AMD 90nm power consumption measured


— 12:24 AM on October 4, 2004

For ages, moving to a smaller fab process has been the key to achieving lower power consumption and higher clock speeds, but the 90nm process hasn't worked out that way, at least for Intel. After watching Intel's struggles with its rather hot and power-hungry 90nm Pentium 4 "Prescott" processors, we've been waiting with some trepidation to see whether AMD's new 90nm chips would have similar problems.

We were finally able to get our hands on a 90nm Athlon 64 3500+ this past weekend, and we've been testing it to see how it compares to the 130nm version. Since this 90nm Athlon 64 3500+ runs at the same clock speed as the 130nm Athlon 64 3500+, we were able to do a direct comparison between the chips running at 2.2GHz.

Let me give you the test setup briefly. We compared power consumption by measuring power usage for our entire testbed system, sans monitor, at the wall outlet. The testbed system included an Asus A8V Deluxe motherboard, 1GB of Corsair XMS 3200XL DDR400 memory, an NVIDIA GeForce 6800 GT graphics card, an Asus DVD-ROM drive, a Maxtor MaxLine III 250GB SATA hard drive, and an OCZ PowerStream 470W power supply. (The 130nm chip we used was actually an Athlon 64 3800+ underclocked to 2.2GHz, for what it's worth.) We also tested power consumption on a similarly-configured Pentium 4 system based on an Abit AA8 DuraMax motherboard with 1GB of OCZ DDR2 533MHz memory and the PCI-E version of the GeForce 6800 GT (all other components were the same as the Athlon 64 rig). The Pentium 4 was a Prescott 90nm core running at 3.4GHz. Update: Cool'n'Quiet was not enabled on the Socket 939 motherboard.

In order to strain the processors several different ways, I ran several of our regular CPU benchmarks and took measurements with each of them running. We've used these tests many times in reviews like this one. I included the Sphinx speech recognition benchmark, Sciencemark's "moldyn" molecular dynamics computation, and Xmpeg video encoding with the DivX codec. I also measured power consumption with the system sitting idle at the Windows XP desktop. Here are the results:

As you can see, our die-shrunk Athlon 64 came out looking pretty darned good. Of course, every chip is a little different, but these differences are probably substantial enough to suggest that AMD's 90nm Athlon 64s should generally consume less power than their 130nm counterparts.

I also took some quick temperature readings, and I'll give them to you, although I wouldn't recommend taking them as gospel. The ambient temperature in my office was about 85F/29C, and it probably varied a bit during the duration of the tests. The test rig was on an open bench and was equipped with a stock AMD CPU cooler. I recorded temps at idle and under load via the Asus Probe utility, with the Folding@Home client providing the CPU loads. The 130nm Athlon 64 idled about 49C, and it ran up to 61C after ten minutes of number crunching. The 90nm version's idle temp was 41C, and it peaked at 55C under load. All in all, a good performance from the new 90nm AMD chip.

Incidentally, there have been rumors that the new 90nm Athlon 64s incorporate some planned enhancements to the K8 core, including SSE3, better data prefetch, additional write combining buffers, and a tweaked memory controller. We haven't yet been able to confirm with AMD whether the new 90nm chips include these changes, but the preliminary indications seem to be negative. CPU-Z identifies this chip as a Winchester core and doesn't list SSE3 among the supported extensions. More tellingly, I've run a handful of synthetic memory benchmarks on the 90nm 3500+, and scores didn't differ significantly from the 130nm chip in my preliminary tests. For now, the new 90nm chips appear to be a successfully die shrunk version of the current Athlon 64: cooler, with less appetite for power, and otherwise largely unchanged.

Like what we're doing? Pay what you want to support TR and get nifty extra features.
Top contributors
1. GKey13 - $650 2. JohnC - $600 3. davidbowser - $501
4. cmpxchg - $500 5. DeadOfKnight - $400 6. danny e. - $375
7. the - $360 8. Ryszard - $351 9. rbattle - $350
10. Ryu Connor - $350
   
Register
Tip: You can use the A/Z keys to walk threads.
View options

This discussion is now closed.