MikeC suggested that I post the results of an experiment I ran with some old NVIDIA cards through the Doom3 benchmark.
The command was "timedemo demo1 nocache" run on the same system to ensure all the tests would be consistent. I want to minimize the relevance of the system, since this is a comparison of the cards to each other, not an absolute performance test. YMMV.
DOOM3 Timedemo FPS (Frames Per Second) at 640×480
- The 5900U was a NV35GL engineering sample with a 5900 BIOS and slow RAMs clocked at 300/500 (250 DDR). The RAM is clocked well below consumer cards and performance is NOT AT ALL representative of consumer 5900 Ultras
. The rest of the cards were consumer cards or engineering samples clocked at the same speeds as a consumer card of the same kind.
The system had a fresh install of Windows XP no service packs, and Doom3 with the 1.1 patch. Hardware baseline: nForce2 A7N8X-E Deluxe / Barton 3000+ 333 MHz fsb / 512M dual channel PC3500 @ 333 MHz / 2x WS Raptors RAID0 on motherboard controller.
So it seems you better have a 128M GeForce4 Ti or higher (excluding the FX 5200) to enjoy Doom3 at the minimum resolution of 640x480.
I didn't see significant image quality changes from 256M, 128M, and 64M cards. Video RAM down to 64M was acceptable, but there was a horrible decrease in image quality dropping below 64M RAM. With 32M RAM, it was really bad:
- coloring on the walls was banded
- shotgun smoke was square
- fog effects had linear edges
- lighting was cartoonish, posterised if you know Photoshop jargon.
What surprised me:
- It appears that 32M cards provide better FPS than 64M cards equipped with better GPUs, at the cost of image quality. The GeForce 256 and GF2 GTS @32M had better FPS than the GF2 MX or GF4 MX @64M.
- The FX 5200 performed dismally in raw speed, and didn't seem to provide any better image quality. I expected a low-end NV3x card to be better than a high-end NV2x card because of DX9 compatability. I'd rather have a faster DX8 card than a slower DX9 card.
- I also used to think 20 FPS was acceptable, but after the testing session, I'd rather play at a lower screen resolution than play Doom3 below 30 FPS.
- A 4-pipeline NV36 (FX 5700 U) with DDR-II was supposed to perform on par with a 4-pipeline NV30 with DDR-II (FX 5800U) at the same clocks, but the 5800U performed consistently better, about 10 FPS better than the 5700U at each setting. (Unfortunately I gave away my GF3 and GF 5700 before this organized testing, so they don't appear.)
The zip attachment is the spreadsheet. There are more measurements, but I got fed up with Excel before I could make it all pretty.
EDITED for spelling, detail, and clarity.