Well, VLB certainly isn't going to be faster than PCI if that's what you were thinking. The results with those cards may also be cpu limited.

But results are strange, s3-vision864 is faster than S3-805, and both faster than a trident 9440. But tests say that this three cards have same speed.

And I know that 3DBench mission is to measure the performance in pure MS-DOS, without any hardware acceleration.

According 3DBench, any graphics card is just as good for my setup (always the same FPS)

Under Windows 3.1 (wintune 2.0) do not have the same results, this makes me think that 3dbench not provide a valid result of the performance of graphics cards.

I also acknowledge that the results under Windows 3.1 depend heavily on the quality of the drivers.

I don't want to say that a vga that works great under Win3x could work great for DOS games...

There are 2 versions AFAIK of 3D Bench.

The first one tops out at 100 fps and the second one goes a lot higher.

1.0 is the first ne 1.0c is the second one.

What I believe is happening is that the software isn't very precise and the gaps from one notch to the other one are quite large.

What scores do you get if you test those cards with an AMD X5-133 or -160?