Question about 3DMark 2001 SE Vertex + Pixel Shading.

[micb]

After looking at Dave's reviews of graphcs cards, I noticed a difference between my 3D Mark2001 theoretical scores for vertex shader and pixel shader numbers on my ATI 9800.

I suddenly realised there might be a CPU/FSB component to the scores/tests as mine were much lower.

To test this I run the following test with my system at 133 FSB = 1667 MHz (2000XP) and then again at 100 FSB = 1250 MHz.

133 FSB = 1667 MHz :

100 FSB = 1250 MHz :

By dropping the FSB by and also CPU speed the scores drop by ~20% in the first two tests.

My main question is why are these first two tests so CPU dependent as I would have thourght they should be only graphics card dependent?

If you look at the Advanced Pixel shader test they stay exactly the same. Is this test correctly graphics card limted?

Note: system specs are:

AMD Athlon XP 2000+ (Palomino)
Epox EP-8KTA3 PRO Motherboard
512 Megabytes RAM (PC133)
Sapphire ATI Radeon 9800 128MB (AGP 4x)
17 inch CTX Monitor
Microsoft Windows 2000 Professional (Service Pack 4)
ATI Catalyst 3.9 Drivers.

 

[horvendile]

This isn't my area of expertise, but since I don't really have any I'll hazard a guess!

I suppose that you still have to send vertex data to the graphics card. It could be that your rather slow memory is the bottleneck. If it's running synchronously it drops to 800 Mb/s (theoretically) with a 100 MHz FSB. This implies that there is less vertex data in the third test than in the former two, but just looking at them (AFAICR) that seems plausible.

 

[Ostsol]

Vertex data, just like textures, can be stored in video memory.