Introduction

What's this?

The Viper II finally made it into our offices this past weekend. The timing couldn't have been worse with everyone preparing for Comdex. A few of us had just left for Vegas hours before the FedEx package arrived, but, luckily, I had to stop by the office to finish up an article for the weekend. That's when I found the Viper II box sitting next to my desk. Of course, I immediately put my other article on hold and started work on the Viper II.

From announcements to reality

The current Savage 2000 has evolved quite a bit from the original specifications S3 first announced back in late August. We put up a little preview of the Savage 2000 the day S3 announced the chip.

As you can see in the preview, the original Savage 2000 specs claimed a 700Mtexels/s fill rate with a 175MHz engine clock. The original specs also stated that core speeds would go up to 200MHz, suggesting an 800Mtexels/s fill rate. Well, it looks like the original specifications were a little too optimistic.

The shipping Viper II card's Savage 2000 chip has a 125MHz core clock instead of the lofty 175MHz clock called for in the original plans. That translates into a 500Mtexels/s fill rate, which is significantly less than the 700Mtexels/s S3 first announced.

Having optimistic numbers isn't new. In the past, Nvidia has been known to promise numbers in early product announcements and have the finished product come up short. Making product announcements on unfinished semiconductor products is very tricky because yields are usually unknown at such an early stage. Nvidia has learned from previous announcement mistakes, and you can bet S3 is learning right now. 3dfx has been the only 3D graphics company that has been pretty good about delivering on promised numbers.

Time to dance

Two weeks ago, we posted a few preliminary benchmarks that we obtained during the S3 Savage 2000 Party. We were only able to test the Viper II cards on the P3-600 and Athlon 600 machines provided by S3 at the party. The benchmark scores were decent and placed the Viper II somewhere between the TNT2U and the GeForce, not spectacular, but the drivers were still very rough.

Now we have a Viper card to test on our own benchmark systems. S3 had two weeks to improve on the drivers. Let's see how the card does in front of our firingsquad.