AlexZ wrote on 2022-04-18, 09:55:To those searching for best drivers for FX5600 this may be of interest. I tested a few driver versions on my PIII 900, 440BX, Wi […]
Show full quote
To those searching for best drivers for FX5600 this may be of interest. I tested a few driver versions on my PIII 900, 440BX, Windows 98. The fastest driver is 53.04, however I cannot comment on game compatibility vs the well known 45.23. I tested drivers using 3D Mark 99, 3D Mark 2000, 3D Mark 2001. 53.04 may be the last usable driver for retro gaming for FX cards.
3D Mark 2000 is a DirectX 7 benchmark, 3D Mark 2001 is DirectX 8 benchmark however DirectX 8.1 was installed.
Benchmarks were run in 1024x768x32bit, 32bit textures, 24bit z-buffer and triple buffering, vertical sync always off in driver settings.
3D Mark 99
- 45.23 - 6190
- 53.04 - 6176
- 56.64 - 5064
- 61.76 - 4960
- 81.98 - 6416
45.23 and 53.04 are neck and neck here within margin of error.
3D Mark 2000
- 45.23 - 5805
- 53.04 - 6151
- 56.64 - 6200
- 61.76 - BSOD
- 81.98 - 5926
53.04 is clearly faster.
3D Mark 2001
- 45.23 - 5512
- 53.04 - 5562
- 56.64 - 5511
- 61.76 - 5224
- 81.98 - 4981
45.23 and 53.04 are neck and neck here.
53.04 is clearly faster in 3D Mark 2000 while performing about the same in 3D Mark 99 and 3D Mark 2001.
56.64 - missing OpenGL settings, 50FPS cap in 3D Mark 99 (vertical sync was off) - explains low score
61.76 - same as above + BSOD in 3D Mark 2000
81.98 - too new, crashes in Fifa games using Direct3D older than 8.1
For FX5600 in Windows 98 don't waste your time with drivers newer than 53.04.
Good testing, only sorry you missed many interesting driver versions. And at least two very important ones - ForceWare 56.55/56.56 for Win9x (The first versions, by the way, which already support extension limit), as well as ForceWare 60.86. In addition, 3DMark, in my opinion, is not very suitable for such thin tests, since certain optimizations were made for them and the best solution when testing drivers is to use real applications (games). So, for example, on any video card of the GeForce FX series that I tested (NV30, NV31, NV38) in Direct3D mode (DirectX 7.0), the best results were with 56.56/56.55 driver versions. And with newer video cards (NV40+), on which, unfortunately, 5X.XX driver versions cannot be installed, in the same Direct3D, in real tests (games), the 60.86 version driver showed itself best. Regarding the OpenGL mode, the picture is different - the newer the driver, the better (And the higher the version of supported OpenGL itself, up to version 2.0 in 77.72 and 2.0.1 in 8X.XX). I managed to get the fastest results in OpenGL with ForceWare 8X.XX drivers as well.
As for ForceWare 77.72 in Direct3D, it showed itself approximately at the level of 45.23 (Same in terms of buffer shadow support in SplinterCell, by the way), but significantly worse than the fastest 56.56 in old Direct3D. In general, a good increase began with version 52.16, reaching its ceiling in version 56.56. 56.64 is already a little slower and further there was a serious regression.
When testing, I think you should not focus only on official versions, since there are not so many of them released for Win9x (The earliest one that I managed to install on Nv30 (adding an identifier to INF) is Detonator 40.52), it is worth avoiding only obviously problematic or specific versions that not suitable for GeForce FX (Like 61.42 and some other 60s).