by Bubba "Masterfung" Wolford

Q) Some have compared S3TL potential to what they believe T-Buffer might look like. Having been one of the few who have actually seen T-Buffer running in San Jose a few weeks ago, I find a T&L comparison to full screen, real-time Spacial AA kind of ridiculous. Can you explain some of the differences between what 3dfx is doing with T-Buffer and what S3 is doing with S3TL?

A: Well, comparing a geometry solution like S3TL to a rendering solution like the T-Buffer is really an "apples to oranges" comparison. Geometry solutions like that offered by Nvidia and S3 with their recent product announcement address the issue of being able to add more geometric complexity to games and applications.

The T-Buffer rendering solution, however, is a technology which is used to improve the rendering image quality of a game, such as removing the "jaggies" on the edges of triangles, adding real-time motion blur to a scene, emulating depth of field focal effects, dramatically improving shadow and reflection quality, etc.

So, the two are really quite different and distinct. S3 and Nvidia have opted to focus on the geometry aspect of 3D graphics, while we have instead opted to focus on improving the overall rendering visual quality. It's really just different viewpoints of what are the real "problems" with 3D graphics today. As we've said, we believe rendering image quality is the most glaring visual "problem" with today's 3D graphics, and that's why we are so excited about the T-Buffer technology.

This is going to be a problem for 3dfx because people are missing "perspective" on what T-Buffer really looks like in a game we can relate to. Having seen it running in a game I am familiar with (NFS 3), I was trying to pick my jaw up off the ground while watching the difference! Ever had a good friend tell you how simply awesome a movie is and you are skeptical because you have not seen it yet? 3dfx is trying to do that job, convincing you that spatial AA is something you cannot live without before you have actually seen it...

We don't want to quite give away our own take on this, but we want you to know that COMBATSIM.COM™ has a little something special coming in this department.

Q) Can you explain why you feel fill-rates (as opposed to the direction NVidia has taken with high poly/sec ) are so important in games?

A: Sure, the example I like to use is Quake3. Let's take a look at Quake3 frame rates running on a TNT2 board (Pentium3-600 Quake3 v1.08, q3testdemo1, vsync off):

640x480

16bpp: 84.5
32bpp: 56.4

1024x768 and 1600x1200

16bpp : 49.7 and 19
32bpp : 29.8 and 11

Notice that at low resolution (640x480) and 16-bit color, the frame rate is fairly high, which here could actually indicate a cpu-limited situation. When a game is cpu-limited, then a geometry accelerator like the GeForce can actually improve overall frame rates. However, remember that increasing resolution or increasing pixel depth has no impact whatsoever on the cpu or geometry horsepower requirements.

Click to continue

So, as you can see by the benchmark numbers below left, either increasing the resolution or increasing the pixel depth to 32-bit dramatically reduces the overall performance. What you are seeing is fill-rate becoming the bottleneck. If fill-rate were not the bottleneck then the 1024x768, 16-bit score would also be 84.5 (just like the 640x480 benchmark score…). So, unless fill-rate is addressed you can have a Cray Supercomputer doing geometry calculations and it's not going to make Quake3 run any faster at high resolutions or high color depths!

We have heard from gamers loud and clear that they want to run their games at high resolution in 32-bit color, so we have focused on delivering a solution that meets those needs. If someone wants to increase their geometric complexity and run at 640x480 or 16-bit color, then maybe the GeForce or Savage 2000 products are right for them. But we know that no one wants to take a step backward in resolution or color depth, and that's why we feel so confident that gamers will embrace our next generation products.

This has been 3dfx's policy for sometime now as Scott Sellers has pointed out. This question is actually pretty simple. No pushing the fillrate, no playing the game at high resolution in 32-bit color with high FPS. Without a good fillrate, an awesome product is hopelessly, "bottlenecked". Period.

Q) Can you talk a little about what kinds of texture compression we are going to see from 3dfx this year?

A: We recently announced a very exciting development for texture compression known as FXT1. There are lots of web sites, including our own at 3dfx, that go into the details of FXT1, but the bottom line is that we've delivered a compression technology to the Open Source community that is not only superior to S3tc but also available to everyone-and-anyone who wants to use it.

So instead of making the technology proprietary, we have instead opted to completely open it up free-of-charge so that it will gain incredibly broad acceptance, not only on the Windows platform, but also on Linux, Mac, BeOS, etc.. We have received a ton of interest about FXT1 since the announcement, and I think you'll see some very interesting announcements moving forward regarding companies' pledged support for FXT1. It certainly already has good momentum.

That being said, because we are committed to offering the best Direct3D accelerator out there, we will not only be supporting FXT1 in all our future products but we will also support S3TC in our future products. This way we believe we give the developers and consumers the best of all worlds.

COMBATSIM.COM™ also put up an article about FXT1. This technology could be very exciting if it becomes a standard. Again, time will tell…

Go to Part V: Features and Bump Mapping