NVIDIA’s New NT-Drivers for Athlon on KX133 — THG.RU

NVIDIA’s New NT-Drivers for Athlon on KX133

If you have read my recent article about VIA’s new Apollo KX133-chipset for AMD’s Athlon processor, you might remember the bad OpenGL-scores that Athlon produced under Windows NT, when running with NVIDIA’s GeForce256-card. Those scores were significantly worse than those of a Pentium III 800 and also much worse than Athlon’s own OpenGL SPECviewperf-scores under Windows98. I suspected that something fishy was going on and received a lot of emails about this issue.

Surprisingly one of the emails to this topic came directly from NVIDIA and, believe it or not, it showed that NVIDIA had reacted to the findings in my article in a very swift manner. The email contained a new driver for GeForce and Quadro, which specifically enables all of GeForce’s features on a motherboard with VIA’s Apollo KX133 chipset.

New SPECviewperf-Results

I couldn’t wait and ran SPECviewperf under NT with this new driver. The scores show that I was right expecting that Athlon should actually beat Pentium III (Coppermine) in this very OpenGL-benchmark. Have a look at the scores:

The chart shows results with Athlon 800 on KX133 with the official NT-driver rev 3.68 and then with the new driver that enables all of GeForce’s features on KX133. The blue bars show Pentium III 800-scores using driver rev. 3.68.

What can I say? Athlon running on KX133 is indeed faster than Intel’s latest Coppermine-processor running on the i840-platform! This is what we would have expected in the first place.

Quadro-Scores

I also tried NVIDIA’s Quadro, to see how the SPECviewperf scores look with this high-end OpenGL-card. Quadro is also using NVIDIA’s GeForce-chip, but it comes with 64 instead of 32 MB of onboard memory.

The results are of course higher than the GeForce-results and Athlon still remains ahead of Coppermine.

No Improvement for AMD’s 750 ‘Irongate’-chipset

Next I ran those drivers on a AMD750-platform, just to see if the other Athlon-chipset is also benefiting from NVIDIA’s new driver.

The chart shows that with the new drivers, KX133 is scoring under NT just as high as or even higher than under Windows98. Irongate’s NT-results however are still lower than the Windows98-scores.

Unfortunately there’s no change here whatsoever, which I consider as rather disappointing. As you can see in the results, Athlon is still scoring higher SPECviewperf-scores under Windows98 than under NT, which is a very strange occurrence. Usually a system scores higher SPECviewperf-results under NT than under Win98 and the opposite finding only proves that the drivers are not fully enabling GeForce’s features under NT when using AMD’s 750 platform. It is weekend right now, but I will try to get back to NVIDIA to receive an explanation for Athlon’s bad OpenGL-performance under NT when running on ‘Irongate’.

Benchmark Setup

NVIDIA Reacted Very Quickly

I am certainly impressed with NVIDIA’s quick response to my article and it makes me feel proud that my influence is strong enough to make things change so quickly. However, the above findings are a strong proof that a driver software can make good hardware look really bad if it wants. It makes me wonder in how many cases a good performing hardware product has been slowed down by software, so that it looked really bad compared to others. We should learn to be very critical about this issue.

‘Irongate’ Still Blows

I’d also like to point out that ‘Irongate’ is either much worse than we thought or that there’s still something fishy with GeForce’s NT-driver used on an AMD 750-platform. After all ‘Irongate’ was the only Athlon-chipset until very recently, and the SPECviewperf-scores show that Athlon running on this chipset is no real alternative for an OpenGL-workstation. Athlon and Irongate are now available for more than 5 months and GeForce is out for about three. In all that time Athlon looked a lot worse than it should if you ran SPECviewperf with a GeForce-based grapics-card. NVIDIA told me in the email containing the new drivers for KX133 that this new driver enables AGP4x on KX133. It cannot really be only that, because the SPECviewperf-scores under Windows98 with only AGP1x and only 100 MHz memory-bus are still higher than the initial NT-scores of Athlon with NVIDIA’s rev. 3.68 drivers. This is a strong hint that more than only AGP4x has been enabled in the new driver. Whatever other features have been enabled in GeForce’s new KX133-driver, they are still not enabled when using the AMD750-platform. There are only a few possible reasons I could think of. Either AMD750 runs so bad under NT that NVIDIA had to disable all those features (I consider this as rather unlikely), or NVIDIA didn’t consider Athlon as a CPU that would be used in an OpenGL-platform, thus they didn’t bother optimizing the NT-driver for AMD750 (in this case I blame AMD’s product marketing for sleeping), or AMD didn’t bother supporting NVIDIA properly for their development of drivers (something that I’ve heard from other companies before). I already asked NVIDIA to give me some kind of statement about that.

Epilogue

This article was addition number one for the KX133-review. I will run Athlon on an AMD750-platform with enabled ‘super bypass’ over the weekend and publish the results by Monday. I hope that by then every Athlon-fan will be satisfied.

Please don’t forget to read the full article ‘New Hope For Athlon — The VIA Apollo KX133 Chipset‘ in case you haven’t done that yet.

Follow-up by reading the article ‘Irongate with Super Bypass vs. VIA Apollo KX133 ‘.