How to boost your Doom3 performance by 40% on ATI hardware

IMHO a very simple way to do the GL_LINEAR / GL_LINEAR_MIPMAP_LINEAR stuff to workaround the app AF bug is to put the following in your autoexec.cfg:

Code:

bind "g" "set image_filter GL_LINEAR; wait 10; set image_filter GL_LINEAR_MIPMAP_LINEAR"

Of course, you may choose the key freely . Then, press the key ONCE in-game to activate.

The wait command is important, it won't work without it ("wait x" tells the game to wait x game ticks - frames AFAIK - before processing the next command). The value 10 is simply the second one I tried and it worked, so no science here (my first try was just "wait" which equals "wait 1", but this didn't work)

BTW putting things like...

Code:

set image_filter "GL_LINEAR"
set image_filter "GL_LINEAR_MIPMAP_LINEAR"

...in your autoexec.cfg won't work. Just mentioning this because a few people suggested that. 8)

 

Hey!, my system is really angry, 31,5 fps is very playable, and he strains to the maximum. Compare above P4 3,2 + 9700Pro give 41 fps. My CPU run only at 2,0 Ghz, so it´s not bad.

I will always defend to my machine.

 

9700 Pro is slower than your 9800 Pro yet the machine with the 3.2 Ghz CPU and 9700 Pro is 30% faster than what you're getting... sounds like a pretty good argument that you are largely CPU limited.

Most of the 2 Ghz CPUs are slow... and your 133 Mhz RAM is not going to help.

-FUDie

 

If you're happy with how D3 runs that's great, but it was you who said you needed a boost for D3, so I assumed you weren't. Also I never said you were CPU limited, I said try timedemo in 640*480 low detail to see if you were CPU limited.

I've got X800 Pro, P4 2.4 and I get exactly the same fps in 640*480 low detail and 1280*1024 high detail - therefore my CPU is the problem.

 

Wait until september?

 

I don't get it? That's funny.

Guess what? I ran a test on the refrast and got the same results as on the R300, minus the angle-dependent AF of course.

Makes no sense? Guess what: You don't write the specs. Last time I checked, you didn't work on 3D HW or SW and are really in no position to judge "what makes sense".

Obviously, ATI is following the refrast. NVIDIA does not follow refrast as setting a mag filter of NEAREST and a min filter of ANISOTROPIC does not give the desired results. It doesn't matter what you think is correct: This is not the world of Chalnoth, this is the world of 3D graphics.

Aliasing is not necessarily an error and can even be a desired effect. Why else would apps or users set a negative (i.e. sharper) LOD bias? Why else would games run without mipmaps?

Of course, I hardly expect you to admit you are wrong, you never do.

-FUDie

 

Humus! question! is this the total fix by downloading it(by that i mean both codes) or just the first code?

 

Xoc3, I think your fps for Doom is really good, considering the bottleneck your RAM and CPU would(should) make. Just shows, how great that 9800P of yours is. My RAM runs effectively %600 faster than yours(not to mention my CL2.2.2.6.8 timings), and my CPU has a 200mhz clock/fsb to match it. My R9700P is in good balance with the rest my system, where your better R9800P is not, I think. If it wasn't for Humus' tweak and CP 16xAF, the rest of my rig is matched by yours(in Doom 3 performance), which I think is pretty damn good, considering the rest of your HW, imo

I agree with FUDie, that the rest of your rig slows down your overall performance, and I'll bet a (rare) Fibble cake, that the test he suggested, would show that your system is indeed CPU limited.

Again, nice performance I think

Humus, now it all makes sense. I tried to search for the POW lines you mentioned in your initial post, but couldn't find them in the shader. Could you post all the final tweaks/alterations you have made? Just being curious and lazy, I guees(don't want to scan the +50 pages of posts)

Again, thank you for shader tweak. Also, is the improved performance only for those who force CP AF, or does it work overall. I haven't run any tests without CP AF, yet. Many comments on the www about your tweak, has stated that they can only see a small fps gain, or none at all with ATI HW. Would forcing AF from the CP, be enough for them, to actually get the great performance gain, that I (and others) have received?

 

I thought WaltC had conclusively proved it couldn't possibly be a driver bug?

Humus,

Can you confirm this? This would imply that if the fix had been released earlier, prior to your shader patch, the ATI cards would have performed considerably worse in D3 than the currently published benchmarks suggest.

Of course, it is possible that if the fix had been released early enough id might have noticed the suprisingly poor performance and taken the time to investigate it beforehand and possibly provide a different code path, but that's hard to say.

 

wheres the link to download the new version? the sticky?

 

First page... If I haven't misunderstood the question.

 

I'm not knowledgeable in the coding arena, but I was wondering if the problem with adding the:

set image_filter "GL_LINEAR" is due to the .cfg needing "seta" vice "set" for the command?

I know that "set" is used in the Console for things like "set com_allowConsole 1", but since this looks like the line that is in the Doomconfig.cfg shouldn't it be "seta" (for the autoexec.cfg)? Just wondering, don't flame if I'm wrong. Didn't know if you needed to change from "set" to "seta" when doing it in a .cfg vice console. Just trying to learn. (Sorry for going off topic of Humus Tweak.)

BTW great work Humus!!!

 

I have just updated to Catalyst 4.8 and it gave me a good boost in performance.

1024x768, High settings, everything advanced on except AA and V-Sync. CP settings: AA=AP, AFx16 Quality, Tex/MIP=HQ. Timedemo demo1 runs.

Catalyst 4.7:
41.1 fps
Catalyst 4.8:
43.4 fps

1024x768, "Ultra" settings, forced machinespecs "3" and Vram set to 512mb, execmachinespec. Else, same settings as above.

Catalust 4.7:
32.2 fps
Catalyst 4.8:
34.8 fps

Nice boost from new drivers

High settings runs and looks really good now. %35-40 better performance, compared to Catalyst 4.7 without the Humus tweak. Humus should work for ATI or ID Software.

Ultra setting is now faster, than my original(post Humus tweak era) high settings(but the constant overflow of memory makes the pauses annoying, most of level1 and level2, I get 40-60 fps in Ultra, that is then dastardly dropped to 0-8fps when going to another area, or when a sequence is activated). Nice little R9700P *looks at it lovingly*

 

Good idea.

 

There are several rendering paths. How do we get to set specific path to use?

"r_renderer [ arb | nv10 | nv20 | r200 | arb2 ]"



What does this do and would it benefit me in any way to set it to a specific path for my 9800pro(128MB)

 

The only difference between "set" and "seta" is that "seta" writes variables (name AND value) to DoomConfig.cfg in addition to setting them, while "set"-ting a variable changes values in DoomConfig.cfg only if the variable name was already defined in there. In other words, if you want your changes reflected in DoomConfig.cfg no matter what, use "seta". Otherwise, use "set". I like "set" because its a bit shorter and doesn't change DoomConfig.cfg when I tweak "exotic" variables

Ok but back to topic, forgot to write in my last post that I really like Humus' tweak! Although I only gain about 2-3 fps from it alone (using app AF and the workaround I mentioned), together with some other tweaks I have done (.cfg / driver) I am getting about ~7 fps "for free" which is quite nice I think 8)

 

Type "gfxinfo" in the console to find out which one is in use.
Type "r_renderer best" for the game to choose (it should choose arb2). Your card can use arb|r200|arb2 (quality in that order).