Radeon 7500=Shader Model 1.4.
Geforce 256=fixed pipeline only.
IMHO - Doom 3 is using different renderers for each card.

Gamecollector wrote:

Radeon 7500=Shader Model 1.4.

False. 8500-9250 have 1.4. The 7xxx series don't (well, they do have some unsupported prototypical pixel shaders that Microsoft dropped the ball on for DX8.0 ....)

However the big difference is really that Doom3 doesn't have a lower non-nVidia rendering path (a lot to do with stencil buffers, dot3 bump and cubemaps than fragment shaders) Radeon 8xxx is also screwed. It gets playable and looking right starting at 9500 though, R300 is a massive improvement on the OpenGL side

It's not just failing performance though. Cubemap exponents are cut in half letting you see screwed up shading showing at the halves of models turning everyone into bolians or twoface or something.

oh yeah, i forgot to mention that the flame at the muzzle looks square! 🤣

i think you are right that doom3 looks horribly with r200 when i switched an r9200se on, the world is almost in complete darkness. its not simply "ugly" or low gamma, just as if it is missing most, if not all of the world polygons.

and i checked the opengl info of the cards and drivers with everest, it reports that even gf256 has full support of opengl1.5(maybe software support provided by driver), while r100 and r200 has only full 1.3, and incomplete 1.4 and 1.5. is that correct and does that matter much?

You can massively speed 7500 up with Doom3 by disabling vertex buffers. 7500 will use the game's generic ARB render path. It isn't great but it is playable I guess.
'r_usevertexbuffers 0'

GF256 will run a optimized NV1x rendering path. GF4MX was far too popular to ignore. There is also a NV20 path for GF3/4, I think.

8500/9000/9200/9250 should work fine. Doom3 does have a optimized R200 rendering path. Ive tried it in the past. Performance and visuals are similar to using a GeForce 3. If it isn't working you probably need a different driver.

swaaye wrote:

GF256 will run a special NV1x rendering path. GF4MX was far too popular to ignore.

I've been saying this for a long time. Between the original GeForce, GeForce 2 and MX and GeForce 4MX, that's a whole lot of video cards that were sold, especially to OEM's looking for cheap 3D capability to add to their pre-built systems. There was no way you could have released a game back then without supporting them because you would cut out too many potential customers. I think they finally managed to do away with NV1x/DX7 renderer support after Half Life 2 and Star Wars Battlefront II, which would be around 2005-06, which was way too long to keep supporting it.

i necro this up... i changed to a r8500 running at 275/275 and doom3 has switched to the r200 rendering path, driver is the known 6.11, but the performance is still disappointing:

r8500: 26fps at low quality, 19fps medium
ti4200: 70fps low, 30fps medium
gf256ddr: 24fps low, medium not run due to video ram size limit.
r7200: 23fps low, medium not run due to video ram size limit.
r9200se: 22fps low, medium not run due to video ram size limit.

as you can see, the r8500 is not only a huge lot slower than ti4200 especially at low image quality, but also only marginally faster than the r7200 and r9200se, which seems really unreasonable. have i done anything wrong? and are there any ways to improve it?

test machine is asrock pt880 board and e5200 oc 3.33g, this platform pushes the r8500 to 12100pts in 3dmark01 so it shall not be the problem.

Now I wonder what the pre-R300 chips could be used for though. Do they have any potential use?

Truform. 😀

I think they're also the final ATI chip with error diffusion dithering...

Don't judge these cards by Doom3, bad things happened between Id and Ati.

I would also try an older driver. Maybe a version from a few months after the game's release. ATI OpenGL doesn't necessarily get better with time. Application fixes seem to be lost sometimes or new bugs break old games.

swaaye wrote:

I would also try an older driver. Maybe a version from a few months after the game's release. ATI OpenGL doesn't necessarily get better with time. Application fixes seem to be lost sometimes or new bugs break old games.

i dug up some driver benchmarking articles, testing various versions of ati drivers in 2004-2005(before and after doom3's release). they all showed same scores on the 8500 without any improvement, and were as slow as mine. but the performance of the r300 series improved quite a bit in those months, and even the 9550 was a lot faster than 8500.
so i assume its either that ati never attempted to optimize the r200 for doom3, or that the r200 gpu had totally incompatible designs with doom3 that couldn't be fixed by drivers.

noshutdown wrote:

i dug up some driver benchmarking articles, testing various versions of ati drivers in 2004-2005(before and after doom3's release). they all showed same scores on the 8500 without any improvement, and were as slow as mine. but the performance of the r300 series improved quite a bit in those months, and even the 9550 was a lot faster than 8500.
so i assume its either that ati never attempted to optimize the r200 for doom3, or that the r200 gpu had totally incompatible designs with doom3 that couldn't be fixed by drivers.

I know that id was working on optimizing for R200. There are .plan posts from Carmack talking about R200 performance and of course there is the R200 render path. But yup I wouldn't be surprised if ATI stopped putting much effort into R200 once R300 was the focus of the company.

Is there a particular fast atioglxx.dll to share for the best R200 experience with that game? 😀

swaaye wrote:

I know that id was working on optimizing for R200. There are .plan posts from Carmack talking about R200 performance and of course there is the R200 render path. But yup I wouldn't be surprised if ATI stopped putting much effort into R200 once R300 was the focus of the company.

i have seen something like that, my thoughts are that carmack had planned optimizing for r200 in early develop stages, but then gave it up as r200 later turned out not to quite meet his request. there is a r200 rendering path in the game, but i doubt it actually did any good. just compare it with r7200 and r9200se results, especially that r9200se is less than half of r8500's speed in most tests.

http://techreport.com/review/7200/doom-3-mid- … e-gfx-comparo/3

8500 is beating 9000 here. RV2x0 chips should be using the R200 path too. The results make R200 and RV250 look pretty bad with a plain FX 5200 beating them. Frankly I think the ATI D3D8 chips were buggy and inefficient and there is a lot of proof in benchmarks to back it up. I don't think one can blame the drivers entirely. The difficulties with driver development could actually be indicative of hardware flaws.

That a 5200 is competitive is interesting. NV34 operates at 2 pixels per clock when shader programs are involved AFAIK (4 px/clk in simpler games). 5200 non-Ultra is only 250 MHz. This means the pixel & texture fillrate should be much lower than that of a 275/275 MHz R8500. They have similar pixel shader resources, but 5200 has considerably less memory bandwidth (275 vs 200 MHz @ 128-bit). So 8500 should have a number of advantages, but it is still losing. Unimpressive efficiency.

On the other hand for a display of massive inefficiency, look to the broken Parhelia...

swaaye wrote:

http://techreport.com/review/7200/doom-3-mid- … e-gfx-comparo/3 […]
Show full quote

http://techreport.com/review/7200/doom-3-mid- … e-gfx-comparo/3

8500 is beating 9000 here. RV2x0 chips should be using the R200 path too. The results make R200 and RV250 look pretty bad with a plain FX 5200 beating them. Frankly I think the ATI D3D8 chips were buggy and inefficient and there is a lot of proof in benchmarks to back it up. I don't think one can blame the drivers entirely. The difficulties with driver development could actually be indicative of hardware flaws.

That a 5200 is competitive is interesting. NV34 operates at 2 pixels per clock when shader programs are involved AFAIK (4 px/clk in simpler games). 5200 non-Ultra is only 250 MHz. This means the pixel & texture fillrate should be much lower than that of a 275/275 MHz R8500. They have similar pixel shader resources, but 5200 has considerably less memory bandwidth (275 vs 200 MHz @ 128-bit). So 8500 should have a number of advantages, but it is still losing. Unimpressive efficiency.

On the other hand for a display of massive inefficiency, look to the broken Parhelia...

yeah whatever, the results are accordant with mine, however there are still some unreasonable problems that i can't understand.
1. in medium quality 8500 is much slower than ti4200 but somewhat acceptable. however when it comes down to low quality, the ti4200's speed skyrockets while the 8500 doesn't.
2. 8500 has similar structure with the 9200se(lowest model of the r200/rv250 family), and in most tests 8500 is over twice the performance of 9200se, however in doom3 its only marginally faster.

Less efficient does not mean broken, there is 1,5 year distance between those two. The FX architecture was just perfect match for Doom3, in other games 8500 is good match for 5200 Ultra.

noshutdown wrote:

...in most tests 8500 is over twice the performance of 9200se, however in doom3 its only marginally faster

So memory bandwidth of R2x0 is somehow irrelevant for the game.

Putas wrote:

Less efficient does not mean broken, there is 1,5 year distance between those two. The FX architecture was just perfect match for Doom3, in other games 8500 is good match for 5200 Ultra.

noshutdown wrote:

...in most tests 8500 is over twice the performance of 9200se, however in doom3 its only marginally faster

So memory bandwidth of R2x0 is somehow irrelevant for the game.

memory bandwidth is not the only difference, 8500 has 2 vertex shader and 4*2 texel pipeline, while 9200se has only 1 vertex shader and 4*1 texel pipeline. the 8500 also runs at much higher clock than the 9200se aswell.