GeForce3 Under Attack: ATi's Radeon 8500 Previewed

Detonator 4 - NVIDIA Spoils It All

You have to admit that ATi is making a gutsy move to release Radeon 8500 today, considering that the final spin of silicon isn't ready and several features of the test samples are not functioning right now. ATi was obviously of the opinion that the brute force of Radeon's superior theoretical fill rate and memory bandwidth would be good enough to beat NVIDIA's GeForce3 badly enough to make possible buyers forget about the missing SmoothVision, video and HydraVision functionality.

NVIDIA, however, has been fully aware of the upcoming R200 launch and the strategists in Sunnyvale decided to drop some kind of bomb close to the release date of Radeon 8500. The new driver set that goes by the name of 'Detonator 4' will be released this week, but the press has already been equipped with it. The new driver with the revision number 2080 is able to boost GeForce3's performance in a very significant way, making it extremely hard for Radeon 8500 to beat its competitor from NVIDIA.

We had a closer look at 'Detonator 4', which has already been hyped quite a lot. We can already now tell you that we don't agree with a 'Thank you, NVIDIA'. We think that Detonator 4 has been ready for weeks now and that consumers had to wait for its release only to ensure that NVIDIA could spoil ATi's Radeon 8500 launch. I personally don't consider this as an example of admirable sportsmanship, but who except me in this business would possibly care about such old-fashioned or even honorable things?

NVIDIA's new 2080-driver has one important improvement over previous driver revisions. The software engineers under Dwight Diercks were able to squeeze an impressive amount of useable memory bandwidth out of GeForce3's admittedly excellent crossbar memory controller design. However, not all applications are benefiting from the new driver, as you can see in the chart below:

The extremely useful synthetic fill rate test that is built into 3D Mark 2001 could show a fill rate improvement of up to 29%, mainly found in high resolutions and 32-bit color. In actual fact, the fill rate at 1600x1200x32 is now as high as at 1024x768x16, which is not only remarkable, but almost suspiciously high.

Quake 3 Arena can nowadays be considered as a game with rather low expectations to hardware, which is why the fill rate increase generated by the new driver is directly translating in up to 29% higher game frame rates (when it is fill rate limited!) at 1600x1200x32. You can see that Radeon 8500 has to beat a much stronger enemy here.

AquaMark (AquaNox) and 3DMark are still benefiting quite nicely from driver 2080 as well, but the picture changes already with Dronez, which in worst-case scenario is actually performing up to 1% slower than with the old driver revision 1241.

It goes downhill from here. Evolva may lose a bit or may gain abit, depending on the resolution. PowerVR's VillageMark doesn't give a rat's butt about the driver revision whatsoever, but Giants is the real loser. 29% SLOWER is pretty much unacceptable.

All in all NVIDIA's Detonator 4 release leaves a bad taste in my mouth. It was obviously timed for a release date that would possibly hurt ATi. Fine, if this makes the guys feel better. I'd prefer a driver release date that gives the customers the best driver as soon as possible and as late as necessary to ensure reliable operation. What I don't like is that the new driver shows its biggest improvements in the typical benchmarking programs, such as 3DMark2001, Quake3 or AquaMark. Rarely used game benchmarks see only little or no gain, while Giants is even losing out considerably. I leave it up to you to make a rhyme on that. Detonator 4 is a good thing for most owners of NVIDIA based cards, but I challenge the choice of its release date as well as the fact that at least one (new!) game does better with the old driver.