Chief among the new developments is the availability of cheaper—err, less expensive—quad-core options like the Core 2 Quad Q6600 and Athlon 64 FX-70. Intel and AMD like to showcase their top performing chips in order to show off what they can do, but top-speed-grade processors are rarely the best values. What’s more, we’ve found that practically any top-speed-grade incarnation of a processor tends to be in a rough spot with respect to heat and power consumption. Lower speed grades promise higher performance per watt.
For instance, the basic power and heat rating, or TDP, of the Core 2 Extreme QX6700 is 130W. Although it’s the same technology and runs only 266MHz slower, the Core 2 Quad Q6600’s TDP is an Al Gore-approved 105W. Officially, the Athlon 64 FX-70’s thermal power rating is the same considerable 125W per chip (in a two-chip solution) as its bigger brother, the FX-74, but we had a hunch the 2.6GHz FX-70 wouldn’t be the same class of double-barreled blowtorch as the 3GHz FX-74. There is, of course, one way to find out: test ’em. And so that’s what we’ve done.
We’ve also recently made the transition to Windows Vista for our test platforms, a move that promises to take better advantage of these quad-core system architectures in various ways. Join us as we fire up our widely multithreaded suite of test applications, many of them 64-bit executables, to see which quad-core solution offers the best mix of price, performance, and energy efficiency.
Four cores, two chips, and how many sockets?
Since we’re already reviewed both of these basic technologies, I will spare you most of the gory details here. If you’d like more info on the “Kentsfield” quad-core processors from Intel, go have a look at our Core 2 Extreme QX6700 article. And if you’re unfamiliar with AMD’s Quad FX platform, you can find our take on it right here. The Cliff’s Notes version goes like this: neither of these products feature a truly native quad-core processor. Kentsfield is comprised of two Core 2 Duo chips cuddled up together in a single package. You get four cores in total, but in groups of twoand the two chips have to communicate with one another by means of the system’s front-side bus. The great advantage of this scheme is that Core 2 Quad variants can use the same motherboard infrastructure as the Core 2 Duo.
Quad FX, meanwhile, uses a variation of the Opteron infrastructure, complete with dual CPU sockets and a pair of dual-core processors, in order to reach four cores in a desktop platform. Quad FX’s main concession to desktop use is that it doesn’t require registered DIMMs, just run-of-the-mill DDR2 memory. Although it makes for larger, pricier motherboards, Quad FX’s dual-socket arrangement arguably offers better integration between two dual-core chips thanks to AMD’s HyperTransport interconnect. Because each CPU has dual memory channels hanging off of it, Quad FX also has twice the peak theoretical memory bandwidth of the Core 2 Quador more, if you count the Core 2 Quad’s front-side bus as the primary constraint.
Making the most of the Quad FX platform’s memory architecture, however, can be tricky. Since each CPU has its own integrated memory controller, access to RAM can be very quick, but if CPU 1 has to access data in the memory space controlled by CPU 2, memory access will be considerably slower. Fortunately, Windows Vista incorporates kernel improvements for non-uniform memory access (NUMA) architectures like Quad FX, so testing with this OS ought to let the Quad FX platform put its best foot forward.
As I’ve mentioned, the Core 2 Quad’s motherboard situation is simple: it will work with most newer motherboards intended for the Core 2 Duo. Quad FX’s motherboard situation is also simple, but in a different way: you can choose any motherboard you want, so long as it’s the Asus L1N64-SLI WS. This beast is indeed a decent enthusiast-class motherboard, but it’s a little too, er, enthusiastic for some of us, with a total of four PCIe x16 slots, 12 SATA ports, and dual Nvidia chipsets that draw nearly as much power as a theater screening An Inconvenient Truth. Here’s another inconvenient truth: the L1N64-SLI WS costs $350.
That little wrinkle complicates life for the Quad FX platform. AMD has said that more Quad FX motherboards are on the way, but as far as we know, no others are immediately imminent. For now, Quad FX remains tied to one board.
AMD has attempted to keep the price equation in line by selling Quad FX processors at relatively reasonable prices. The original plan was to sell them in pairs at $599 per pair for the FX-70, $799 per pair for the FX-72, and $999 per pair for the FX-74. In fact, they’re still listed in pairs on AMD’s price sheet today. If you go to buy ’em at Newegg, though, you’ll find the CPUs priced and selling individually at a bit of a premium over list. Two FX-70s will set you back about $614, while two FX-72s are $820, and a couple of FX-74s total up to $1040. That’s still not a bad price for the FX-70, relatively speaking, but then you have to take the $350 mobo into account. All told, the cheapest Quad FX mobo and processor combo will add $964 to your next Visa bill, while the most expensive one rings up just ten bucks short of $1400.
By contrast, quality mobos for Intel’s quad-core processors can be had for under $180well under, if you’re not too concerned about high-end features like dual PCIe x16 slots. Probably the most apt comparison to the L1N64-SLI WS, if there is such a thing, is eVGA’s version of the Nvidia nForce 680i SLI motherboard, which currently lists for $220. The combination of a Core 2 Quad Q6600 and this mobo costs $1066 at Newegg. Switching to a Core 2 Extreme QX6700 would raise the total to $1190 for the mobo and CPU.
None of these are exactly value products, but Quad FX systems wind up costing more than you might think, if you were to start with the AMD processor price list. Of course, the question of value becomes cloudy when talking about four-core rigs whose two main components add up to a grand or more. I should say right here, at the outset, that few of us really need four cores, especially since few of today’s applications can really put ’em all to use at once. Glance over the test results in this article if you want to see that fact illustrated. We have now compiled a nice set of apps for our test suite, most of which can use four cores, but we had to dig a little bit to find them. So the results you’re about to see are as much about potential as about what most folks would get out of a quad-core system today—something to keep in mind.
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.
We’ve tested our quad-core systems against one another and against the top dual-core desktop processors, as well. In some cases, getting the results meant simulating a slower chip with a faster one. For instance, our Core 2 Duo E6700 is actually a Core 2 Extreme X6800 processor clocked down to the appropriate speed. Its performance should be identical to that of the real thing. Similarly, our Athlon 64 FX-72 results come from an underclocked pair of Athlon 64 FX-74s.
Our test systems were configured like so:
|Processor|| Core 2 Duo E6700 2.66GHz
Core 2 Extreme X6800 2.93GHz
Core 2 Quad Q6600 2.4GHz
Core 2 Extreme QX6700 2.66GHz
|Athlon 64 X2 6000+ 3.0GHz (90nm)||Athlon 64 FX-70 2.6GHz
Athlon 64 FX-72 2.8GHz
Athlon 64 FX-74 3.0GHz
|System bus||1066MHz (266MHz quad-pumped)||1GHz HyperTransport||1GHz HyperTransport|
|Motherboard||Intel D975XBX2||Asus M2N32-SLI Deluxe||Asus L1N64-SLI WS|
|North bridge||975X MCH||nForce 590 SLI SPP||nForce 680a SLI|
|South bridge||ICH7R||nForce 590 SLI MCP||nForce 680a SLI|
|Chipset drivers||INF Update 18.104.22.1680
Intel Matrix Storage Manager 6.21
|ForceWare 15.00||ForceWare 15.00|
|Memory size||2GB (2 DIMMs)||2GB (2 DIMMs)||2GB (4 DIMMs)|
|Memory type||Corsair TWIN2X2048-6400C4
DDR2 SDRAM at 800MHz
DDR2 SDRAM at 800MHz
|Crucial Ballistix PC6400
DDR2 SDRAM at 800MHz
|CAS latency (CL)||4||4||4|
|RAS to CAS delay (tRCD)||4||4||4|
|RAS precharge (tRP)||4||4||4|
|Cycle time (tRAS)||12||12||12|
|Audio||Integrated ICH7R/STAC9274D5 with
Sigmatel 22.214.171.12474 drivers
|Integrated nForce 590 MCP/AD1988B with
Soundmax 126.96.36.19900 drivers
|Integrated nForce 680a SLI/AD1988B with
Soundmax 188.8.131.5200 drivers
|Hard drive||Maxtor DiamondMax 10 250GB SATA 150|
|Graphics||GeForce 7900 GTX 512MB PCIe with ForceWare 100.64 drivers|
|OS||Windows Vista Ultimate x64 Edition|
Thanks to Corsair for providing us with memory for our testing. Their products and support are far and away superior to generic, no-name memory.
Also, all of our test systems were powered by OCZ GameXStream 700W power supply units. Thanks to OCZ for providing these units for our use in testing.
The test systems’ Windows desktops were set at 1280×1024 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:
- SiSoft Sandra XI 2007.2.11.17 64-bit
- CPU-Z 1.39
- POV-Ray for Windows 3.7 beta 19a 64-bit
- Cinebench 9.5 64-bit Edition
- Windows Media Encoder 9 x64 Edition
- picCOLOR 4.0 build 598 64-bit
- 3DMark06 1.0.2
- notfred’s Folding benchmark CD 10/31/06 revision
- The Panorama Factory 4.4 x64 Edition
- CASE Lab Euler3d CFD benchmark 2.2
- MyriMatch proteomics benchmark
- Valve Source Engine particle simulation benchmark
- Valve VRAD map build benchmark
- LAME MT 3.97a 64-bit
- 3DMark06 1.10
- The Elder Scrolls IV: Oblivion 1.1
- Rainbow Six: Vegas 1.02
The tests and methods we employ are generally publicly available and reproducible. If you have questions about our methods, hit our forums to talk with us about them.
The Elder Scrolls IV: Oblivion
We tested Oblivion by manually playing through a specific point in the game five times while recording frame rates using the FRAPS utility. Each gameplay sequence lasted 60 seconds. 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 frames 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.
For this test, we set Oblivion’s graphical quality to “Medium,” but with HDR lighting enabled and vsync disabled, at 800×600 resolution. We’ve chosen this relatively low display resolution in order to prevent the graphics card from becoming a bottleneck, so differences between the CPUs can shine through.
Notice the little green plot with four lines above the benchmark results. That’s a snapshot of the CPU utilization indicator in Windows Task Manager, which helps illustrate how much the application takes advantage of up to four CPU cores, when they’re available. I’ve included these Task Manager graphics whenever possible throughout our results. In this case, Oblivon really only takes full advantage of a single CPU core, although Nvidia’s graphics drivers use multithreading to offload some vertex processing chores.
I guess you could say things don’t start well for the Quad FX rigs, but all of these systems can run Oblivion smoothly without breaking a proverbial sweat. You simply don’t need a high-end processor to run many of today’s games quite well. This one is mostly a speed contest involving one or two cores, as the dual-core Core 2 Extreme X6800’s first-place finish indicates. Overall, the Intel systems sweep the top spots.
Rainbow Six: Vegas
Rainbow Six: Vegas is based on Unreal Engine 3 and is a port from the Xbox 360. For both of these reasons, it’s one of the first PC games that’s multithreaded, and ought to provide an illuminating look at CPU gaming performance.
For this test, we set the game to run at 800×600 resolution with high dynamic range lighting disabled. “Hardware skinning” (via the GPU) was disabled, leaving that burden to fall on the CPU. Shadow quality was set to very low, and motion blur was enabled at medium quality. I played through a 90-second sequence of the game’s Terrorist Hunt mode on the “Dante’s” level five times, capturing frame rates with FRAPS, as we did with Oblivion.
The Task Manager plots and the benchmark results agree: Rainbow Six: Vegas doesn’t really seem to use more than two cores to any great effect (at least on the PC). Regardless, all of our test rigs are able to run the game very smoothly, with very little separation between them performance-wise.
Valve Source engine particle simulation
Next up are a couple of tests we picked up during a visit to Valve Software, the developers of the Half-Life games. They’ve been working to incorporate support for multi-core processors into their Source game engine, and they’ve cooked up a couple of benchmarks to demonstrate the benefits of multithreading.
The first of those tests 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 is limited by the available computing horsepower. Valve’s particle system distributes the load across multiple CPU cores.
Both the Intel and AMD systems scale up nicely from two cores to four here. The Core 2 Duo E6700 is the dual-core equivalent, at 2.66GHz, of the Core 2 Extreme QX6700, and its performance is just about half that of the QX6700. Similarly, the Athlon 64 X2 6000+ is a single 3GHz processor, while the FX-74 system uses two of the same; the FX-74 is nearly twice as fast as the X2 6000+. The big contrast here is that the Intel processing cores execute the particle simulation much faster, so they come out on top.
Valve VRAD map compilation
This next test processes a map from Half-Life 2 using Valve’s VRAD lighting tool. Valve uses VRAD to precompute lighting that goes into its games. This isn’t a real-time process, and it doesn’t reflect the performance one would experience while playing a game. It does, however, show how multiple CPU cores can speed up game development.
Again we see reasonably good scaling from two cores to four, but the Intel systems are faster overall.
3DMark06 combines the results from its graphics and CPU tests in order to reach an overall score. Here’s how the processors did overall and in each of those tests.
As you’ll see below, performance in the graphical tests in 3DMark06 are limited almost entirely by the graphics card (which is as it should be for a graphics benchmark). 3DMark06 also incorporates a couple of multithreaded CPU tests that involve physics simulations, AI, and game logic. That’s where we’ll see separation between the CPUs.
3DMark’s CPU test give us a glimpse of the potential for quad-core CPUs in games, and indications are good. Even the slowest quad-core solution, the FX-70, comes out well ahead of the top dual-core processor.
The Panorama Factory
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 multithreaded. I asked it to join four pictures, each eight megapixels, into a glorious panorama of the interior of Damage Labs. The program’s timer function captures the amount of time needed to perform each stage of the panorama creation process. I’ve also added up the total operation time to give us an overall measure of performance.
Virtually everything this program does is widely multithreaded, and the quad-core systems outperform the dual-cores substantially. Yet again, the Core 2 processors are faster than the Athlon 64s. Those of you who are exceptionally bright may be noticing the beginnings of a pattern there.
picCOLOR was created by Dr. Reinert H. G. MÃ¼ller of the FIBUS Institute. This isn’t Photoshop; picCOLOR’s image analysis capabilities can be used for scientific applications like particle flow analysis. Dr. MÃ¼ller 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. Eight of the 12 functions in the test are multithreaded, and in this latest revision, five of those eight functions use four threads.
Scores in picCOLOR, by the way, are indexed against a single-processor Pentium III 1 GHz system, so that a score of 4.14 works out to 4.14 times the performance of the reference machine.
Look at those trippy colors in the picCOLOR screenshot. I’ve replicated them in the lower graph. Sweet, eh? Oh, and the Core 2 processors come out ahead again.
Windows Media Encoder x64 Edition
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. 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.
We keep varying the tests and trying new things, and the Core 2 processors keep beating the Athlon 64s. This time around, the Core 2 Quad Q6600 finishes 20 seconds before the Athlon 64 FX-74.
LAME MP3 encoding
LAME MT is the multithreaded version of the LAME MP3 encoder that we discussed earlier. 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, as we have done in many of our previous CPU reviews.
With only two threads at work, the fastest dual-core solution grabs the top spot, and that’s definitely the Core 2 Extreme X6800.
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 are available.
At last, an AMD system picks up an undisputed victory as the FX-74 comes in first. The FX-70 even outperforms the Core 2 Quad Q6600 for once.
We’ve finally caved in and moved to the beta version of POV-Ray 3.7 that includes native multithreading. The latest beta 64-bit executable is still quite a bit slower than the 3.6 release, but it should give us a decent look at comparative performance, regardless.
Chalk up another win for AMD in POV-Ray, where the three FX rigs finish 1-2-3. Obviously, both the Athlon 64 and Core 2 systems scale well to four threads in this app.
The POV-Ray benchmark scene uses some features that our Chess2 scene does not, and some of those features are not (yet?) multithreaded. As a result, the dual-core processors finish stronger here. The Intel CPUs also make up some ground on the AMDs, as well.
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 recently offered to provide 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 four threads. Also, this is a newer version of the MyriMatch code than we’ve used in the past, with a larger spectral collection, so these results aren’t comparable to those in some of our past articles.
Here’s an intriguing result. Notice that the FX-72 and FX-74 are neck and neck with four threads, and the FX-72 is actually faster with one and two threads. I believe that’s the result of a quirk of Athlon 64 processors. Since they base their memory clocks on overall CPU speeds, they don’t always run their RAM at the precise frequency requested. In this case, the FX-74 is running at 3GHz, and its memory is running at one eighth that speed, or 375MHz. The FX-72, on the other hand, can run its memory at one seventh its 2.8GHz clock speed, or right at 400MHz (that’s 800MHz DDR) on the nose. MyriMatch looks to be hitting the memory subsystem hard enough that the FX-72’s faster RAM essentially offsets its lower CPU frequency. If you watch, you’ll see this effect subtly at work in some of our other results where the FX-74 isn’t much faster than the FX-72, but it’s more prominent here.
Even in this case where the Athlon 64 systems appear to be memory limited, the quad Core 2 systems scale up nicely. The Core microarchitecture’s memory disambiguation feature can be very effective in avoiding memory bottlenecks, and I expect it’s helping out quite a bit in this test.
STARS Euler3d computational fluid dynamics
Our next benchmark is also a relatively new one for us. Charles O’Neill works in the Computational Aeroservoelasticity Laboratory at Oklahoma State University, and he contacted us recently 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. (I believe the score you see there at almost 3Hz comes from our eight-core Clovertown test system.)
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 45Âº. 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. I understand the STARS Euler3D routines are both very floating-point intensive and oftentimes limited by memory bandwidth. Charles has updated the benchmark for us to enable control over the number of threads used. Here’s how our contenders handled the test with different thread counts.
The Quad FX systems struggle an inordinate amount here, as witnessed by the fact that the FX-74 is only marginally faster than the Athlon 64 X2 6000+ when four threads are in use. With two threads, the 6000+ is faster. I asked Charles to enable thread control for us because I suspected such scaling problems might be under the surface, given the results we’ve seen. This application may be a case where the program needs to be coded with an explicit awareness of NUMA architectures in order to achieve optimal performance.
Even in the dual-core processors, though, the E6700 is 30% faster than the X2 6000+.
Next, we have another relatively new addition to our benchmark suite: a slick little Folding@Home benchmark CD created by notfred, one of the members of Team TR, our excellent Folding team. For the unfamiliar, Folding@Home 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 Folding@Home 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, Folding@Home 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.
As we’ve seen before, the performance race between Intel and AMD for Folding is divided pretty strongly by work unit type. For the Tinker and Amber WUs, the Athlon 64s are easy winners. Just the opposite is true in the two Gromacs WU types, where the Intel processors reign supreme. If you want to pump out lots of work units for your team, though, any brand of quad-core setup will do quite well.
SiSoft Sandra Mandelbrot
Next up is SiSoft’s Sandra system diagnosis program, which includes a number of different benchmarks. The one of interest to us is the “multimedia” benchmark, intended to show off the benefits of “multimedia” extensions like MMX, SSE, and SSE2. According to SiSoft’s FAQ, the benchmark actually does a fractal computation:
This benchmark generates a picture (640×480) of the well-known Mandelbrot fractal, using 255 iterations for each data pixel, in 32 colours. It is a real-life benchmark rather than a synthetic benchmark, designed to show the improvements MMX/Enhanced, 3DNow!/Enhanced, SSE(2) bring to such an algorithm.
The benchmark is multi-threaded for up to 64 CPUs maximum on SMP systems. This works by interlacing, i.e. each thread computes the next column not being worked on by other threads. Sandra creates as many threads as there are CPUs in the system and assignes [sic] each thread to a different CPU.
We’re using the 64-bit version of Sandra. The “Integer x16” version of this test uses integer numbers to simulate floating-point math. The floating-point version of the benchmark takes advantage of SSE2 to process up to eight Mandelbrot iterations in parallel.
The results from this test are predictably dramatic, since the Core microarchitecture can process 128-bit SSE instructions in a single cycle. We’ll keep tracking the results, though, because we expect big things from AMD’s next-gen Barcelona core when it arrives. That chip will be a native quad-core processor with enhanced 128-bit SSE capabilities, and a pair of ’em should drop right into any Quad FX motherboard to yield an eight-core monster.
Power consumption and efficiency
We’re trying something a little different with power consumption. 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 systemthe CPU, motherboard, memory, video card, hard drives, and anything else plugged into the power supply unit. (We plugged the computer monitor and speakers into a separate outlet, though.) We measured how each of our test systems used power during a roughly one-minute period, during which time we executed Cinebench’s rendering test. All of the systems had their power management features (such as SpeedStep and Cool’n’Quiet) enabled during these tests.
You’ll notice that I’ve not included the Athlon 64 FX-72 here. That’s because our “simulated” FX-72 system is based on underclocked FX-74s, and we can’t enable Cool’n’Quiet on the L1N64-SLI WS motherboard when manual multiplier control is in use. I have included our simulated Core 2 Duo E6700, because SpeedStep works fine on the D975XBX2 motherboard alongside underclocking. The simulated E6700’s voltage may not be exactly the same as what you’d find on many retail E6700s. However, voltage and power use can vary from one chip to the next, since Intel sets voltage individually on each chip at the factory
The results are dramatically different for the various processors, especially when you look at that 460W peak for the double-barreled blowtorch, the FX-74. Rather than get ahead of ourselves, let’s analyze the data a little bit. 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.
There’s not much of a power premium at idle for the Intel quad-core processors—about 10-15W over their dual-core counterparts. The Athlon 64 X2 6000+ draws a little more power on our Asus M2N32-SLI Deluxe motherboard than the competing Intel dual-core processors, but the gap is pretty small. And then we come to the Quad FX system, where two separate CPU sockets and two full Nvidia chipsets are onboard. Idle power use is between 50 and 60W higher than the quad-core Intel systems.
Next, we can look at peak power draw by taking an average from the five-second span from 10 to 15 seconds into our test period, during which the processors were rendering.
Here’s where things become even more interesting. The first effect I want to point out here is how the top speed grades of the quad-core processors tend to consume quite a bit more power than the lower ones. That’s obviously most apparent in the case of the FX-74 versus the FX-70. Despite sharing the same 125W thermal rating, the FX-70 system peaks at “only” 334W, while the FX-74 tops out at 466W. A similar dynamic is at workin a smaller waywith the Intel quad-core processors, where the Q6600 draws 32W less the QX6700.
The other result that pops out is how much less power draw we see from the Core 2 CPUs. The Q6600’s peak power draw with multithreaded rendering is a few watts below the idle power draw of the FX-74.
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, equivalent to joules.
All of the systems are more power efficient over the time period when multithreading is employed. However, the dual-core systems draw less power over the duration of the test period than their quad-core counterparts, thanks mainly to their lower power draw at idle.
Finally, we can consider 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 chosen to identify the end of the render as the point where power use begins to drop from its steady peak. There seems to be some disk paging going on after that, but we don’t want to include that more variable activity in our render period.
We’ve computed the amount of energy used by each system to render the scene, expressed in watt-seconds. This method should account for both power use and, to some degree, performance, because shorter render times may lead to less energy consumption.
This may well be our best measure of energy-efficient performance, and sure enough, quad-core systems with multithreading require the least energy to render the scene. The key thing to notice here is that among their respective types, the lower-speed quad-core processors are most efficient. Even though it takes longer to finish rendering, the Q6600 uses less energy than the QX6700. The same is true of the FX-70 versus the FX-74, although the Quad FX systems are clearly much less efficient overall.
When the Quad FX platform first debuted, we were disappointed with its very high peak power consumption and its lack of cheaper, more power-efficient motherboard options. We also had some concerns about whether a NUMA platform was being used optimally in Windows XP. Now, time has passed, Vista has debuted, and we’ve gotten our hands on a pair of FX-70s, but not enough has changed. We were right to suspect that the lower speed grade Quad FX processors might not have the outrageous peak power consumption of the FX-74s. The FX-70s are much tamer on that front, to the tune of roughly 130W less. But Vista’s NUMA-aware kernel hasn’t allowed Quad FX to gain much, if any, performance ground on Intel’s quad-core processors. And the nice-sounding idea of a pair of FX processors for $599 doesn’t truly translate into a good value for a quad-core system. In fact, moving up to a Q6600-based system is worth every penny of the extra hundred bucks or so. As much as we like the idea of a dual-socket platform for PC enthusiasts, Quad FX isn’t likely to gain any traction unless and until we see some more reasonablein every sense of the wordmotherboards for it. Here’s hoping somebody comes out with a Quad FX mobo that has the admirable power efficiency of Opteron systems before AMD scuttles the Quad FX concept entirely.
Meanwhile, Intel’s quad-core processors are just amazing. Their performance doesn’t seem to be significantly impacted by a front-side bus bottleneck, contrary to what one might expect. They typically scale up to four threads at least as well as Quad FX systems. Their power use is restrained, and as we anticipated, the Core 2 Quad Q6600 looks to be even more power-efficient than the Core 2 Extreme QX6700. In fact, the Q6600 proved to be the most power-efficient processor in our render energy test, beating out dual-core chips in the process. That’s poster-boy-type behavior for the multi-core revolution.
As I said earlier, not everyone needs a quad-core CPU, and our gaming tests illustrated that abundantly. Even the higher-end dual-core CPUs aced those tests. For most folks, an affordable dual-core CPU will probably be more than enough processor for the next couple of years. Still, there’s something to be said for quad-core rigs, if you know you’ll put those cores to good use. A Core 2 Quad Q6600 or Core 2 Extreme QX6700 system purchased today should offer ample power for now and could achieve extensive longevity if widely multithreaded applications become more common.