View Single Post
Old 11-07-04, 09:07 AM   #1
msxyz
Hybrid Microprocessor
 
msxyz's Avatar
 
Join Date: Jul 2004
Location: Milan (Italy)
Posts: 2,050
Default Benchmarking Doom3 Renderpaths...

Ok, this week I had some spare time on my side and I finally performed some experiments I had in mind for a while...

Since Doom3 supports multiple render paths, involving the use of different pixel shader instruction sets, I made some comparison between them. I was particulary interested in ARB2 vs NV20 path performance on NV3x hardware. I happened to have one spare NV34 (a XFX GeForce5500) so I could also benchmark my NV36 card against the "older generation" NV34 core. The results are interesting.

Test metodology: Detonator 65.73, Quality settings, no trilinear optimizations. Doom3: high quality, 1024x768, Anisotropic filtering disabled. I used a custom timedemo recorded from the intro scene (1785 frames). Tests repeated 5 times, I dropped the highest and lowest value and made the average from the remaining three...

I've also tested NV10 and ARB paths but I don't feel they're interesting to my research. Instead, I concentrated on the ARB2 vs NV20 path comparison because they are both usable on the NV3x hardware and they look quite similar.

Now, to the numbers...

GeForceFX5500 (270MHz core / 200 MHz DDR)

ARB2 path: 12.0 fps ; NV20 path: 14.0 Difference +16.6%
Shadows/Advanced SFX disabled:
ARB2 path: 18.7 fps ; NV20 path: 23.5 Difference +25.6%

My comment: On the NV34, the NV20 path is much faster than the ARB2 path, despite the fact that it requires multiple passes to render the frame. I was expecting such results, but the gap really impressed me. If any of you has a low-end 5200/5600 on your PC and wants to play Doom3, the NV20 path may be a better solution. The NV20 path does not support the "heat haze" effect and it has a slightly worse image quality but it may be a better compromise than reducing texture quality or disabling some features. PS: As you can see, a GeForceFX5500 is not the ideal card to play Doom 3 at 1024x768!

GeForceFX5700U (475MHz core / 450 MHz DDR)

ARB2 path: 25.2 fps ; NV20 path: 24.3 Difference -3.5%
Shadows/Advanced SFX disabled:
ARB2 path: 32.4 fps ; NV20 path: 29.9 Difference -7.7%

My Comment: on the NV36, it seems that switching from the ARB2 to the NV20 path will cause a little performance drop. NVidia did redesign the NV35/6 core to boost its performance with PS2.x shaders. Using the NV20 render path on a GeForce 5700/5900 affects both performance and quality in a negative way.

Well, these are my findings. I do not claim them to be 100% accurate. Both cards were probably held back by my slow CPU. I wish I had a FX5600 and a FX5800 to throw into the test. Maybe some of you can try and see if the results are similar
__________________
Powered by:
Pentium Dual E2180 (2GHz, 200MHz FSB)
Abit L95C (The one and only mATX with dual PCIe slots!) + 2x1 Geil Black Dragon PC6400
XFX 9600GSO 384MB
Little freak wonder

Last edited by msxyz; 11-25-04 at 02:06 AM.
msxyz is offline   Reply With Quote