboard it recommends pairing with one of AMD’s new dual-core processors, the MA770T-UD3P. Unfortunately, it arrived in Damage Labs too late for our use in testing, but this compact AMD 770 chipset-based Socket AM3 board offers pretty much everything you’d need for a decent system, along with DDR3 support. The price? 89 bucks, American money. In fact, some places online are selling it for even less.

Choosing the parts for the Econobox in our next system guide just got more interesting, I think.

Should a mobo price like that one free up a few dollars in your CPU budget, you might wish to step up to a slightly faster processor. The Phenom II X2 550 Black Edition is intended to fill that role. Pretty straightforwardly, this one really is a Phenom II with two cores disabled. Each remaining core has 512KB of L2 cache, and the two cores share a common 6MB L3 cache. The X2 550 runs at 3.1GHz, with 2GHz HyperTransport and north bridge/L3 cache speeds, and it has a TDP rating of 80W. As AMD’s top dual-core product, the X2 550 is also a Black Edition, which means its clock multiplier is unlocked to facilitate easier overclocking. The Phenom II X2 550 lists for $102, so for 15 bucks more than the Athlon II X2 250, you get the L3 cache, another hundred megahertz, and an unlocked multiplier.

The final piece of the 45nm puzzle for AMD is a pair of low-power Phenom II processors also introduced last week. The Phenom II X4 905e ticks away at 2.5GHz, has 6MB of L3 cache, is rated for a 65W TDP, and will set you back $195. If you’re willing to drop from four cores to three, the Phenom II X3 705 offers the same basic specs for just $125. We’ve not yet had a chance to test these power-efficient processors, which will face off against some low-power Core 2 Quads from Intel, but we hope to do so soon, so stay tuned.

The natural counter to the Athlon II X2 would come from Intel’s Pentium E series of value dual-core processors. As if in anticipation of the Athlon II, Intel very discreetly let the Pentium E6300 slip into the wild last month. This chip is based on the 45nm “Wolfdale” Core 2 Duo silicon, but with only 2MB of its L2 cache enabled. The rest of the vitals: 2.8GHz core clock, 1066MHz front-side bus, and a 65W TDP. The E6300 is priced just opposite the Athlon II X2 250 at $84. At that price, the E6300 should be a formidable competitor, despite the fact that its name tends to engender confusion with the (quite different) 65nm Core 2 Duo E6300.

The most direct competition for the Phenom II X2 550 would probably be the Core 2 Duo E7400, which is the same thing as the Pentium E6300, only with 3MB of L2 cache instead of 2MB. We unfortunately don’t have an E7400 on hand for testing, but it should be only slightly quicker than the Pentium E6300, which is surely the better value.

We do have a Core 2 Quad Q8400 on hand, though, another product Intel released rather quietly recently. We have not been big proponents of past “value quad-core” processors, including the Core 2 Quad Q8200 and the Phenom II X4 810, because their modest clock speeds limit performance in lightly threaded applications, including games. The Q8400 aims to remedy this situation somewhat by bumping core clocks up to 2.66GHz. Otherwise, the Q8400 mirrors its siblings with a 1333MHz front-side bus, 4MB total L2 (2MB per chip), and a 95W TDP. The $183 Q8400’s closest competitor is probably the Phenom II X4 940, which lists for $195. The Q8400 steers well clear of the Core i7-920 at $284, so it’s positioned nicely as an affordable quad-core option.

Intel has also released a low-power version of this product, dubbed the Q8400S, with a 65W TDP rating for $245.

The CPU hogging all of the attention, though, is Intel’s new flagship, the Core i7-975 Extreme. This puppy steps directly into the role of “fastest desktop processor on the planet” courtesy of its quad-core Nehalem architecture and 3.33GHz core clock speed.

Well, clock speed is a tricky thing with a Core i7, thanks to its Turbo mode dynamic clock scaling. In reality, the Core i7-975 Extreme will spend much of its time above 3.33GHz, at up to 3.6GHz in single-threaded applications or 3.46GHz with multiple threads, depending on thermal headroom.

Like the Core i7-965 before it, the 975 has a QPI link speed of 6.4 GT/s.

The 975 Extreme is based on a new D-stepping of Nehalem silicon, which brings additional newness and possibly additional goodness in the form of higher clock speeds at lower voltages, if the rumors are true. Some extra headroom might be useful, since the 975 is an Extreme Edition with an unlocked upper multiplier. The Core i7-975 is indeed extreme, too, with a 130W TDP and a sticker price of $999. As you may know, AMD has nothing yet to compete with the Core i7-975 Extreme, although some interesting possibilities do suggest themselves, don’t you think?

If one dollar short of a grand is a little too rich for your blood, you might instead be interested in the Core i7-950, another part of June’s bumper CPU crop. With a 3GHz core clock and a 4.8 GT/s QPI link speed, the Core i7-950 essentially replaces the 2.93GHz Core i7-940. Both occupy the same $562 slot in Intel’s price sheet, which suggests the Core i7-940 isn’t long for this world. The i7-950 should be a minor step up in performance, but obviously not terribly different, which is why we didn’t bother testing this speed grade separately.

Test notes

Pictured above is a trio of DIMMs from OCZ that are intended for use with the Core i7-975 Extreme. These 2GB Blade series DDR3 modules are rated for operation at a blistering 2133MHz. Interestingly enough, the 975 Extreme presents the correct multipliers for both 1866MHz and 2133MHz memory via the BIOS of our Gigabyte EX58-UDR3 motherboard, a TR Recommended mobo we’ve selected for our new Core i7 testbed. This board is one of the newer sub-$200 X58 motherboards that offers better overclocking options than the Intel board we’ve used previously.

We did test the Core i7-975 Extreme with the Blade modules, but we used, ahem, a more pedestrian memory speed of “only” 1600MHz in our main testing, though with relatively tight timings. We found that we had to use looser timings to achieve higher memory clocks, and the tradeoff wasn’t always worth it. We continue to tinker, though.

Speaking of which, in order to gauge the impact of memory type on performance and power use, we’ve tested the Phenom II X4 810 both with DDR2 memory on a Socket AM2+ board and with DDR3 memory on a Socket AM3 board. You’ll find the results in the following pages, labeled appropriately.

The Core 2 Quad Q8300 processor we used for testing came to us courtesy of the good folks at NCIX and NCIXUS. Thanks to them for making this comparison possible. We’ve underclocked our Q8300 to simulate a Q8200 for this review.

We’ve simulated several other speed grades via underclocking, too. Specifically, the Phenom II X4 920 is an underclocked 940, and the Core 2 Quad Q9550 is an underclocked Core 2 Extreme QX9650. We expect the performance of these “simulated” speed grades to be identical to the real things, but we sometimes exclude these processors from our power consumption testing because we do anticipate power use could vary slightly from the actual products.

Our testing methods
As ever, we did our best to deliver clean benchmark numbers. Tests were run at least three times, and the results were averaged.

Our test systems were configured like so:

Processor Core
2 Quad Q6600
2.4 GHz
2 Duo E8400
3.00 GHz
2 Duo E8600
3.33 GHz
Core 2 Quad Q8200 2.33 GHz
Core 2 Quad Q9300 2.5 GHz
Core 2 Quad Q9400 2.66 GHz
Core 2 Quad Q9550 2.83 GHz
2 Extreme QX9770 3.2 GHz
2 Extreme QX9775 3.2 GHz
i7-940 2.66 GHz
Core i7-940 2.93 GHz
Extreme 3.2 GHz
Extreme 3.33 GHz
64 X2 6400+
3.2 GHz
X3 8750
2.4 GHz

Phenom II X4 920
2.8 GHz
Phenom II X4 940
3.0 GHz
II X4 810
2.6 GHz

Phenom X4 9950
Black 2.6 GHz
II X3 720
2.8 GHz
Phenom II X4 810
2.6 GHz
II X4 955
3.2 GHz
E6300 2.8 GHz
2 Quad Q8400 2.66 GHz
II X2 250 3.0 GHz
II X2 550 3.1GHz
System bus 1066
(266 MHz)
(333 MHz)
(400 MHz)
(400 MHz)
4.8 GT/s
(2.4 GHz)
6.4 GT/s
(3.2 GHz)
6.4 GT/s
(3.2 GHz)
2.0 GT/s
(1.0 GHz)
3.6 GT/s (1.8 GHz)
3.6 GT/s (1.8 GHz)
4.0 GT/s (2.0 GHz)
4.0 GT/s (2.0 GHz)
4.0 GT/s (2.0 GHz)
Motherboard Asus
P5E3 Premium
P5E3 Premium
P5E3 Premium
M3A79-T Deluxe
M3A79-T Deluxe
DKA790GX Platinum
M4A79T Deluxe
BIOS revision 0605 0605 0605 XS54010J.86A.1149.
F5 0403 0403 11/25/08 0703
0802 0802 0089
North bridge X48
Express MCH
Express MCH
Express MCH
790FX 790FX 790GX 790FX
South bridge ICH9R ICH9R ICH9R 6321ESB ICH ICH10R ICH10R ICH10R SB750 SB750 SB750 SB750
Chipset drivers INF

Matrix Storage Manager


Matrix Storage Manager


Matrix Storage Manager

INF Update

Matrix Storage Manager

Matrix Storage Manager
Matrix Storage Manager
Matrix Storage Manager
controller 3.1.1540.61
controller 3.1.1540.61
controller 3.1.1540.61
controller 3.1.1540.61
Memory size 4GB
(2 DIMMs)
(2 DIMMs)
(2 DIMMs)
(2 DIMMs)
(3 DIMMs)
(3 DIMMs)
(3 DIMMs)
(2 DIMMs)
(2 DIMMs)
(2 DIMMs)
(2 DIMMs)
Memory type Corsair
ECC DDR2-800
speed (Effective)
CAS latency (CL) 7 8 8 5 7 8 8 4 5 5 8
RAS to CAS delay (tRCD) 7 8 8 5 7 8 8 4 5 5 8
RAS precharge (tRP) 7 8 8 5 7 8 8 4 5 5 8
Cycle time (tRAS) 20 20 24 18 20 24 24 12 15 15 20
2T 2T 2T 2T 2T 1T 1T 2T 2T 2T 2T
Audio Integrated
with SoundMAX drivers
with SoundMAX drivers
with SoundMAX drivers
with SigmaTel 6.10.5713.7 drivers
with Realtek drivers
with Realtek drivers
with Realtek drivers
with SoundMAX drivers
with SoundMAX drivers
with Realtek drivers
with Realtek drivers
Hard drive WD Caviar SE16 320GB SATA
Graphics Radeon
HD 4870 512MB PCIe with Catalyst 8.55.4-081009a-070794E-ATI
OS Windows Vista Ultimate x64 Edition
OS updates Service
Pack 1, DirectX redist update August 2008

Thanks to Corsair and OCZ for providing us with memory for our testing.

Our single-socket test systems were powered by OCZ GameXStream 700W power supply units. The dual-socket system was powered by a PC Power & Cooling Turbo-Cool 1KW-SR power supply. Thanks to OCZ for providing these units for our use in testing.

Also, the folks at hooked us up with a nice deal on the WD Caviar SE16 drives used in our test rigs. NCIX now sells to U.S. customers, so check them out.

The test systems’ Windows desktops were set at 1600×1200 in 32-bit color at an 85Hz screen refresh rate. Vertical refresh sync (vsync) was disabled.

We used the following versions of our test applications:

The tests and methods we employ are usually publicly available and reproducible. If you have questions about our methods, hit our forums to talk with us about them.

Memory subsystem performance

This test gives us a visual representation of the cache and memory subsystems of these CPUs. I’ve limited the results to the five new processors we’re testing, plus one more for comparison. As you can see, the higher-end CPUs tend to have larger, faster caches, as one would expect. The performance of the L2 and L3 caches of the Core i7-975 is truly remarkable. Meanwhile, the only real difference between the Core 2 Quad Q8400 and Q9400 is illustrated at the 4MB block size, where the Q9400’s larger L2 caches come into play.

Since it’s difficult to see the results once we get into main memory, let’s take a closer look at the 256MB block:

The Core 2 processors’ throughput is generally limited by their front-side bus speeds, while the AMD processors and the Core i7s have no such bottleneck.

The front-side bus also adds a latency penalty to memory accesses, which is why the Core 2 processors and the Pentium E6300 cluster near the bottom of these results. Interestingly, with no L3 cache onboard, the Athlon II X2 250 is a little quicker getting out to memory than the Phenom IIs. Again, though, the Core i7 is just an absolute monster, with the fastest memory subsystem by far in every way we’ve measured.

Crysis Warhead
We measured Warhead performance using the FRAPS frame-rate recording tool and playing over the same 60-second section of the game five times on each processor. This method has the advantage of simulating real gameplay quite closely, but it comes at the expense of precise repeatability. We believe five sample sessions are sufficient to get reasonably consistent results. In addition to average frame rates, we’ve included the low frame rates, because those tend to reflect the user experience in performance-critical situations. In order to diminish the effect of outliers, we’ve reported the median of the five low frame rates we encountered.

We tested at at relatively modest graphics settings, 1024×768 resolution with the game’s “Mainstream” quality settings, because we didn’t want our graphics card to be the performance-limiting factor. This is, after all, a CPU test.

Right out of the gate, Warhead demonstrates that one need not buy a fancy quad-core processor in order to play even the latest games. In fact, you’re better off with a high-frequency dual-core than you are with a slower quad, as testified by the fact that the Core 2 Duo E8600 outperforms the Core i7-940 here. Similarly, the Phenom II X2 550 essentially ties the Phenom II X4 955.

Warhead does appear to be sensitive to cache sizes, though, judging by the fact that the Athlon II X2 250 falls well behind the Phenom II X2 550. In cases like this one, paying a little more for the extra cache helps. But then that’s a pretty large difference in total effective cache size: 2304KB for the Athlon II X2 250 versus 7424KB for the Phenom II X2 550. The difference between the Core 2 Quad Q8400 and Q9400 is smaller—4096KB vs. 6144KB, respectively, of total effective cache—and doesn’t appear to cross any major dividing lines: only two frames per second separate their averages.

Keeping score: Phenom II X4 940 over Q8400, the Pentium E6300 and Athlon II X2 250 essentially tie, and the Core i7-975 Extreme obliterates everything.

Far Cry 2
After playing around with Far Cry 2, I decided to test it a little bit differently by recording frame rates during the jeep ride sequence at the very beginning of the game. I found that frame rates during this sequence were generally similar to those when running around elsewhere in the game, and after all, playing Far Cry 2 involves quite a bit of driving around. Since this sequence was repeatable, I just captured results from three 90-second sessions.

Again, I didn’t want the graphics card to be our primary performance constraint, so although I tested at fairly high visual quality levels, I used a relatively low 1024×768 display resolution and DirectX 9.

The minimum frame rates here are a little higher than in Warhead, generally speaking, which means nearly any of these processors should play this game pretty smoothly. Even the Pentium E6300’s minimum frame rate is 30 FPS.

Unreal Tournament 3
As you saw on the preceding page, I did manage to find a couple of CPU-limited games to use in testing. I decided to try to concoct another interesting scenario by setting up a 24-player CTF game on UT3’s epic Facing Worlds map, in which I was the only human player. The rest? Bots controlled by the CPU. I racked up frags like mad while capturing five 60-second gameplay sessions for each processor.

Oh, and the screen resolution was set to 1280×1024 for testing, with UT3’s default quality options and “framerate smoothing” disabled.

Although the frame rates involved here clearly show that even the slowest processor is very much up to this task, this is one of the rare situations in current games where we can exploit more than two cores to gain additional performance. As a result, the dual-core CPUs congregate in the bottom third of the chart, along with the rather troubled original Phenoms.

Half Life 2: Episode Two
Our next test is a good, old custom-recorded in-game timedemo, precisely repeatable.

This is a very strong showing from the Phenom II X2 550, nearly in league with the Core 2 Duo E8400—and, notably, faster than the Core i7-940, amazingly enough.

Source engine particle simulation
Next up is a test we picked up during a visit to Valve Software, the developers of the Half-Life games. They had been working to incorporate support for multi-core processors into their Source game engine, and they cooked up some benchmarks to demonstrate the benefits of multithreading.

This test runs a particle simulation inside of the Source engine. Most games today use particle systems to create effects like smoke, steam, and fire, but the realism and interactivity of those effects are limited by the available computing horsepower. Valve’s particle system distributes the load across multiple CPU cores.

There will be no upstart dual-cores taking the lead here. This one is about core counts, instructions per clock, and clock speed. The Q8400 nearly ties the Q9400 in this test, and I believe this is the first time we’ve seen the Q8400 finish ahead of the Phenom II X4 940. Need I even say that the Core i7-975 Extreme is at the top of the heap?

WorldBench’s overall score is a pretty decent indication of general-use performance for desktop computers. This benchmark uses scripting to step through a series of tasks in common Windows applications and then produces an overall score for comparison. WorldBench also records individual results for its component application tests, allowing us to compare performance in each. We’ll look at the overall score, and then we’ll show individual application results alongside the results from some of our own application tests.

So how in the world did the Core i7-975 Extreme finish behind the 965? Simple: for whatever reason, our new Core i7 test rig didn’t play well with the Nero 7 Ultra test. The application would hang on each attempt through the test. As a result, we couldn’t include results for Nero in the WorldBench composite score for the 975 Extreme, which held it back. The Nero test has long been a sticking point in the WorldBench suite, although it usually runs sucessfully for us after multiple tries. Hopefully we can overcome this issue next time around.

Beyond that, the general strength of the Intel processors is apparent here, as the blue clusters in the top half of the results and the green in the bottom half. The Pentium E6300 outdoes the Phenom II X2 550 by a single point, and the Q8400 scores another win over the Phenom II X4 940.

Productivity and general use software

MS Office productivity

Firefox web browsing

This very common PC activity emphasizes fast dual-cores over slower quads in fairly dramatic fashion. Hence the very strong showing the for Phenom II X2 550—with its dual cores, large cache, and high clock speeds—and the weak placement of the Core 2 Quad Q8400, with its quad cores, smaller caches, and relatively modest frequencies. I understand this test is also fairly sensitive to memory access latencies, which helps explain why the Athlon II X2 250 so dramatically outruns the Pentium E6300.

Multitasking – Firefox and Windows Media Encoder

WinZip file compression

The new dual-core AMD processors’ relatively strong performance here is a bit puzzling, but it seems this test just doesn’t play well with AMD’s quad-core processors. Ye olde Athlon 64 X2 6400+ outperforms all of the AMD quad-core CPUs, as well.

Nero CD authoring

These results are pretty clearly stratified by disk controller type.

Image processing


AMD processors haven’t performed well in WorldBench’s Photoshop test, and as we’ve noted before, some of the blame apparently lays at the feet of AMD’s disk controller. Between this test and Nero, AMD comes by its poor rankings in WorldBench rather honestly. One suspects, though, that with a better disk controller, the Phenoms and Athlons would put in a better showing.

The Panorama Factory photo stitching
The Panorama Factory handles an increasingly popular image processing task: joining together multiple images to create a wide-aspect panorama. This task can require lots of memory and can be computationally intensive, so The Panorama Factory comes in a 64-bit version that’s widely multithreaded. I asked it to join four pictures, each eight megapixels, into a glorious panorama of the interior of Damage Labs.

In the past, we’ve added up the time taken by all of the different elements of the panorama creation wizard and reported that number, along with detailed results for each operation. However, doing so is incredibly data-input-intensive, and the process tends to be dominated by a single, long operation: the stitch. So this time around, we’ve simply decided to report the stitch time, which saves us a lot of work and still gets at the heart of the matter.

The Intel CPUs come out on top here in every category: E6300 over both new dual-core AMDs, Q8400 over Phenom II X4 940, and Core i7-975 Extreme uber alles.

picCOLOR image analysis
picCOLOR was created by Dr. Reinert H. G. Müller of the FIBUS Institute. This isn’t Photoshop; picCOLOR’s image analysis capabilities can be used for scientific applications like particle flow analysis. Dr. Müller has supplied us with new revisions of his program for some time now, all the while optimizing picCOLOR for new advances in CPU technology, including MMX, SSE2, and Hyper-Threading. Naturally, he’s ported picCOLOR to 64 bits, so we can test performance with the x86-64 ISA. Many of the individual functions that make up the test are multithreaded.

This final image manipulation test is much closer than the others above. The Intel and AMD products in the same basic categories are pretty evenly matched.

Media encoding and editing

x264 HD benchmark
This benchmark tests performance with one of the most popular H.264 video encoders, the open-source x264. The results come in two parts, for the two passes the encoder makes through the video file. I’ve chosen to report them separately, since that’s typically how the results are reported in the public database of results for this benchmark. These scores come from the newer, faster version 0.59.819 of the x264 executable.

If you want to encode video quickly, a cheap dual-core processor definitely isn’t the way to go. You’re vastly better off with a Core 2 Quad Q8400 or its rival, the Phenom II X4 940, which is a little quicker in both tests. Of course, with price as no object, the best option is the Core i7-975 Extreme—or perhaps more than one CPU. If you want to see something really cool, have a look at the dual-socket Nehalem systems reaching into the 50 to 60 FPS range in the more broadly multithreaded second pass of this process.

Windows Media Encoder x64 Edition video encoding
Windows Media Encoder is one of the few popular video encoding tools that uses four threads to take advantage of quad-core systems, and it comes in a 64-bit version. Unfortunately, it doesn’t appear to use more than four threads, even on an eight-core system. For this test, I asked Windows Media Encoder to transcode a 153MB 1080-line widescreen video into a 720-line WMV using its built-in DVD/Hardware profile. Because the default “High definition quality audio” codec threw some errors in Windows Vista, I instead used the “Multichannel audio” codec. Both audio codecs have a variable bitrate peak of 192Kbps.

Among the value dual-cores and the mid-range quad cores, the AMD processors prove to be a little faster here.

Windows Media Encoder video encoding

Roxio VideoWave Movie Creator

I’m not a big fan of WorldBench’s video encoding tests, with aren’t as multithreaded as real video encoding apps tend to be these days. I’ve included them for the sake of completeness.

LAME MT audio encoding
LAME MT is a multithreaded version of the LAME MP3 encoder. LAME MT was created as a demonstration of the benefits of multithreading specifically on a Hyper-Threaded CPU like the Pentium 4. Of course, multithreading works even better on multi-core processors. You can download a paper (in Word format) describing the programming effort.

Rather than run multiple parallel threads, LAME MT runs the MP3 encoder’s psycho-acoustic analysis function on a separate thread from the rest of the encoder using simple linear pipelining. That is, the psycho-acoustic analysis happens one frame ahead of everything else, and its results are buffered for later use by the second thread. That means this test won’t really use more than two CPU cores.

We have results for two different 64-bit versions of LAME MT from different compilers, one from Microsoft and one from Intel, doing two different types of encoding, variable bit rate and constant bit rate. We are encoding a massive 10-minute, 6-second 101MB WAV file here.

The performance differences between the CPUs here are pretty minor. This is, though, one more example where fewer cores and higher clock speeds win out.

3D modeling and rendering

Cinebench rendering
Graphics is a classic example of a computing problem that’s easily parallelizable, so it’s no surprise that we can exploit a multi-core processor with a 3D rendering app. Cinebench is the first of those we’ll try, a benchmark based on Maxon’s Cinema 4D rendering engine. It’s multithreaded and comes with a 64-bit executable. This test runs with just a single thread and then with as many threads as CPU cores (or threads, in CPUs with multiple hardware threads per core) are available.

The Core i7’s dominance here is staggering. I won’t belabor the point, but do have a look at the single-threaded results, where multiple cores and Hyper-Threading are no help at all. The Core i7-975 Extreme still leads the field by a considerable margin, in part because its Turbo mode mechanism allows it to run at 3.6GHz while rendering with a single thread. We’ve been talking some about the value trade-off between fewer, faster cores and more, slower cores in the context of the other processors. Thanks to Turbo mode, the Core i7 takes the edge off of that trade-off.

POV-Ray rendering
We’re using the latest beta version of POV-Ray 3.7 that includes native multithreading and 64-bit support. Some of the beta 64-bit executables have been quite a bit slower than the 3.6 release, but this should give us a decent look at comparative performance, regardless.

3ds max modeling and rendering

Valve VRAD map compilation
This next test processes a map from Half-Life 2 using Valve’s VRAD lighting tool. Valve uses VRAD to pre-compute lighting that goes into games like Half-Life 2.

Some trends emerge from our remaining 3D rendering tests. Among them: of course, having more cores is good. But also look at cache size. The Q8400 and Q9400 shadow one another, with very little daylight between them, and the Phenom II X2 550 barely stays ahead of the Athlon X2 550, with the only difference likely due to its 100MHz clock speed advantage.

[email protected]
Next, we have a slick little [email protected] benchmark CD created by notfred, one of the members of Team TR, our excellent Folding team. For the unfamiliar, [email protected] is a distributed computing project created by folks at Stanford University that investigates how proteins work in the human body, in an attempt to better understand diseases like Parkinson’s, Alzheimer’s, and cystic fibrosis. It’s a great way to use your PC’s spare CPU cycles to help advance medical research. I’d encourage you to visit our distributed computing forum and consider joining our team if you haven’t already joined one.

The [email protected] project uses a number of highly optimized routines to process different types of work units from Stanford’s research projects. The Gromacs core, for instance, uses SSE on Intel processors, 3DNow! on AMD processors, and Altivec on PowerPCs. Overall, [email protected] should be a great example of real-world scientific computing.

notfred’s Folding Benchmark CD tests the most common work unit types and estimates performance in terms of the points per day that a CPU could earn for a Folding team member. The CD itself is a bootable ISO. The CD boots into Linux, detects the system’s processors and Ethernet adapters, picks up an IP address, and downloads the latest versions of the Folding execution cores from Stanford. It then processes a sample work unit of each type.

On a system with two CPU cores, for instance, the CD spins off a Tinker WU on core 1 and an Amber WU on core 2. When either of those WUs are finished, the benchmark moves on to additional WU types, always keeping both cores occupied with some sort of calculation. Should the benchmark run out of new WUs to test, it simply processes another WU in order to prevent any of the cores from going idle as the others finish. Once all four of the WU types have been tested, the benchmark averages the points per day among them. That points-per-day average is then multiplied by the number of cores on the CPU in order to estimate the total number of points per day that CPU might achieve.

This may be a somewhat quirky method of estimating overall performance, but my sense is that it generally ought to work. We’ve discussed some potential reservations about how it works here, for those who are interested. I have included results for each of the individual WU types below, so you can see how the different CPUs perform on each.

This one splits interestingly among product categories. At the low end, the Pentium E6300 outperforms the two value duallies from AMD. In the mid-range, the Phenom II X4 940 ends up just ahead of the Q8400, but it’s a pretty much a tie for practical purposes. And after a little turmoil due to low scores in each individual WU type, since the CPU is coping with two threads per core, the Core i7-975 Extreme only barely establishes its supremacy over its predecessor, the 965 Extreme.

MyriMatch proteomics
Our benchmarks sometimes come from unexpected places, and such is the case with this one. David Tabb is a friend of mine from high school and a long-time TR reader. He has provided us with an intriguing new benchmark based on an application he’s developed for use in his research work. The application is called MyriMatch, and it’s intended for use in proteomics, or the large-scale study of protein. I’ll stop right here and let him explain what MyriMatch does:

In shotgun proteomics, researchers digest complex mixtures of proteins into peptides, separate them by liquid chromatography, and analyze them by tandem mass spectrometers. This creates data sets containing tens of thousands of spectra that can be identified to peptide sequences drawn from the known genomes for most lab organisms. The first software for this purpose was Sequest, created by John Yates and Jimmy Eng at the University of Washington. Recently, David Tabb and Matthew Chambers at Vanderbilt University developed MyriMatch, an algorithm that can exploit multiple cores and multiple computers for this matching. Source code and binaries of MyriMatch are publicly available.

In this test, 5555 tandem mass spectra from a Thermo LTQ mass spectrometer are identified to peptides generated from the 6714 proteins of S. cerevisiae (baker’s yeast). The data set was provided by Andy Link at Vanderbilt University. The FASTA protein sequence database was provided by the Saccharomyces Genome Database.

MyriMatch uses threading to accelerate the handling of protein sequences. The database (read into memory) is separated into a number of jobs, typically the number of threads multiplied by 10. If four threads are used in the above database, for example, each job consists of 168 protein sequences (1/40th of the database). When a thread finishes handling all proteins in the current job, it accepts another job from the queue. This technique is intended to minimize synchronization overhead between threads and minimize CPU idle time.

The most important news for us is that MyriMatch is a widely multithreaded real-world application that we can use with a relevant data set. MyriMatch also offers control over the number of threads used, so we’ve tested with one to eight threads.

I should mention that performance scaling in MyriMatch tends to be limited by several factors, including memory bandwidth, as David explains:

Inefficiencies in scaling occur from a variety of sources. First, each thread is comparing to a common collection of tandem mass spectra in memory. Although most peptides will be compared to different spectra within the collection, sometimes multiple threads attempt to compare to the same spectra simultaneously, necessitating a mutex mechanism for each spectrum. Second, the number of spectra in memory far exceeds the capacity of processor caches, and so the memory controller gets a fair workout during execution.

Here’s how the processors performed.

Make of these rather gratuitously complex results what you will. I’ll just note that the Core i7-975 Extreme finishes in under half time it takes the Core 2 Quad Q8400 to finish.

STARS Euler3d computational fluid dynamics
Charles O’Neill works in the Computational Aeroservoelasticity Laboratory at Oklahoma State University, and he contacted us to suggest we try the computational fluid dynamics (CFD) benchmark based on the STARS Euler3D structural analysis routines developed at CASELab. This benchmark has been available to the public for some time in single-threaded form, but Charles was kind enough to put together a multithreaded version of the benchmark for us with a larger data set. He has also put a web page online with a downloadable version of the multithreaded benchmark, a description, and some results here.

In this test, the application is basically doing analysis of airflow over an aircraft wing. I will step out of the way and let Charles explain the rest:

The benchmark testcase is the AGARD 445.6 aeroelastic test wing. The wing uses a NACA 65A004 airfoil section and has a panel aspect ratio of 1.65, taper ratio of 0.66, and a quarter-chord sweep angle of 45º. This AGARD wing was tested at the NASA Langley Research Center in the 16-foot Transonic Dynamics Tunnel and is a standard aeroelastic test case used for validation of unsteady, compressible CFD codes.

The CFD grid contains 1.23 million tetrahedral elements and 223 thousand nodes . . . . The benchmark executable advances the Mach 0.50 AGARD flow solution. A benchmark score is reported as a CFD cycle frequency in Hertz.

So the higher the score, the faster the computer. Charles tells me these CFD solvers are very floating-point intensive, but oftentimes limited primarily by memory bandwidth. He has modified the benchmark for us in order to enable control over the number of threads used. Here’s how our contenders handled the test with different thread counts.

Again, the Core i7-975 Extreme is over twice as fast as the mid-range quad-cores here. This is the sort of application for which the Nehalem architecture was intended.

Power consumption and efficiency
Our Extech 380803 power meter has the ability to log data, so we can capture power use over a span of time. The meter reads power use at the wall socket, so it incorporates power use from the entire system—the CPU, motherboard, memory, graphics solution, hard drives, and anything else plugged into the power supply unit. (We plugged the computer monitor into a separate outlet, though.) We measured how each of our test systems used power across a set time period, during which time we ran Cinebench’s multithreaded rendering test.

All of the systems had their power management features (such as SpeedStep and Cool’n’Quiet) enabled during these tests via Windows Vista’s “Balanced” power options profile.

I’ve whittled down these results to just the new processors being tested. You can see the results for the other processors in prior reviews.

Let’s slice up the data in various ways in order to better understand them. We’ll start with a look at idle power, taken from the trailing edge of our test period, after all CPUs have completed the render.

One questions pops out immediately as we look at these results: Why does the Core i7-975 Extreme draw so much more power at idle than the 965 Extreme? Chalk it up to our new motherboard. The Gigabyte EX58-UD5 simply draws more power than the Intel DX58SO that we used with the other Core i7 processors. Gigabyte has a purported solution for this problem in the form of its Dynamic Energy Saver Advanced utility, which is supposed to reduce power consumption. I spent some time trying various versions of this utility, including the latest from Gigabyte’s website, on the EX58-UD5, and none of them worked—they were somehow incompatible with the BIOS revision I was using (the latest publicly available). Frustrating. I had expected Gigabyte to have this issue sorted by now.

Beyond that one issue, the rest of the CPUs tested look well within expectations. Notice that with its two disabled cores still present, the Phenom II X2 550 consumes as much power at idle as any AMD quad-core processor. Still, the Phenom II’s idle power draw is respectably low.

Next, we can look at peak power draw by taking an average from the ten-second span from 15 to 25 seconds into our test period, during which the processors were rendering.

Intel has the edge here in both the value dual-cores and the mid-range quads. The 975 Extreme again draws more power than the 965, probably largely due to the motherboard.

Another way to gauge power efficiency is to look at total energy use over our time span. This method takes into account power use both during the render and during the idle time. We can express the result in terms of watt-seconds, also known as joules.

We can quantify efficiency even better by considering specifically the amount of energy used to render the scene. Since the different systems completed the render at different speeds, we’ve isolated the render period for each system. We’ve then computed the amount of energy used by each system to render the scene. This method should account for both power use and, to some degree, performance, because shorter render times may lead to less energy consumption.

These final results should be no surprise to anyone who has been paying attention. In multithreaded applications, multi-core processors are much more efficient. That’s one reason server processors have been racing toward six and eight cores per socket. Even with its motherboard handicap, the Core i7-975 Extreme places among the top of the pack, because it spent so little time at peak utilization rendering the scene.

The rest of the results are close, yet Intel has a clear edge. The Q8400 proves more efficient than the Phenom II X4 940, and the Pentium E6300 requires less energy to render the scene than the two X2s.

Yep, I overclocked all five of these processors. Took a while, but eh. Just know that these overclocking results are of the quick-and-dirty variety. I didn’t test stability for hours on end, and I didn’t resort to heroic measures in an attempt to squeeze a few extra megahertz out of these CPUs. Instead, I took reasonable steps with common clock and voltage tweaks to reach the best stable speed I could, with air cooling used in all cases. I used smaller stock AMD and Intel coolers for the cheaper processors, but I pulled out the big dawg from Thermalright, a disturbingly large air cooler, for the Core i7-975 Extreme. Let’s take it CPU by CPU:

  • Athlon II X2 250 — I started overclocking this chip by shooting for 3.6GHz, and that’s where I wound up in the end: at 3.6GHz on a 240MHz base HT clock, with the CPU voltage at 1.375V and the HT multiplier dialed back so the effective HyperTransport speed was 1.92GHz. Attempts to go higher were no use, even up to 1.4125V.
  • Phenom II X2 550 Black Edition — With an unlocked multiplier, this one was definitely easier. My final destination was 3.7GHz at 1.4125V. The system never could boot into Windows at 3.8GHz.
  • Core 2 Quad Q8400 — I started off here by aiming for a logical stopping point: at 3.2GHz on a 1600MHz front-side bus, where all system clocks are in harmony once again. That proved possible on the first attempt with CPU voltage set to “auto” in the Asus BIOS, and I tested performance at that speed. With some tweaking, this baby then reached 3.68GHz on a 460MHz base FSB clock (1840MHz effective). For that, the CPU voltage was at 1.4125V, RAM was at 1226MHz, and FSB voltage was at 1.4V. Unfortunately, getting to this lofty goal involved some unexpected reboots, one of which corrupted my Steam install and prevented me from getting Half-Life 2: Episide Two scores for the Q8400 at 3.68GHz. Re-imaging and running them would have delayed this article another day.
  • Pentium E6300 — The first logical stopping point for the E6300 was at 3.5GHz on a 1333MHz FSB, which it handled perfectly on the first try, with the CPU voltage set to “auto.” Unfortunately, the corruption of my Steam install dampened my enthusiasm for seeing exactly how high I could take the E6300. Perhaps in a future article, I’ll test its limits further.
  • Core i7-975 Extreme — With its unlocked multiplier, the 975 Extreme was another easy overclock, and it was another one where my initial attempted speed—4GHz, in this case—proved to be the highest one it achieved, at 1.3875V. The system would boot into Windows at 4.1GHz, but it would reboot during a Prime95 stability test, even at 1.4125V.

There you have it. Here’s a quick performance test at our overclocked speeds, along with a few overclocked results from other recent processor reviews.

Some nice performance gains there in each case. Obviously, overclocking one of the value processors is a great thing, since it will put you in league with some of the fastest desktop CPUs around—witness the Phenom II X2 550 Black Edition at 3.7GHz. Still, as nice as that is, the Core i7-975 Extreme at 4GHz is just incredibly impressive.

The value proposition
We’ve taken a long and meandering route through several truckloads of performance data, and in order to help you make sense of it all, we have ripped a page from our last CPU value article.

To create a synthetic “overall performance” score, we computed an unweighted average of the results for a subset of our tests consisting of the benchmarks used in the CPU value article. Our formula includes 22 different benchmarks, but since our aim is practicality, it excludes a few more esoteric ones like the scientific computing applications. As our baseline, the Athlon X2 6400+ gets a 100% score. Other scores are all relative to it.

Of course, what you see below is a crazy experiment and probably meaningless, but some folks may find it a worthwhile thought exercise, at least. These scatter plots show price versus performance in a fairly intuitive way. To oversimplify slightly, the best CPU values tend to be located closer to the top and left edges of the plot.

As one might expect, some of the new CPUs we’re reviewing today come out looking good in this analysis. The totality of our benchmarks is somewhat biased toward multi-core processors, so the Core 2 Quad Q8400 shows up in a nice spot on this plot—as does its rival, the Phenom II X4 940, whose overall performance is slightly higher. The Pentium E6300 appears to have a clear edge over the Athlon II X2 250, but the Phenom II X2 550 is also a strong value with a higher performance rating.

Of course, the 975 Extreme is no great value, but it is progress over the like-priced Core i7-965, which it replaces.

Now, here’s another crack at the same issue with total system cost taken into account. To get our pricing numbers for the X axis, we’ve added the cost of a motherboard, memory kit, graphics card, and hard drive to that of our processors. Wherever it made sense, we picked components from our latest system guide. Also, we got all our prices from Newegg. Here’s a complete breakdown:

Intel LGA775 platform AMD Socket AM2+ platform Intel Core i7 platform
Gigabyte GA-EP45-UD3P $135 Gigabyte GA-MA790X-UD4P $110 Gigabyte GA-EX58-UD3R $200
4GB Kingston DDR2-800 $51 4GB Kingston DDR2-800 $51 6GB Corsair DDR3-1600 $104
Sapphire Radeon HD 4870 512MB $165 Sapphire Radeon HD 4870 512MB $165 Sapphire Radeon HD 4870 512MB $165
Western Digital Caviar Black 640GB $75 Western Digital Caviar Black 640GB $75 Western Digital Caviar Black 640GB $75
$426 $401 $544

Notice that we are making some assumptions here that may not be entirely valid. For instance, we’ve priced the Socket AM3 Phenom II processors on a Socket AM2+ motherboard with DDR2 memory, though we tested most of them with DDR3 memory. As you may have noticed, memory type didn’t make much difference at all to the performance of the Phenom II X4 810, and we expect the story will be similar for the rest. In the same vein, we priced the Core 2 processors with DDR2 memory, though we tested them with DDR3. Our goal in selecting these components was to settle on a standard platform for each CPU type with a decent price-performance ratio, not to exactly replicate our sometimes-exotic test systems.

Thanks to a lower overall cost for the Socket AM2+ platform, the AMD processors separate themselves from the Intels in this plot. Rather dramatically, a line of green dots runs down the left edge of the performance results between 100% and 175%. Assuming this difference in motherboard prices holds up as typical in the market, the Athlon II and Phenom II chips at various price points look to be the better deals, generally. Among the strongest values from Intel are the Core 2 Quad Q8400 and the Core i7-920, which is in a class by itself.

For what it’s worth, the Core i7-950 would presumably sit atop the i7-940 just like the i7-975 does above the i7-965 at the same price point, as a slightly better value.

We’ll tackle this from the top down, starting with the Core i7-975 Extreme at the peak of the market. Obviously, this processor isn’t intended to be an astoundingly good value or anything like that, but boy, you will be getting an awful lot of CPU for your money, regardless. Intel has placed the performance crown a long way out of reach for AMD, and that isn’t likely to change any time soon, especially with the debut of this new model at the same price as the 965 Extreme before it.

These days, many folks don’t feel the need for a faster processor quite as acutely as they did back when we started doing these reviews, nearly ten years ago—and rightly so. PCs are often fast enough for everyday tasks, and even the latest games don’t take great advantage of, or tend to require, more than two CPU cores. There’s a legitimate question to be asked about who really needs a processor like the Core i7-975, with four cores, eight hardware threads, and three channels of DDR3 memory. But if you’re running a seriously performance constrained application that chews up loads of time—be it video editing, scientific computing, or just compiling a large software development project—you’re probably acutely aware of the fact that more speed would be nice and that the Core i7-975 Extreme’s thousand-dollar price tag is, compared to the time you might save with a faster CPU, just a drop in the bucket.

Such applications don’t usually mix well with extreme overclocking, but the 975 Extreme’s potential to reach 4GHz with a minimum of effort makes it a heckuva nice plaything, as well. Just think: at a 4GHz core clock, it’s hitting 4.26GHz in single-threaded applications via Turbo mode, which helps explain the ridiculous performance we saw out of it. If the Extreme Edition’s job is to be the aspirational performer, the top-of-the-line exotic that makes you want to buy something related at a more affordable price point, well, it plays that role awfully well. I’d suggest a Core i7-920 plus a little overclocking as the sensible alternative, for what it’s worth.

The next sensible step below that may be the Core 2 Quad Q8400, which acquitted itself reasonably well in our tests. At 2.66GHz, the Q8400 is fast enough to play games and handle most single-threaded applications without issue, although in such cases, it was oftentimes slower than value dual-cores like the Pentium E6300 that offer higher clock speeds. I suppose my grumpiness over “value quad cores” has been largely blunted by the Q8400’s 2.66GHz frequency and its rather ample overclocking headroom (at least in our chip). Let’s put it this way: I wouldn’t pay more for a Core 2 Quad Q9400, which is the next step up the ladder, because the extra cache isn’t worth the extra cash. The Q8400 is the better value.

The Phenom II X4 940 outperformed the Q8400 by a small margin in our overall performance analysis, so AMD may have the superior offering here. We only wish AMD didn’t charge 30 bucks more for the Socket AM3 version of that same part, which is known as the X4 945. Also, if you’re a gamer, you’d do better to go for a Core 2 Duo E8500 at the same price as the Q8400. You’ll step up to 3.16GHz and a larger cache, and you won’t miss the two cores in any of today’s games (and probably few of tomorrow’s).

But for the all-around PC enthusiast with diverse needs, the Q8400 is a solid option.

Of the three value dual-core processors, the Phenom II X2 550 is the obvious choice. Both the Athlon II X2 250 and the Pentium E6300 are reasonably attractive cheap CPUs, but the Phenom II X2 550’s combination of a much larger cache, generally better performance, and low-effort Black Edition overclocking makes it easily worth the extra $15 or so over the other two options. The fact that it’s a cut-down quad-core part doesn’t sting in the least; even its idle power consumption is fairly low. I didn’t expect to find myself saying this, but I think the gimpy quad-core will be the best option for most folks. I suppose that’s why we test these things.

0 responses to “A bevy of new desktop processors arrives

  1. “The final piece of the 45nm puzzle for AMD is a pair of low-power Phenom II processors also introduced last week. The Phenom II X4 905e ticks away at 2.5GHz, has 6MB of L3 cache, is rated for a 65W TDP, and will set you back $195”

    not so much of a problem, but I’d prefer to scan the data in tabular format. Thanks!

  2. Especially since you can unlock the two disabled cores on the X2 550.
    I just got one from newegg and easily unlocked it to a quad core. Read the newegg reviews and I haven’t seen anybody who tried and could not. Now that’s a nice perk. 🙂 Did you try that with yours Scott?

  3. Not cheap stuff. That would be the $30-$50 Sempron/Celeron line. I don’t think that segment is as popular now as it once was. The VFM pie is in two main segments- the sub $80 (budget duals) and the sub $150 (performance duals/ budget quads).
    This is the reason why Intel is shifting the performance bar in these segments. Most HTPCs and almost all new office systems use at least dual cores (Pentium/X2).
    Intel can run over its competition with good products, but its not doing that. See how it denied nVidia a chipset license and how it threatened to withdraw AMD’s x86 license over the GlobalFoundries issue (only to shut up when the EU slapped a hefty fine). This is reminiscent of the P4 days when Intel tried to arm-twist the OEMs to not buying AMD’s CPUs.
    Not the right way.

  4. On a funny note- right now, the way SATA HDDs are getting bummed, I’m more concerned about getting my SATA HDD’s to keep working!

  5. Yeah just stay away from AMD chipsets unless you’re after something for a HTPC box. They do well for that. But the SATA performance is tangibly worse than other chipsets. I’ve even had USB issues over the past year that I’ve never seen before.

  6. Well, I haven’t used an AMD chipset as yet. I’ve used the old (and golden) Via K8M800 and now the Nvidia 7000m (notebook) and 6100 (desktop). Somehow, the hard disks seem to perform better on these than on the Intel ICH7- on which they have a perceptible drop in performance.
    And the poor performance of the AMD on Sis(?)/AMD chipsets is because of the SB performance. The system only performs as fast as the slowest component. If I loaded my E7200 on an ICH10R platform, my system would surely fire up!
    AMD has been guilty of not beefing up its execution units (and now on the Phenom II they haven’t put in SMT). That is why the Opteron surpassed the old Xeons in quad socket systems- the Xeons FSB used to get swamped, while the Oppie’s IMC had plenty of room for the ‘guests’.
    Further, I again reiterate that the IMC is a deciding factor in Quads performance if it is made out to be so. Intel managed it quite easily because each core used the FSB for all its business (even the cores in the dual core dies)- easy to manufacture, but they had to change the design ultimately to get in an IMC. AMD uses its inter core connects on a different place, quite distinctly from the IMC.
    But in the end, its not the thought that counts. I always considered AMD’s engineers to be much better than Intels at getting things to work without blowing off the roof (going overboard). Its unfortunate that they had what was possibly the worst management at a time when they were putting out something revolutionary. They couldn’t handle the mess up caused by the 65nm fiasco.
    AMD’s design was and is not faulty, the 65nm implementation was. Now that they are earning again, I hope to see some more products like the K8 (oh, btw the K8 took about 8 months to become a P4 killer- in the beginning it languished in the 1.4-1.6GHz range and underperformed).

  7. What I said is that it’s not possible for AMD to build a cheap quad out of 2x dual cores because the IMC doesn’t allow for that flexibility. That’s been one hell of a nice plus for Intel.

    I also get the feeling that you think I’m saying that a CPU->FSB->RAM design is somehow technologically superior compared to a built in memory controller. I’m not saying that. What I am saying is that it hasn’t been an advantage for most situations. It has been costly too for AMD. They only place it is a considerable advantage is multi socket systems. That’s why we now have i7 and its Xeon counterpart, ready to smash AMD’s last advantage.

    And I really do wish you’d drop the “IMC = snappier” claim. I’ve personally built and used many K8 and K8x2s (and use some still today daily). I have the Phenom 2s here as well. Actually, unless you go with an NV chipset for your AMD machine, you will probably end up with a system that’s definitely not as smooth as an Intel-chipset system because of how subpar SIS, VIA and AMD’s chipset SATA performance is. Don’t get me started on the various quirks of 780G.

  8. It is possible to build quads with 2 duals side by side- either the data is transferred through the FSB, as in the case of the Core series or through the chip interconnect (QPI in the case of the i7).
    Core 2 Quads do not benefit from FSB boosts because of the very problem that inter core data transfer is through the FSB, so a faster FSB does not compensate the higher latency.
    Insofar as the Core 2 having a lot of memory branch predictors improved, it was just to overcome that issue. In any case the 45nm Core 2 Quads, in which the improvements were made, were only 10% faster than the 65nm versions. The true benefit can be seen in the i7 where the IMC can more than handle the multi threaded cores. Could an FSB have been enough for the i7? A point for nordic hardware or who ever says that the FSB is better than the IMC.
    The Phenom failed because of architectural inefficiencies, mostly because they did not make the execution units wide and strong enough. Minor changes have brought the AMDs back with the Core 2 Quads.
    By the way, the IMC does not have anything to do with the way the cores communicate with each other, the IMC is basically the FSB-on-chip. So that the CPU is not at the mercy of the chipset. With BIOS upgrades, you can drop any 65W AMD chip in an AM2+ socket, but you cannot drop a Core 2 in an old LGA775 with FSB 800. It won’t work.
    I have a lot of the better reviews of TR/Anand/XBit/Toms saved on my PC, so I won’t insult the reviewers by saying that I didn’t read them properly.
    Most of all, I have my own experience that though AMD cpus (K8) get overpowered quickly compared to the Core 2, in normal conditions, the AMD is more snappier- the chips simply get the data in and out very quickly.

  9. P4s ran on 400, 533, 800 and 1066 MHz FSBs, actually.

    You might want to go re-read some of your favorite articles. I know of a few offhand that show Core 2 Quads gaining little in the majority of apps from FSB overclocks. You also might want to go read up on the substantial improvements made to Core 2 for memory accessing and branch prediction.

    §[<<]§ As far as I'm aware, the only time a FSB hurts Intel is for multi socket systems. On the desktop I think it's been an advantage, frankly. It's allowed them to build cheap dual cores that still wipe the floor with AMD's best. You apparently can't build a quad out of 2 duals if you use an IMC. AMD's been eating some crazy costs cuz of that!

  10. I dunno. Sure, most people are out to buy cheap stuff that gets them by. But Intel wipes its rear with dollar bills for a reason. They know how to sell stuff and make big bucks on it. They don’t even need to have the best tech to do it, although they definitely do have the best tech right now.

    Sometimes I wonder if the only reason AMD is still around is to keep the US govt away. Otherwise I think Intel would have purchased AMD by now, honestly.

  11. The bulk of the market is the VFM (value for money) market- which purchases the <$150 CPUs. This is why both AMD and Intel enter price wars.
    Even if the Lynnfield is good, its not the end for AMD. Remember that it took Intel 3 full years to get the Core series out, and it suffered only a 10% market share drop.
    AMD’s fiasco has been the Phenom (I), which they are rectifying with the Phenom II. As long as AMD produces a competitive and maybe better platform (keyword- platform, not CPU) that provides better value for money than Intel, its market share will grow.

  12. I don’t have the energy for a complete response to this. Suffice it to say that I don’t think AMD’s intention with these was to “stay ahead” of Intel, rather it was to fully transition to 45nm and DDR3.

    Core i7 975 is WAY more than .2 MHz faster than the 965, perhaps you meant .2GHz, which would also be incorrect. Regardless of whether or not you think it is worthwhile, 133 MHz is a very standard MHz increment for Intel, and it has been for some time.

  13. What makes you expect the 9550 to be more efficient?

    The 9550 is clocked faster and so probably uses a little bit more power just due to that.
    The 9550 has a larger cache than the 9400/8400 and definitely uses more power due to that.
    The 9400/8400 may take a little bit longer to complete a task, but will none the less use less energy to do so.
    Since power efficiency is a measurement of how much power it takes to do something, the less power used, the more efficient.

  14. hey the power efficiency shows intel’s q9400/8400 is more efficient than q9550!!?? how come thats possible?? i dont understand!

  15. AMD has to release dual cores with faster clock speeds to stay ahead of the competition not just by disabling two cores on a quad core processor. and hey what is the idea of introducing Ci7 975 extreme when its barely 0.2 mhz faster than Ci7 965 extreme? dosent make much diference though. the core i7’s biggest enemy are their hefty price tags and only people with really big budget would pick them up. many would prefer processors with best price/performance ratio. as for amd they should try to bring out their bulldozer processors as soon as possible. at this rate when intel will introduce lynnfield then every thing is over for AMD.

  16. Swaaye- A lot depends on what apps you are running. Intel has been stronger in the media encoding always, even from the days of the P4. AMD excels in those apps where the memory thoughput is more significant than the raw processing power.
    I can’t afree with your view that the FSB 800 Pentiums are stronger than the AMD ones- maybe where one app with uniform data flow takes place, yes, but not otherwise. Even reviews from the P4 days (P4s ran on FSB 533, remember?) advocated that OCs should be more productive with higher FSB but lower multipler as the FSB chocked even the P4!
    When the C2D is more than 4 times as powerful as the P4, even the FSb1066 is a limiting factor, which is why Intel depends on higher cache to compensate for the slow FSB. ALso Core 2 cores communicate with each other via the FSB, unlike AMD which always had an interconect.
    One need not look at tests to judge the merits of Intel CPUs, just have a look at the cache/FSB variations from the lowest to highest ones. AMD’s IMC is stronger than what any reasonable CPU can handle. So it needs less cache and any limitation is CPU bound. And lets forget the Phenom- I hope that its buried with Hector Ruiz.
    The above can be confirmed by looking up historical reviews from TR/XBitLabs/Toms/Anandtech- 4 sites I referred to for over 4 years.

  17. Not the most up-to-date but should paint a nice picture of what you are saying:

    §[<<]§ AMD definitely has nothing on Intel for the very high end but competes very well on perf/dollar. And looking at the charts, a Q6600 barely better than a Phenom X4 9950 but not Phenom II X4 810 (and definitely not a 920). As for server side, AMD manages to hold its own with Shanghai and Istanbul in the 2 sockets configurations: §[<<]§ And still generally dominates Intel in the 4 and 8 sockets configurations, Nehalem EX seeks to counter that. Couldn't find something -[<]§ EDIT: typo...

  18. Seriously, what are you talking about? This very article includes the Q6600, some original Phenoms, and a representation of every type of current 45nm CPU. None of it backs up what you are saying, and there are countless other sites with the same results.

    You seem intent on inventing things to complain about that aren’t real. Several times now, you have specifically stated things regarding Phenom IIs that aren’t just unrelated, they’re downright inaccurate.

    Phenom II does not have a TLB bug.

    The Q6600 definitely is not faster clock to clock than a Phenom II. It didn’t even have much, if anything, on the *[

  19. What Intel did was put together a quad core that works and performs amazingly well. That’s not what AMD did. Hell Phenom 2 didn’t even work right on a few levels, including CnQ and that TLB bug which was huge for commercial clients. i7 and K10 are quite different in just about every respect other than die photo. Sure they have similarities, but that comes down to things such as modern CPU architectural techniques being similar for everyone and plain ‘ol x86 core design being evolutionary.

    What I think K10 proves is that AMD should have designed a better native dual core instead of a native quad. Intel has enjoyed cheap, extremely fast dual cores that are easily turned into quads that defeat AMD even with the supposed “AMD native quad core” marketing advantage.

  20. How can you say a Phenom 2 “walks all over C2Q” when the C2Q is clearly faster per clock in the majority of apps, even when it has gimped L2 cache?

    When I say Intel has superior memory utilization efficiency, I’m referring to articles from RealWorldTech primarily. And the fact that a 2MB Core 2 on a 800MHz FSB isn’t exactly a slow processor that can’t keep up with AMD’s best in per-clock performance.

    My personal experiences with x264+heavy AVISynth scripts on my Q6600 @ 3.0 / DDR2 800 and my PII-940 @ 3.3 / DDR2 1066 / 2.2GHz NB tell me a lot about these CPUs. The Q6600 outperforms that PII even with a 300 MHz clock deficit. That’s a 65nm 2 year old C2Q beating a rather brand new 45nm “native quad core” per clock by a sizable margin.

    Sure if you load up Everest or Sandra and run some synthetic memory tests you’ll see the PII win, especially in ganged mode. But for real apps it doesn’t mean jack as far as I can tell. I’m sure there are some special situations that really can leverage memory bandwidth though.

  21. DDR3 for one. It is faster in nearly every bench than the x2-6400 while running 6.25% slower, for another. Cheaper to produce for another.

    AMD’s next real chance is Bulldozer.

  22. By “crippling” I mean, turning off that (pesky) Turbo Mode that makes it “overclock” itself. That’s it. I’m curious to see how fast, would the i7 family be without Turbo Mode assistance. (Turbo Mode while being a cool thing, does indeed interfere with an “apples to apples” comparison, where the cpus in question would have a FIXED operating frequency.)

    In fact, would it be possible for the MB manufacturers to devise a special BIOS SW that’ll constantly watch the cpu core temperature and as long as the temperature stays below an a priori set maximum value, it will overclock the CPU? MB-assisted turbo mode. It would have to work much like that “fancy” fan speed controlls, only that the bus speed/multiplier would be adjusted instead of the fan speed. In fact, you could take this up to the next level, where the user would be allowed to choose between different control curves, as in, each time you go up in frequency, do it 1Mhz at time with a rate of 1Mhz/3 seconds and each time you need to go down in frequency – do this 10Mhz at a time, with a rate of 10Mhz/second.

    Am I making any sense, here?

  23. I doubt the native dual-core Phenom cost much to R&D. The quad-core flavors already did all of the hard work. AMD’s engineers and architects just had to rescale and retool a couple elements to make a dual-core version.

    The point of these chips is affordability. Performance difference and power consumption difference between older Wolfedale-based Core 2s/Pentiums and new dual-core Phenoms/Athlons is marginal at best. The market segments that go after these chips don’t care about such differences. They only see and care about platform cost.

  24. The percentage of CPU utilization has nothing to do with the hardware and everything to do with the OS and the application being used.

    My perspective on things is that K8’s IMC was a brute force way to solve their inability to build a high quality memory controller into the chipset and to design their CPUs to be highly efficient with memory accesses. It really was basically a K7 but with the IMC added on and that brought about most of the performance increase. K7 always had rather craptastic RAM performance. Phenom is definitely a lot better than K8 but Intel definitely has the superior design in C2D and later with regards to using RAM efficiently. The evidence is everywhere. You can go read about it in depth at Realworldtech if you’d like.

    When you get into multi-socket configs, AMD did clearly have the advantage until Nehalem. Unfortunately for AMD, Nehalem is Phenom done right in a big way. Phenom was good in the server space (Opteron I mean) until Nehalem (which is an unproven CPU still but it’s not going anywhere methinks).

    FSB vs. IMC is a pointless argument on the single-socket machine. Hell there are benchmarks out there that show C2Qs gaining very little from huge differences in FSB clock. They don’t saturate the FSB. They outperform AMD chips regardless. As far as I’m concerned, my old-n-crusty Q6600 is clearly the superior chip compared to my “state of the art” Phenom II chips.

  25. Professional and business use, yeah, that’s why it’s in Win7 PROFESSIONAL and Ultimate…home users and enthusiasts, not so much. I’m an ‘enthusiast’ and for over a year I have had absolutely zero problems running programs on Vista, at least problems due to Windows itself. It’s nice that it’s available but Virtual XP in Win7 is getting way too much inappropriate attention.

  26. No prob. I wouldn’t normally go all terminology nazi but you used it like seven times and it kind of kept grating on me. We’ve got enough virtual ambiguity as it is.

    But I agree with you that hardware virtualization support (VT-d or AMD-V) is an important feature for a a lot of professional or enthusiast users and therefore can make the difference when choosing a processor. For other users it’s a complete non-issue (unfortunately, since many of them might actually benefit from it via XP Mode in Win7… if it was done right, and if the Home version of Win7 supported it).

  27. Yes and no. I mean, broadly speaking, it is good that this Turbo Mode came into being since, after all, it represents a technological progress and this is always good. However, how can you keep a straight face when you’re comparing two processor, each rated at say – 2.5Ghz and one will always stay at 2.5 Gigs and the other will seldom be at 2.5Ghz since it has the ability to, on the fly, increase its own frequency (within limits) provided it does not exceed its safe thermal envelope?

    The second one, being more advanced will most probably make full use of the Turbo Mode and win most contests. But, from here to saying that it also enjoys more “number crunching” prowess per clock is a long way.

    You could only say this if the two processors would be clocked all the time at the same frequency THROUGHOUT the tests. But since, not even Intel itself can predict just how often the i7 will accelerate itself beyond it’s base frequency, you cannot draw such a conclusion.

    So – yes. i7 wins them all – but no more of that “[…] and has more clock for clock prowess[…]” bussiness. This is the only reason that I may have for “crippling” the i7 for a lab – academic if you will – test.

  28. What about the E7600? This is a 3.06GHz Core2 CPU with 3MB L2 cache, SSE 4.1 and VT
    §[<<]§ §[<<]§ (ignore the inflated initial price - it should level off around $130 in the next couple of weeks) Also, I don't quite agree with the comment "We unfortunately don't have an E7400 on hand for testing, but it should be only slightly quicker than the Pentium E6300". Outside of the L2 cache size that would have some moderate impact depending on the test, the E6300 has VT, but the E7400 does not. And the E7400 has SSE4.1 and the E6300 does not. Those can be significant differences depending on the test. - The E6300 can run the XP mode on Windows 7 Professional but the E7400 cannot. - The SSE4.1 of the E7400 can provide a noticeable boost in some application.

  29. Gimping features is interesting in an academic sense, which I think is what you’re getting at, but I’d much rather see full-featured comparisons first and gimped ones second.

  30. So you want to gimp Intel’s CPU to somehow prove that AMD isn’t “quite that bad”. Considering “Turbo Mode” seems to offer performance with no caveats, I would say that would be completely wrong to do. Intel figured out a way to gain a good bit of extra single-threaded performance while AMD couldn’t come up with anything similar.

  31. Yep, noticed it yesterday in some online stores, so I ordered one (Retail pack) this morning for €215 (+shipping).

  32. Yeah it also makes the other Intel quad cores less attractive, the extra cache alone seems to give ~5% increase (going by some Q9450 benchmarks for clock-clock vs the Q9400.) Something mentioned in the article I wanted to comment on too – the ‘official prices’ didn’t mean much for Q8400 vs Q9400, one could easily find the Q9400 for about $15-20 more and for a generally ~5% increase due to cache alone that isn’t bad.

  33. Thanks for that correction! That just goes to show you that I shouldn’t post anything before I have had my morning coffee!

  34. You’re right! See? You reminded me of one of Intel’s features that I’ve always wanted to talk about. That “Turbo Mode”…is there any way it can be inhibited? I mean… pitting the Nehalems against anything the Phenoms have to offer, on “fair” terms. No user would do that, but just for argument’s sake, maybe it will put a stop to that “Intel’s waaaaay faster, clock-for-clock, than AMD” mantra that everybody is chanting since a few years now. I mean, if this can be done, and the a “crippled” 2.8Ghz Nehalem still wins, then I’ll “forever hold my peace”. But, seeing the benchmark results, and seeing the Phenom IIs more than holding their own as it is, I cannot help to think that, maybe, Turbo Mode is just about the only thing that’s preventing Phenom to sweep the floor with i7, much like the Athlon did with P3 and P4 back in the day. Now, if Intel had Turbo Mode with P3 and P4, maybe the K7 era would have been shorter.

  35. I’m sure they’d love to do that, but I wouldn’t be surprised if the specific reason that they do not is that they can’t really afford it. It would probably cost them more than just lowering Opteron prices, which is the only way it’s really hurt them so far. They don’t lose their market in doing that, as they still have a number of advantages over Intel. They just don’t get to charge as much.

    What they’re really missing is something like Turbo Mode. That helps just about *[

  36. Sorry if I’m intruding MMO, but looks like I put that phrase somewhere.
    What i’d like to stress is that my comments are made along the VFM line and not the performance line.
    In the value line- NBs are differentiated mainly by the IGP quality, the RAM speed supported, and the PCI-e specs, most of which are nearly equal. It all boils down to the SB and the IGP drivers.
    In this matter AMD/NVidia IGP are much better than Intels, also Intel bundles the old ICH7 in most of its budget chipsets, while AMD/nVidia have better ones (even the nV 430a has RAID and much better expeience than the ICH7).
    So when AMD/NV drivers have support for video/TV quality and allow setting of profiles in which one can save all settings through the IGP rather than fiddling with the monitor, its a godsend.
    And all budget CPUs are more or less equal on different counts.
    Even when I don’t game much (MotoGP max), I ended up putting a Graphics card on my Intel build to reduce the monitor gamma and set up different profiles for different viewing scenarios, and for good support for my TV tuner playback quality (nVidia drivers for the 8000 series also provide support for noise reduction in video playback).
    Intel IGP drivers are at the most basic.
    My overall platform cost for intel was significantly higher because of poor IGP and slow SB (I had to stripe across different HDDS).

  37. Man…If was Dirk Meyer right now, I’d get out the CEO clothes, I’d call up the design team from the good old “AMD K7” era and forge in some version of HT into the new Athlon x2 dual-core. I’m talking about “Skunk Works” style working. Get the team together. Re-locate them somewhere out of the main corporate headquarters, to a farm or and old abandoned warehouse where no one can reach them. Temporarely re-locate their families it necessary, but have them spend every hour they’re awake on either inserting HT into at least one of their current chip or, re-design something there – execute more instructions per clock – I don’t know – SOMETHING!

    I feel they really have an opportunity on this new Athlon x2 front. Since it’s a clean next gen dual-core design (not a Phenom with two cores gone AWOL), they really should (have) spend (spent) more time trying to make it faster. They trick is not adding another core. The trick is making the two-core literally breathe down the Phenom X3’s neck. THEN we can speak about adding another core. When you say “well…we’re gonna add 5 more cores and then we’ll get them” it just seems like sloppy design.

  38. I agree with the CnQ issue. But one thing that may affect what we are talkign about is that in encoding (WMM and AnyVideoEncoder), the Core 2 seems to run at 85% tops, while the X2 runs at over 95%. I remember seeing some tests during multi threaded scaling that proves this also.
    So maybe your X4’s cores ran out of steam.
    That’s the reason why Intel scrapped the FSB1600 from general use. The Intel FSB chokes before the CPU does and the AMD CPU chokes before the IMC does.
    I did mention that AMD lags in pre horsepower. And the Phenom is frankly best forgotten (AMD had to bring out a P4!)- the Phenom was 15% faster than the 65nm X2, which in turn was 10% slower than the 90nm X2.
    I use 945/ICH7 on intel and nV 6100 on AMD. All budget builds. No sense in going for G31 as that too has ICH7.

  39. Actually, I prefer Nero 9 for simple things, but kept Nero 8 because Vision was buggy in the latest one. Nero 7 had a tough ride to even become compatible with new systems.

  40. You can’t equate OC on an FSB CPU to an IMC CPU. The AMDs OC can be only compared to the i7, which they can do well right from the word go in 45nm.
    Also note that the i7 975 OCs less than the 920. So Intel’s not always going forward.
    And all tech steps are planned years in advance. Intel started prepping for 32nm 2 years back and its still 2 years away.

  41. As I said though, AMD is not alone on 45nm. It’s a joint effort with IBM. They’ve been working with IBM on manufacturing processes since K8.

    BTW, you could’ve picked up a day-one 45nm Wolfdale and cranked the clock to high hell. They weren’t rough around the edges at any point in time. Maybe they’ve gotten even better, but they have always been great overclockers.

  42. Huh? AMD is winning with their dual core Athlon II and Phenom II X2 against the price comparable Intel parts. Which is all that matters to me as a consumer. The E6300 didn’t look so great in this article compared against the Athlon II or Phenom II X2.

  43. Oh I don’t really see it that way. My two Phenom 2 w/DDR2-1066 systems get noticeably sluggish when I peg all 4 cores on something. My Q6600 doesn’t do that.

    All that IMC does is get AMD better memory latency. Intel actually has smarter memory usage than AMD, believe it or not. That’s partly why Core 2 Duo can squash AMD’s chips even though it is supposedly hindered by that “retro” CPU->NB->RAM design.

    Your problems could be caused by the motherboard. I’ve found that 780G, for ex, has really poor SATA and USB performance compared to competing chipsets from NV and Intel. My Q6600 is the smoothest system I own. Also have a old E6300 @ 3.2GHz here that’s wickedly snappy. Q6600 is on a Gigabyte P35 mobo and the E6300 sits on a Gigabyte P965 board.

    One thing’s for sure, Intel has had fewer issues with Core 2 than AMD with Phenom. Even Phenom II has serious CnQ performance issues in XP because AMD didn’t bother to fix the core clocks to be equal. They only fixed Vista. All that hoopla over “more advanced power management” with Phenom 1 was nonsense and even turned out to be troublesome.

  44. Actually, it is on sale, but just as of yesterday, I think.

    I figured there were some potential BIOS issues. I didn’t realize even this site mentioned it. Thanks for pointing that out.

  45. Indeed, Intel introduced a completely different extension called virtual mode extensions (VME) with the Pentium and even backported it to later Intel SL-enhanced 486s (including later Intel 486DX2s and all Intel 486DX4s). That extension was for the Virtual 8086 mode used to run DOS apps on modern OSes and was completely different. It fixed some of the problems of the original 386 version of Virtual 8086 mode. For example, it allowed the processor itself to maintain the Virtual Interrupt Flag (VIF), instead of having to trap every instructions that reads and writes to EFLAGS and letting the OS itself maintain the VIF (as was done in the original 386 version of V86 mode), which was much slower.

  46. OAS- The Athlon II’s efficiency is built into the hardware C1E state. However, given window’s inefficiency at switching cores, and the lack of proper BIOS code in many mobos- the X2 250 shows slightly lower performance and higher power than normal. This is the explanation given in some sites (I refer to TR, Xbitlabs, Toms Hardware and Anandtech ususally).
    However, the figures in TR seem ok to me. In some reviews, I saw the 250 languish behind the 550 even in media encoding tests,which seemed hard to believe, but the above seems to be the reason. I think BIOS code is on the way, and that seems to be why the 250 is not on sale yet.

  47. man, you get the pedant award today 🙂

    I much prefer calling them victual extensions anyway…

  48. About the power difference between the Athlon II and Phenom II dual-core, I’m now completely lost on it. Taking out 1/3 of the chip ought to have some change, but it’s not clear what that is.

    I looked up a ton more reviews that include both CPUs, and some of them have basically no difference in power at all. I have no clue why Tech Spot shows such a difference, and they even point it out in the explanation, so it’s not just a typo. I don’t think you can really chalk up so many variations simply to power supply efficiency.

    However, it does appear that TR used a separate BIOS version for all 4 CPUs that were tested with that particular motherboard. That further confuses things…

  49. cocobongo, Hyper Threading can be disabled. In fact, TR used to include benchmarks with it on and off for Core i7s:

    §[<<]§ I imagine that was taken off because it cluttered the now very long list, and people are just going to leave it on. The funny thing is, turning it off can do the opposite of what you're suggesting. Hyper threading is an additional overhead on the CPU, and it can slow it down a little in work that's not heavily threaded. That's one reason why every CPU under the sun doesn't have it, even though it's nothing new. Even if it's not slowing it down, it's generally not helpful in desktop applications, as Core i7s have already got 4 cores, and not much is going to use more threads than that. For some people, it might, and so those people know to get a Core i7. This is why only 1% of Intel's shipped CPUs are Core i7s, and they don't expect it to be more than 2% even after they release the new Lynnfields. For everyone else, Hyper Threading doesn't matter. It's largely a benefit for certain types of server work loads, just like with the triple channel memory interface. Nehalems are really designed to be server CPUs. With that out of the way, back to derailing this thread... Xylker, I don't think anyone is disputing any of that, but AMD didn't intend to just avoid Hyper Threading altogether. For no particularly good reason, they just thought they'd wait it out a little longer, and rely on CPU designs based on higher core counts. Unfortunately, AMD gambled and lost. Regardless of who wins the "performance crown" between the 12 core Opterons and 8 core Nehalems, all the way through 2010, AMD has to continue to make larger chips and sell them for less. Almost all of the blame for that rests on their intentional exclusion of Hyper Threading. They can continue to increase bandwidth and clock speed, but even if their Hyper Threading had been worse than Intel's, it still would be completely necessary to even the score. They openly admitted that this plan was a mistake, and because of their mistake, it will take them until 2011 to include Hyper Threading and start to play catch up. As you noted, they can't just tack it on to existing CPUs, which is what they're stuck with for at least another year.

  50. I’m not completely sure they would know, but I have no pity for Home buyers who lament lost features either. Fair point though.

  51. I purchased a Pentium dual core after reading the tests about how it bashed up the X2. Then I came face to face with something called the FSB. Now even after switching to the E7200 OC’d to 3.16GHz (FSB1333), switching power tasks sunning simultaneously is way better on my 90nm 2.2 GHz X2 than on the C2D.
    The CPU is just the one part. And if the Core 2 was indeed perfect, why did Intel licence AMD’s memory controller tech and follow the Phenom’s design in the i7? (Sure sure- they made everything better- they can afford to- Intel made profits from the P4 and now the Atom).
    Honest point is- you cannot realize on tests to reflect what the platforms will be like in real life. AMD’s $80 cpus are compensated by slightly less horsepower by excellent memory controller and IGP’s, to which Intels answer a $250 CPU and $150 mobo that still sucks in the IGP dep’t.
    My end take is that Intel’s CPUs may finish that encoding a few minutes faster (wait for 2 hours or one hour- you still wait don’t you?), but they still are not at par with AMD in terms of system responsiveness in real life.
    This is not the finish line- AMD needs to first of all recover its losses through low risk methods and designs. Then we’ll see why 5 years after the K8, and 2 years after the Phenom, Intel copied all that into its i7 (ah but scored extra for its handwriting)

  52. AMD’s entire manpower is less than Intel’s manufcturing strength. So they will be behind. But they are behind only in process tech. We all know since the days of the P4 that its Inte’s marketing that is much better than AMD’s. Even during the Athlon 64-P4 days, Intel had over 75% market share. See how the TLB bug pummelled the Phenom, though it wouldn’t affect 99% of users, while Intels CPUs are selling in spite of not supporting VT overall.
    Intel’s puny share in a few watts of power consumption is offset by AMD’s excellent platforms.
    AMD has a lot more than 45nm to catch up in- namely profitability. They won’t and can’t risk drastic changes in tech, but will evolve.
    Any by the way, process tech cannot be perfected after 6 months- Intel is still doing it after 1.5 years.

  53. Whoa!

    Hyperthreading is a LOT more than just [bolt on parts] [check marketing box] [promote as better] and it has some drawbacks as well…

    In the P4 days when Intel learned how to do HT it was borderline necessary to the P4’s competitiveness. I am certain that they learned a lot about how real code reacts to “logical processors” They also learned about cache thrashing and some not so fun side effects of HT.

    At the time, AMD had zero need to mask long pipelines or access latencies. Now, 4-5 years later and Intel has gained experience with HT and knows how much cache is required to keep it fed, plus their execution engine is a third wider than AMD’s. and it no longer carries P4 baggage…

  54. Why should it outperform Wolfdale? It’s a cut down Phenom II. That’s like saying an E5000 Pentium dual-core should outperform a Wolfdale, just because they came out later.

    The fact that it mostly equals the Phenom II version is pretty impressive. I don’t see why you think it performs so poorly.

    Selling a cheap CPU doesn’t put them in the league of Intel’s $500-1,000 CPU margins, but the “affordable” range has pretty much always been AMD’s place. It only makes sense that they’d adapt to take as much advantage of it as possible. Their strategy of cutting down full size, and rather large, chips, isn’t feasible for a huge chunk of their market.

    To my knowledge, Intel sells a lot more Pentium dual-cores than anything else. That’s what goes into most computers on store shelves, and even lots of “enthusiasts” buy them. Heck, I have two of them. This will put AMD in a position to retake that portion of the market, and also expand on it in a way that Intel would never even consider, once they have tri and quad-core Athlon IIs.

    As for the power, it’s still 6 watts less on the TR test. That can give you a pretty substantial amount of extra battery life on a laptop.

    However, TechSpot’s comparison showed a 20w difference, using the same motherboard:

    §[<<]§ That makes me wonder if the board TR used just didn't get along with it as well as it could, since it's a new type of CPU.

  55. Not necessarily. This computer has a Q9400 and a Radeon 4850. It idles at 90w, and it doesn’t have the world’s most efficient power supply or feature-devoid motherboard, either. I’m quite certain that it could be lowered down to 80w with the right setup.

    I have another computer which also uses a fanless 9600GT and an E5200, and it idles at 105w. Just because it’s less powerful, doesn’t mean it uses less power.

    As time goes on, graphics cards, especially, have been becoming more efficient at idle. There are plenty of examples, even very recently. The 4890 sits at the same idle level as the 4870, but it’s a good deal more powerful in outright clock speeds, and even runs higher GPU voltage. The 4770 is pretty much neck and neck in performance with the 4850, but it uses less power.

    Even the 65nm GTX 200s have better idle power than the 55nm versions of the 9000 series, as unbelievable as that may seem.

    A lot of sites tend to use motherboards and power supplies that are overkill, but that doesn’t necessarily explain all of it.

    The “missing link” is probably that the 4870 is very commonly used. It definitely use more power at idle than a 4850, and most other cards. While they say that GDDR5 is more efficient, I think that’s only in 1:1 performance compared to GDDR3. It’s not twice as fast AND more efficient at the same time.

  56. Sure. But it’s still disappointing that they can’t beat their competitor’s finely aged dual core part. They R&D’ed this thing and have to sell it on day one for $100 or less. Amazing profits to be had there I’m sure……..

    I suppose it is better than selling those Kuma garbage cores for $60. Better for everyone because those chips are really not good on any level other than their Phenom-esque HT clock that benefits IGPs.

  57. Well it’s certainly not impressive from many standpoints. It’s a fancy new 45nm dual core that can’t outperform ‘ol Wolfdale. It’s probably slower than Wolfdale 2M per clock. It’s not amazingly low power either really.

    At least it’s cheap, but that’s not good for AMD’s margins. If it performed better than Wolfdale, they could charge a premium for it. But their tech is subpar so they can’t.

    Back in the day with Duron, AMD slapped around Intel’s low end stuff. This thing doesn’t really do that. It’s just competitive from a pricing standpoint because AMD’s marketing people are realistic. So they developed a new core that they have to sell for cheap. That doesn’t seem like an answer to very many problems to me.

  58. There are a heck of a lot of reasons for Athlon IIs, every last one of which is very good.

    In most things, it’s right up there with the Phenom II version. I’m not sure what you’re complaining about when it’s cheaper and has relatively limited drawbacks. Unless there’s some specific thing you do that the L3 cache is necessary for, then it’s just a more affordable CPU, as is the case for most people.

    Less cache cuts unnecessary power consumption for normal use, which, as noted, makes it a much better option for laptops.

    The L3 cache is about 1/3 of the chip size. If they’re actually making them without it altogether, that should increase their yields drastically. It will likely result in both lower prices and higher margins for AMD. Everyone wins.

    AMD bleeds too much cash and has too low of margins compared to Intel. They need something to put in their pockets to keep up the competition. A cheap and effective budget to midrange CPU line will accomplish that for them. That’s always been their forte, anyways, so it’s best that they pull all the stops to hold onto that.

  59. This is a case where AMD just flat out made a mistake, rather than something else working against them. They’ve said they didn’t intend to include hyper threading until 2011. They didn’t think they needed it, because they have more physical cores, but in relying on that, they handed Intel a free ginormous performance boost – at least, for server work loads.

    Luckily, it’s pretty much inconsequential for most everyone else in desktop use.

  60. 1.) Do not make claims you can’t possibly keep. Not every chip will reach those levels. Even if most will you’re simply setting up many people for dissapointment.

    2.) Scott certainly has the skills but having the time or even believing such a thing should belong in the review is entirely different.

    3.) They use the motherboards as provided by Intel and AMD for testing.

    4.) To an earlier point; I still wouldn’t purchase a Core 2 system if gaming is your reason for the hardware. Your upgrade path is pathetic from now on. The Phenom II’s also OC very very well. It’s a literal toss up swayed by the fact that you’ll have upgrade options in a year from now that don’t involve retooling your motherboard, processor, and RAM all at the same time.

  61. Err uhh. Consider it is much cheaper to manufacturer AND it will be used in laptops. I think it makes total sense.

  62. Your last sentence is the entire reason they made a native dual-core so it’s certainly not a waste of R&D. AMD doesn’t make Core 2…

  63. What a waste of time and money it was to R&D that new native dual core. Even worse perf/clock than Phenom, even with double-sized L2s. Hopefully it’s at least saving them serious money on chip manufacturing costs… That certainly seems to be the only gain here.

  64. AMD has been on 45nm making quads for more than half a year now. If they and their IBM 45nm “Alliance” don’t have the process tweaked yet, well that’s ridiculous considering how far behind they are in general.

  65. Having used the pentium E2160 and the C2D E7200, I long to go back to AMD and its super fast memory controller. Problem with tests is that the testing software generally takes over the system and pummels it with data along predictable lines.
    I feel that the Core 2 line’s FSB is way behind AMD’s Hypertransport, and the FSB frequently chokes even the dual cores when operating more than one program which load the memory sub system.
    Along that line, the AMD 250 and 550 should be unbeatable options, especially since AMD and nVidia chipsets are way better than Intel chipsets in the graphics dep’t (can’t say about the disk controller issue- maybe AMD used seagate’s ‘cuda 11 series for testing).
    But hate to say that even a fortnight after the new AMD chips have been reviewed, they are not to be seen on Newegg. Are they paper launches? I thought that AMD left that behind.

  66. this review is amazing… so complete, so informative, and still easy to read..

    great work scott.

  67. What is your graphics card? My graphics card adds 50watts to the idle power (4850).

  68. I/O is not that much of a factor outside a few benchmarks. I doubt it will bridge any performance difference between AMD and Intel platform.

  69. Kinda whack that your idle power consumption figures are so high. I have an e8500 on an ASUS P5B Deluxe with three SATA Drives, a Blu ray Drive, xi-fi, 4G RAM, W7, 64-bit and my idle is a good 40-50 W less than your test platform. You mention that it is after cinebench has run, but even after I run benches it goes right back down to 87-95W.

    I use my APC Back UPS 1500VA realtime data to test, and I turn off my my monitors and other equipment as wellg{<.<}g

  70. Yeah, damm shame.

    Nero 7 works so much better than Nero 8 and 9. Although, Nero 7 lacks BR and -[

  71. Damage…the platform cost thing under the value proposition page seems a little arbitrary. I think people recycle KB, mice, HDD’s, optical drives, etc. But even if they buy new, I don’t think the gfx card should be included in the value proposition as it is a sizable chunk of money for a good one, compared to other components. For a value comparison, I’d only put in:


    Because CPU/board type dictates which type of memory to use, etc. that seems OK. And with DDR3 being competitive, an upgrader will likely take the plunge. Even though I’ve built Intel, AMD would look even better value-wise looking at the Board/CPU/Memory $$$.

  72. The Nero Test should be excluded. My how that company has fallen hard in making quality softwareg{<.<}g

  73. I have gotten to the point where the CPU indeed doesn’t matter as much for home use..
    .. where I WOULD like a much faster & multi-core CPU is at work.

    At home I generally have one or two apps running at a time.. at work it’s 5 or 6 on average and my poor work PC cant handle it well.

  74. Both the Pentium E6300 and the Core2 Q8400 has VT. The E5000 series and below and the other CPUs in the Q8000 series doesn’t have VT (yet?).

  75. nvidia sucks. I’ve had way better stability with AMD chipsets than Nvidia ones. .. a little performance for a lot more stability is a trade I’ll take.

  76. Nope. Dual electrical loads on the FSB make the Penryn Quads more limited, along with the fact that you have twice as many gates switching, of course.

    Doh. Must hide any indication of “skillz.”

  77. He didn’t overclock the E8400 at all I don’t think. He overclocked the Q8400.

    Had he attemped a E8400-E8600 yes, we should expect around 4Ghz with a halfway decent heatsink. It seems like a foregone conclusion these days.

    I don’t think the Q8400 matches the dual cores in overclocking.

  78. Chill man, remember that this is the first stage of AMD’s 45nm process when Intel’s 45nm is mature. ALso thegap is not that far off, remember that the 250 and 550 are over 200MHz faster than the E6300.
    Wait for a few months for AMD’s 45nm to mature out. The X4 955 OC’s much higher than the first 945’s, so they’re improving.
    Also for the slightly higher output, you get a chipset that beats the pants out of an Intel chipset.

  79. He must be an engineer – Scott’s done an excellent article on marketing, and this guy wants to screw it all up with a more apples-to-apples comparison. Where’s the fun in that? 🙂

  80. Not meaningful imo for home users in that any DIY home user already knows whether they need it or not regardless of Win7. also only Business and Ultimate have virtual XP.

  81. It’ll be interesting to see how VT will factor into the Value equation once W7 comes out properly. Although the Q8400 has it also according to Intel’s website.

  82. Yeah i7 isn’t for gaming really so if that is one’s primary or especially only concern it shouldn’t be under consideration.

  83. I agree with you. If I was upgrading my whole rig now I’d go for an i7.

    But come on!

    £200+ for mobo, £200+ for CPU, £80+ for RAM

    compared to:

    £100+ for mobo, £130+ for CPU, £40+ for RAM

    Do you think you got £210+ worth of performance increase?
    Why even bother when my (overclocked) E8400 can actually match any of i7s in gaming performance (which is all I’m interested in)?

    (I know the newer mobos cost less but, at the time…)

  84. Sadly SB800 was originally going to be 1H 2009 and was meant to coincide with AM3 socket launch but it got pushed back to Q4. I’d be really interested in getting an integrated chipset AMD board with that SB and new NB but alas the timing is wrong.

  85. I am newly impressed with the x4-955. For the most part, when the disk controller is not in the way, the x4-955 meets or beats the q9550 and it draws less power at both idle and load, and it costs less. And it’s an unlocked CPU. Dam, that ain’t bad at all.

    Waiting for a competitive AMD south bridge is very tiresome. Bring it already. Anyone know when that this is going to hit the market?

    November/December should be exciting for me, that is when I plan to do a new build and it will either be an x4-955 or a -[

  86. 125 dollars for the X3 720… now that makes for an attractive chip, especially with its low power draw. Might be worth putting into a HTPC.

  87. So, basically, as long as your processor is Core2 or Phenom2, it is pretty much fast enough.

    The x4-955 and the q9550 look like not such bad values either, IMO.

    Thanks for the hard work Wasson.

  88. There’s nothing stopping AMD from putting HyperThreading technology in their processors. I remember reading somewhere before that AMD was considering putting HT in their chips but decided not to.

  89. I love overviews of new chips compared to older ones. It is a nice way to keep up with what is available, even if I don’t have a good excuse to build a new system.

    One thing that is often overlooked when comparing processors is virtual extensions. All things being equal (or close), I go for the chip with virtual extensions. This is one area where AMD has been a leader. They include virtual extensions in almost every processor, while Intel has always held them back in the lower price ranges. Now that AMD is closer to Intel in performance and tdp, virtual extensions could be the tipping point.

    Now that MS is talking about virtual XP in W7 as a way to entice XP users. having virtual extensions could be plus. My understanding of the E6300 is that it was supposed to fill that hole for Intel.

    I have been using Linux for years. I used to keep a dual boot system with XP on it. For the past two years or so I have only used XP in a virtual machine for the occasional legacy application. This is why I run a Q6600 instead of a Q8200 in my current system. I am always on the lookout for cheap/low power quad processors with virtual extensions. I went Intel with the last build anticipating cheaper 45nm processors with virtual extensions, but they have not obliged so far. Maybe I’ll go AMD next time.

  90. To be fair there is nothing ‘extreme’ about getting an E8400 (E0) to 4GHz. It is very common and a 24/7 setup, at that!

    Also, I did not slave at all. It took 10 mins of Google searching, came across the thread, and 20 mins to apply the settings.

    Most E8400s can do 4GHz with low/high voltage, so for people with a decent motherboard and a third party air cooler they represent a real and viable alternative to the extortionist prices of an i7 setup.

    I agree with Meadows, in principle. A specialist blogger who can unearth the max 24/7 OC of a CPU will shed a whole different light on the value and particular benefits of a certain CPU.

  91. Nice review. I’d like to see an article solely on the Athlon X2 250. I kind of wish this chip to be AMD’s “Celeron Mendocino”…as in…the cheap CPU that packs as much punch as the big boys.

    On a second note…I understand that AMD and Intel have a very complicated legal relationship where they have some kind of cross-licensing agreements that allowed Intel to include 64-bit mode in its CPU lineup. Now, why can’t AMD include that Hyper-Threading nifty trick that Intel has where a dual-core can supervize 4 “physical threads”? Is there a way to completely disable the HyperThreading mechanism in Intel’s chip and put them against AMD in a core-for-core battle? That’ll be interesting.

  92. I’m disappointed in the Athlon II’s power consumption. I guess having four power states and no L3 cache didn’t really do anything.

  93. Just figured I’d throw this out there since the Gigabyte DES utility was mentioned. I’ve been messing around with it on my WHS box and it does provide notable power draw savings even over a CPU with voltage tweaked to one step in BIOS above the minimum needed for very good stability (Linpack, 24 hour P95 small FFT and blend.) Idle draw was reduced by ~5W and load by ~15W with the power saving options in DES all set to best including level 3 voltage setting. One weird thing about the utility is that it defaults to ‘CPU throttling – Off’ which results in almost no power savings at load. Changing that setting drops the load draw by ~15W and also drps CPU temps by ~4C. What’s even more interesting about that is the voltage reported by CPU-Z doesn’t change appreciably so there must be some pretty good efficiency tricks going on in the mobo CPU voltage supply section.

    It’s a neat utility and will make me look toward Gigabyte boards first in the future.

  94. Scott Wasson, i don’t understand you!
    In every single AMD review you mention that the problem is the AMD disk controller but in every single review you keep using a motherboard with an AMD chipset. WHY? Is someone forcing you to do this?
    Last time i checked, you can still buy nVidia based AMD motherboards.
    Why do you let a weak southbridge cloud people’s judgment over a CPU purchase? Please avoid AMD chipsets for reviews until AMD releases SB800.

  95. Yeah…we need something like “The Stig” here at TR.
    Some mad overclocking geek ,deep in the restricted area(51?) of the Damage Labs who can really see what these puppies can do.
    If there ever was a “dream CPU” to drool over…the i7 975 is a worthy contender…yum.

  96. Yeah I more or less ignore overclocking results from TR – no offense guys.

  97. Intel is going to push LGA1336 to server or high-end workstation as fast as they can, hence the pending cancellation of the i7-920. It makes me chuckle a bit at the early adoptors of LGA1336 who thought it was a great upgrade path but didn’t realize that it’s the new LGA771, after Lynnfield launch CPUs for LGA1336 will be $500+ only.

  98. Look at actual power draw not TDP. The 45nm AMD chips are far better than past AMD chips especially at load (check various sites too, some show improvements at idle as well) and finally on par with Intel CPUs of the same node unlike 65nm CPUs. Now if you don’t need the additoinal clock speed a 5050e or so might make sense just based on cost but there’s not much power advantage of going with a 65nm AMD CPU any more, even a ‘low power’ 45W rated one. Maybe they’ll come out with lower speed lower TDP 45nm CPUs, that might make the TDP watchers happy 😉

  99. I think they should recruit a separate blogger or reviewer specialising in overclocks.

  100. I love my E8400 as well, but I don’t think TR aims for the “extreme OC”. From all the OC results on TR reviews, it always is what can simply be done, not what takes 6 hours slaving in BIOS to get.

    I’ve always felt the OC section is more what you could expect to get not what skill and expertise take to coax out. If the counter-argument is “It gauges the CPU’s real potential for OC” this isn’t that far off from the people who use LN to “Gauge CPU potential”.

  101. Aaaahh, Scott

    Your overclocking skills leave a lot to be desired.
    An E8400 E0 can run at 8*500 = 4GHZ with a mere v1.272, stable, on air cooling (others have managed it with less), 24/7. Would it outperform an i920 at those settings?
    §[<<]§ My bandwidth exceeds 10GB/s, with some decent memory, at that speed. §[<<]§ Perhaps an article should be made about the detailed BIOS options that can allow such a high OC with such low voltage and the subsequent special value the Core 2 Duo's have to offer for gamers.

  102. Thanks a lot for that review Scott…it was a good read and very timely.
    Interesting to see the total platform costs compared….Core i7 platform appears more affordable now.
    Nice results for the new Phenom 250….hopefully AMD will sell a heap of them and get back on its financial feet.

  103. For anyone interested in doing some benchmarking with Far Cry 2 themselves: It has a very nice benchmarking tool that does most of the work for you.

  104. I think you’re missing my point. It’s obviously due to the cache, but the other games show a negligible difference. That calls the significance of the Athlon II’s “handicap” into question. Is it really a handicap if, of all things, only Crysis Warhead shows it to be? We don’t really know.

    More important than the difference in cache size, is the difference in the *[

  105. While its looking like the i7 is going to be the performance leader until the next die shrink, I can only wonder about the future of the lga 1366 socket with Lynnfield just around the corner.

    Rumors seem to point to the fact that a Lynnfield chip, clocked reasonably, would meet or out do the lower end i7 models, thanks to some tricks on Intel’s part. Combined with cheaper motherboards and dual channel RAM, I could see how such a chip could steal some market share from the i7.

    Is Intel really positioning 1366 to house only the $300+ chips?

  106. Cache size. Scott mentions this specifically and the results, both for AMD and Intel CPUs, seem to validate the theory.

  107. I cannot stress enough how glad I am that minimum FPS is measured in games by TR, and that the benchmarks include pretty much all available CPUs at once. I don’t put much stock in the game benchmarks of any other sites, for those reasons.

    But why no minimum FPS in HL2 episode 2? It almost defeats the purpose of having so much other useful information (as far as games go, but that’s obviously important to lots of people).

    I’m not asking for all the CPUs to be redone for it (that’s obviously not worth it), but I’m just curious, for a specific reason that only just now became apparent.

    What really intrigues me is that the minimum FPS on two of the three other tests shows that the Athlon II is pretty much negligibly “slower” than the Phenom II dual-core. Part of it could even be clock speed.

    However, there is a pretty drastic difference between the two in Crysis Warhead, which confuses things.

    If I had to guess, I would assume it’s largely inconsequential, knowing how well “optimized” even Warhead turned out to be. But nonetheless, it’s no fun just assuming things when it ultimately may determine how people spend their money.

  108. The X2 550 and X3 720 are both extremely attractive processors. Great performance PLUS you have a friendlier upgrade path. I just couldn’t see justifying a Core 2 build of any sort right now unless you strongly favored some of the non-gaming benchmarks. Even then… I think the upgrade path would sway me. Phenom II or Core i7, a very split market.

  109. Ah, this made my day…


    ::goes back to reading and looking at graphs:::