A quick look at AMD’s 65nm Athlon 64 X2 processors

AMD HAS BEEN PROMISING for some time to deliver chips made on its 65nm fabrication process by the end of 2006. The year’s end is perilously close, but a small package arrived at Damage Labs last week bearing evidence of AMD’s success: a pair of 65nm Athlon 64 X2 processors. Looks like they’ve just made it in under the wire. In fact, Athlon 64 X2 processors built in this 65nm process are filtering out into the market now.

Process shrinks often bring with them some nice benefits, because smaller chips tend to require less voltage, consume less power, and generate less heat. They also sometimes allow more headroom for clock frequency increases. The question is: how is AMD doing on these fronts? What benefits does its 65nm process bring to the Athlon 64 X2? Let’s have a look.

The Athlon 64 X2 at 65nm: Less of the same
First things first: the Athlon 64 X2’s conversion to 65nm is a die shrink and not much more than that. AMD says these chips ought to perform just like their counterparts at 90nm, so one shouldn’t expect any notable performance improvements. The company does have microarchitectural improvements planned for its so-called “K8L” design, which is expected to debut around the middle of next year.

Between now and then, AMD will reap two main benefits from this conversion. First, the chips themselves are smaller, down from 183mm² at 90nm to 126mm² at 65nm. The 65nm and 90nm versions share the same estimated transistor count of 153.8 million. Second, these processors are manufactured exclusively at AMD’s Fab 36 facility using 300mm wafers. These wafers have 2.25 times the area of the 200mm wafers produced at AMD’s adjacent Fab 30 facility. Taken together, the smaller chips and larger wafers should make for much lower per-chip production costs, provided AMD is able to get good yields out of its 65nm process. That cost savings is especially crucial because Intel has had multiple fabs producing 65nm chips on 300mm wafers for quite some time now.


A pair of 65nm Athlon 64 X2 processors

The two 65nm Athlon 64 X2 processors we received for testing are the 4800+ and 5000+ models, both intended for Socket AM2 motherboards. Although nothing much has changed in terms of CPU performance at 65nm, the transition gives AMD an opening to jack with its CPU model numbers, and of course, it didn’t let the opportunity pass without action. These processors have the ability to support half-step multipliers, so clock frequencies can now be controlled in 100MHz increments. Combine this ability with AMD’s apparent determination to kill off or at least significantly reduce the number of Athlon 64 X2 processors with 1MB of L2 cache per core, and you have a recipe for a substantially revised set of model numbers. The new lineup of Athlon 64 X2 models at 65nm looks like so:

Model

Clock speed L2 cache
(per core)
TDP Price
Athlon 64 X2 4000+ 2.1GHz 512KB 65 W $169
Athlon 64 X2 4400+ 2.3GHz 512KB 65 W $214
Athlon 64 X2 4800+ 2.5GHz 512KB 65 W $271
Athlon 64 X2 5000+ 2.6GHz 512KB 65 W $301

Before it was canceled, AMD offered an Athlon 64 X2 4800+ (at 90nm) with 1MB of L2 cache per core and a 2.4GHz clock speed. Now, the 4800+ model number is back, inhabited by a 2.5GHz processor with 512K of L2 cache per core. Cache size and clock frequency don’t always affect performance in the same way, or equally, but I doubt most folks will lose any sleep over these differences.

Notice that all of the Athlon 64 X2 chips at 65nm have a thermal design power (TDP) rating of 65W, down from 89W for standard-issue Athlon 64 X2s at 90nm. The range of recommended core voltages is also lower, from 1.30-1.35V at 90nm to 1.25-1.35V at 65nm. Due to the lower TDP rating, these processors get the same “Energy Efficient” label that AMD previously reserved for specially manufactured 90nm parts. So the full, official name of the 5000+ CPU we’re testing is “Athlon 64 X2 5000+ Energy Efficient,” burdened with four last names like a second-generation feminist. (What was that, Mr. Snerdley?) Fortunately, none of the 65W Energy Efficient models carry a premium any longer, so the 65nm parts are priced identically to their 89W/90nm counterparts.

A trio of late 90-nano additions, too
We’d be remiss not to point out the addition of a few new 90nm models to the Athlon 64 lineup, as well. AMD has stealthily slipped these into its product mix, just below the Athlon 64 FX-62 and just above the 5000+. They are:

Model

Clock speed L2 cache
(per core)
TDP Price
Athlon 64 X2 5600+ 2.8GHz 1MB 89 W $505
Athlon 64 X2 5400+ 2.8GHz 512KB 89 W $485
Athlon 64 X2 5200+ 2.6GHz 1MB 89 W $403

Yes, you’re looking at two new Athlon 64 X2 models with 1MB of L2 cache per core. I won’t pretend to understand AMD’s strategy here, which seems to involve confusing the customer, then conquering him. The most confusing model of the bunch may be the 5600+, which shares the same cache config and clock speed as the Athlon 64 FX-62, but lacks the FX-62’s unlocked upper multiplier. The 5600+, though, has an 89W TDP, while the FX-62’s is 125W. And the 5600+ costs over $200 less than the FX-62. Go figure.

Unfortunately, we don’t have any samples of the 5600+, 5400+, or 5200+ for testing, so that will have to wait for another day.

 

Test notes
Our aim in the following tests is two-fold: to determine how much the conversion to 65nm has affected the energy efficiency of the Athlon 64 X2, and to see how much overclocking headroom these new parts have. We’re not really testing performance today, so consider this a down payment on a future CPU comparo that has performance in its scope. For now, if you’re wondering about processor performance, I suggest you check out our Core 2 Extreme QX6700 review for a broad-based comparison or our Quad FX review for a more focused look at very high-end solutions.

I have included some results from Core 2 Duo processors in our tests, but they come with a caveat. We’re testing the AMD processors with a relatively low-power motherboard, because our other option on Socket AM2, the Asus M2N32-SLI Deluxe, had some problems (innately high power consumption due to dual 16-lane PCIe 16X slots and dual Nvidia core-logic chips, as well as some crashes with Cool’n’Quiet enabled.) I had hoped to test the Intel CPUs with an Asus P965 Express-based motherboard because the P965 was the lowest power consumer in our Core 2 chipset roundup, but I fried that board attempting to update its BIOS. Doh! Instead, I used Intel’s D975XBX2 mobo, which is fairly efficient but still has some additional auxiliary chips and expansion slots that our AMD mobo lacks.

Also, you’ll see that we’ve test a 90nm version of the Athlon 64 X2 3800+ labeled “EE SFF” in the table below. That’s the Energy Efficient Small Form Factor version of the 3800+ with a 35W power rating, not the 65W Energy Efficient version.

You will see test results for the Core 2 Duo E6400 processor in this article. That processor came to us courtesy of the fine folks at NCIX. Those of you who are in Canada will definitely want to check them out as potential source of PC hardware and related goodies.

Finally, because we’re focusing on power testing, I’ve only included processors for which I have actual samples—no simulated results achieved by underclocking higher speed-grade CPUs. I did underclock the Core 2 Extreme X6800 to E6600 speeds for our one performance test, though.

Our testing methods
Our test systems were configured like so:

Processor Core 2 Duo E6300 1.86GHz
Core 2 Duo E6400 2.13GHz
Core 2 Extreme X6800 2.93GHz
Core 2 Extreme QX6700 2.66GHz
Athlon 64 X2 3800+ EE SFF 2.0GHz
Athlon 64 X2 4600+ EE 2.4GHz
Athlon 64 X2 4800+ EE 65nm 2.5GHz
Athlon 64 X2 5000+ EE 65nm 2.6GHz
Athlon 64 X2 5000+ 90nm 2.6GHz
Athlon 64 FX-62
2.8GHz
System bus 1066MHz (266MHz quad-pumped) 1GHz HyperTransport
Motherboard Intel D975XBX2 Asus M2NPV-VM
BIOS revision BX97520J.86A.2333 0603
North bridge 975X MCH GeForce 6150
South bridge ICH7R nForce 430 MCP
Chipset drivers INF Update 8.1.1.1010
Intel Matrix Storage Manager 6.2
ForceWare 8.26
Memory size 2GB (2 DIMMs) 2GB (2 DIMMs)
Memory type Crucial Ballistix PC2-6400
DDR2 SDRAM
at 800MHz
Crucial Ballistix PC2-6400
DDR2 SDRAM
at 800MHz
CAS latency (CL) 4 4
RAS to CAS delay (tRCD) 4 4
RAS precharge (tRP) 4 4
Cycle time (tRAS) 12 12
Audio Integrated ICH7R/STAC9221D5
with SigmaTel
5.10.5208 drivers
Integrated nForce 430 MCP/AD1986A with
Soundmax 5.10.2.4153 drivers
Hard drive Maxtor DiamondMax 10 250GB SATA 150
Graphics GeForce 7950 GX2 1GB PCI-E with ForceWare 93.71 drivers
OS Windows XP Professional x64 Edition
OS updates DirectX 9.0c update (October 2006)

Thanks Crucial for providing us with memory for our testing.

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:

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.

 

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 draw 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.) In this case, we tested power use while running Cinenbench’s multithreaded rendering benchmark. We chose this benchmark because it scales up pretty well with multicore processors, and it runs in a reasonable amount of time.

Here’s the power draw of the various processors before, during, and after the Cinebench test run. I’ve broken the results into three different graphs and overlaid the 65nm Athlon 64 results on each to make comparison easier.

Several things are obvious right off the bat, the most prominent of which is the 65nm processors’ improved power efficiency over their 90nm predecessors. The most direct comparison here is between the two versions of the Athlon 64 X2 5000+. The 65nm version accomplishes the same work in the same time period with lower power draw both while rendering and at idle.

We’ve also learned that our early Core 2 Duo E6300 sample is something of a runt, requiring more power than our newer retail Core 2 Duo E6400. I decided to go ahead and include its results here as a reminder of how much power consumption can vary from one chip to the next.

We can break down the power consumption data in various useful ways. We’ll start with a look at idle power, taken from the trailing edge of our minute-long test period, after all CPUs had completed the render.

With the 5000+ as our guide, we can see that 65nm power consumption is lower by a few watts, even at idle, than the regular 90nm parts. The 90nm Energy Efficient processors, though, match the 65nm models.

As for the Core 2 processors, their higher idle power use may be in part due to higher overall platform or motherboard power draw, but it may also be related to the aggressiveness of the power-saving mechanisms involved here. AMD’s Cool’n’Quiet takes the CPU all the way down to 1GHz, but Intel’s SpeedStep only slows the Core 2 processors to a minimum of 1.6GHz.

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 all of the processors were rendering.

Again, with the 5000+ as our point of reference, we can see that the 65nm chips look pretty good. The 90nm Energy Efficient parts have even lower peak power draw, but those are running at lower clock speeds and tend not to have much frequency headroom.

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.

The 65nm 5000+ achieves a substantial reduction in energy use versus the 90nm 5000+.

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 then 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 be our best indicator of how AMD’s 65nm process transition changes things. The die shrink brings a modest increase in power efficiency, not a revolution. The 65nm 5000+ improves on its 90nm counterpart, but it only barely surpasses the 90nm Energy Efficient 4600+.

The top spots here are taken by Intel’s Core 2 Extreme processors, thanks in part to their modest power draw but more prominently because of their outstanding performance. Because they finish rendering the scene sooner, they’re able to consume less energy for the duration of the task. AMD’s 65nm processors have improved on one part of the so-called “performance per watt” equation by reducing power draw, but not the other.

 

Overclocking
In order to check out the overclocking headroom of the new 65nm processors, we moved over to an Asus M2N32-SLI Deluxe motherboard, which has a robust set of overclocking options in the BIOS. The cooler was just a stock unit from AMD.

I started my overclocking adventures with the 5000+. I’ll spare you the graphic details, but after quite a bit of experimenting, I was able to get it stable at 2925MHz with a 225MHz HyperTransport clock by raising the voltage to 1.425V in the BIOS—which read as 1.472V in a CPU-Z and in the Asus PC Probe utility. I used two instances of Prime95’s torture test to verify stability, and I’d say the 5000+ was a pretty solid overclock at this speed. The 4800+ came close to the 5000+, hitting 2875MHz on a 230MHz bus using the same 1.425V voltage setting in the BIOS.

At these speeds, with dual Prime95 instances running, the 5000+ registered 55°C, while the 4000+ hit 56°C.

These chips are a little odd as overclockers, as early samples from a new fab process sometimes tend to be. They 5000+, for instance, would POST at 3120MHz and boot into Windows just fine, but when I ran Prime95, it wouldn’t just throw a computational error when things got bad—it would up and reboot the whole system. There seemed to be little room between stability and “prone to fiery death.”

Just to verify that the overclocked CPUs scaled well in terms of performance, here’s a look at Cinebench rendering speed. Looks like the overclocked chips scale up more or less as expected. However, the Athlon 64 X2 can’t match the Core 2 clock for clock, as the matchup of overclocked 5000+ versus Core 2 Extreme X6800 reminds us.

 
Conclusions
AMD has consistently touted its model of continuous improvement for chip fabrication techniques, and that model has served it well in the recent past. The products of a well-refined 90nm process have more or less held their own on power efficiency against 65nm processors from Intel, except for at the very high end. Chips produced with AMD’s special low-power 90nm process tweaks, like the Energy Efficient 3800+ and 4600+ we tested, have been especially impressive. The flip side of that coin is that AMD’s move from 90nm to 65nm does not instantly produce huge improvements in energy efficiency or clock frequency headroom. Instead, the 65nm transition brings a welcome but incremental improvement in power consumption over current Athlon 64 X2 products and little or no additional headroom. If you’re looking to buy an Athlon 64 X2, then I’d definitely try to grab the 65nm version, but don’t expect miracles from it.

AMD will no doubt continue its trajectory of gradual improvement to its process tech, leading to additional power savings and headroom at 65nm in the future. For now, though, making it over this hurdle isn’t nearly enough to overcome the sizeable performance gap between the Core 2 Duo and Athlon 64 X2—and it does very little to counter the formidable presence of Intel’s quad-core Core 2 Extreme QX6700, which is coming in lower-power, lower-cost Core 2 Quad form soon.

That said, AMD could make one considerable stride toward countering Intel’s quad-core chips and salvaging its poorly received Quad FX platform by replacing the Athlon 64 FX-70-series processors—and their scorching 125W TDP ratings—with chips made on this 65nm process. A pair of FX-74 CPUs, each with a 65W TDP, would be the perfect answer to Intel’s Core 2 Extreme QX6700 and its 130W TDP. Pair ’em up with a decent, affordable motherboard, and Quad FX might start grabbing some attention—for the right reasons. Note to AMD: You need to do this. Stat. 

Comments closed

Pin It on Pinterest

Share This

Share this post with your friends!