- Page 1nForce 5 and AM2 Revealed
- Page 2 LinkBoost
- Page 3 LinkBoost Results
- Page 4 Quake 4 Results
- Page 5 Counter-Strike: Source Results
- Page 6 Call of Duty 2 Results
- Page 7 3DMark & Battlefield 2 Results
- Page 8 Testing
- Page 9 SLI Memory / EPP
- Page 10 Networking
- Page 11 NVIDIA GPU Ex Results
To test the platform, we have a Foxconn C51XEM2AA motherboard based on the top-of-the-line 590 SLI chipset. The CPU used is an Athlon 64 FX-62 which is a dual-core 2.8GHz processor with 1MB of Level 2 cache per core. Thanks to Corsair we had two sets of memory to use – 2GB of 6400C4 (rated up to 800MHz 4-4-4-12) and 2GB of 8500C5 (rated up to 1,066MHz 5-5-5-15). This meant we could experiment with frequency as well as latency to see how it affected performance.
For comparison, we used our tried and tested nVidia graphics card test rig which is an Asus A8N32-SLI with an FX-60 and 2GB of 3500LLPRO (rated up to 436MHz 2-3-2-6). We overclocked the CPU to 2.8GHz (by increasing to a 14x multiplier and a 1.5V core voltage) in order to match the FX-62 in our AM2 test setup. Memory was kept at 400MHz, which is the default.
Using ‘SpodeMark 06’, we compared our spanky new AM2 rig to our FX-60 test bed. This gave us a like for like comparison, with the only differences being the new DDR2 memory and the nForce5 chipset.
We used the new 91.27 graphics drivers which also use the new nVidia control panel (worth keeping a look out for).
Everything in the BIOS on both machines was kept at the default (auto in most cases). This meant the memory was running at a 2T command rate in both instances. We actually had a lot of trouble getting 1T timings to work on the AM2 platform and we know for a fact this can make a big difference to performance.
An exception to this rule was the “NVIDIA GPU Ex” setting, which was switched on in the Foxconn BIOS. This is apparently an optimisation specifically for SLI and the 91.27 drivers (hence being turned off by default).
Both machines used a set of default clocked 7900 GT graphics cards, running in SLI.
Before going in to the specifics, it is worth noting that 3DMark showed little or no variance in each setting and as such proved to be a fairly useless comparison benchmark. Equally, it also proved that it is only testing the graphics cards and not the rest of the system. So any analysis is based on the gaming results only. For those interested, we have still included these results.
The first thing we did was to test using the 6400C4 memory, running at 800MHz. This showed on average close to seven per cent increase in performance, with as much as 14 per cent in places. This is an instant win for AMD, as there was initial concern that the relatively higher latencies of DDR2 over DDR may well cause performance issues.
Sign up for the newsletter
Get news, competitions and special offers direct to your inbox
Second up, we put in the 8400C5 memory. Once SLI memory was switched on, this ran at 940MHz. This is higher latency than the 6400C4, but the extra memory bandwidth proved to be more important than latency with an average of just over nine per cent increase in performance over a DDR system. This was as high as 19 per cent difference when at 2,048 x 1,536 with 4x FSAA and 8x AF in Counter-Strike: Source, increasing from 39.5 to 47.1 frames per second. That’s a definite improvement worth noting.
Finally, I decided to test what “NVIDIA GPU Ex” does. According to nVidia this stands for “GPU Extra Performance”. I found this a little concerning at first; as I know that my idea of an optimisation can often differ from the graphics chip companies’. The only game that was slightly affected by turning this feature off was Quake 4 with a drop of five frames per second. This is indeed a minor optimisation.
nForce5 isn’t a revolutionary new product. If it was, nVidia might have considered amalgamating the SPP and MCP of the 590 SLI into one package. Instead, this is basically a revision of an already excellent product – nForce4. I have no complaints that it has decided to add in some extra features that shouldn’t cost us any more than nForce4 products do. nForce4 has been amazingly successful and I expect nForce5 to do just as well, with or without the help of useless features like LinkBoost.
AM2 performance is looking healthy offsetting the higher latency of DDR2 quite happily. However, if the hype is to be believed it won’t be fast enough to compete with Intel’s Core 2 Duo. I will return to both AM2 and Core 2 Duo at a later point, hopefully with an nForce5 Intel Edition to boot.