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

My results can be found at:

http://s95069928.onlinehome.us/stuff/doom3/

An abridged version of my conclusions: thank you, Humus, for a wonderful tweak.

 

Holy shit! I seem to be unable to make a single post in this tread without making an error! I guess that would pretty much turn the entire screen white (or rather, turn every specular object that is not in total shadow super bright). Thanks for catching that.

 

How to instal Humus's Doom 3 tweak?

Ok, I've been wonder what should I do with his file since the .vpf is full with instructions and things pretty confusing. I thought that I should just extract it to the Doom 3 base folder, but I'm not sure if I will be doing something bad with the game. That's the right way to do it? If it isn't, please, teach me how to...

 

Found this over at HardOCP, not sure if its been posted here yet:

http://www.hardforum.com/showthread.php?t=791383
(click for AVI picture animations)

 

Eek, you're cursed! *runs away*

 

I doubt any reviewers would use a hack like this to do a real performance benchmark, so it's kind of a non-issue. Humus never said it was a perfect replacement. It's an approximation, but a very good one. I'm interested in seeing what other things people can come up with.

 

So now a few questions come to mind -
- Did the guy do the comparisons correctly, and with the correct version of the new shader code?
- Is the 'hack' better quality or worse than the original?

Seeing as the change to the shader, as far as I am aware, is supposed to be the same mathematically as the texture lookup that it replaces (John Carmack seems to have implied this in the interview on B3D) then the question is which is more accurate - a two stage ALU calculation using 24-bit floating point, or a texture lookup into a relatively low bit-depth texture map...

 

Since Carmack designed and tweaked the look of the engine with the texture lookup method, I'd say the texture lookup method is the ideally the more accurate presentation of the game, especially based on the critical role brightness and light plays in Doom3 specifically. But I wouldn't lose too much sleep over the differences.

 

Cool stuff

But can somebody post a complete, stable, nice looking VDF file here?
Would be nice, i don't know much about this kinda modding of files like that, so it would be much more helpful for me ^^

BTW: i copied a already modified file into my pk4 file and somehow i get some snowy artifacts when complex light sources appear, like the one on the very beginning which casts some awesome shadows.... this bright, circling light. There are appearing some strange curruptions at the door nearby when the light comes.... didn't happen with the normal/standard VDF file here, and my card clocks aren't that high at the moment (summer oc) ....

 

Differences generated by slightly different texturing implementations between hardware (for example different filter precisions or levels of sub-pixel accuracy) could easily be of the same sort of magnitude or larger than those between using ALU operations and a texture lookup, so I wouldn't have thought that you could really 'tune' a function to a texture lookup with the sort of precision you're implying other than on one given piece of hardware.

 

Specs and you're using medium, high low, or Ultra quality?

 

This is Nvfanboy's talk, being scared that they lose the crown in the only game they own. I gain an average of 20% ingame performance with my X800XT, and looking at the latest screenshots, and comparisons by myself, this is not something that we can "simply forget".
I have no doubt that if JC and Ati will sit around the table they'll find something that will give a 100% equal picture, with the performance gain Humus gave us.

If this thing is not gonna be official one way or the other, personally I never take a future benchmark serious any more if there is no "help for Ati" involved.

 

HELP HELP HELL HELPã!

How to instal Humus's Doom 3 tweak?

Ok, I've been wonder what should I do with his file since the .vpf is full with instructions and things pretty confusing. I thought that I should just extract it to the Doom 3 base folder, but I'm not sure if I will be doing something bad with the game. That's the right way to do it? If it isn't, please, teach me how to...

 

Re: HELP HELP HELL HELPã!

There's a big FUCKING sticky in this very forum. Now open your eyes and stop spamming.

 

I get 0.5 fps hit with Humus' patch on my Radeon 9600Pro on medium quality. I can't read the whole topic (because it's lagging a lot), so what is wrong with my machine?

 

That's what you might say, but it's not what Carmack says...

From the B3d "interview" with Carmack:

(emphasis mine)

If Carmack says no strong argument can be brought against it, and Carmack himself brings no argument against it at all when directly asked about it, it's a bit futile for you to be putting words into his mouth, wouldn't you say?...

But he does more than refuse to argue with it, he actually exlains why he didn't do it himself:

(emphasis mine)

In other words, Carmack says that under different conditions it is the very thing he might well have done himself. He explains why he didn't do it himself--as his concern at the time was *older hardware* support (hardware incapable of anything except texture lookups) coupled with an expectation that his D3 development time was much more constrained than it actually turned out to be.

So I think here Carmack allows us to dispense with the speculations that "lookup tables are better," since he states he did it for older hardware compatibility and does not assert he did it for reasons of IQ; and that "it's a hack," since Carmack himself cites it as entirely legitimate and something he himself might well have done if his development conditions had been a bit different (if he'd actually have known how much time he was going to have with the engine.)

So, I think I'll stick with Carmack on this one...

 

I'll bet dollars to donuts that some reviews will come out showing the X800 vs 6800 and they'll run the X800 on the default and on the "Humus tweaked" version of D3 just to show what kind of difference the tweak can make.

If they show default as well as tweaked, I ain't got a problem with it....but just using the tweaked version would be wrong.

 

LOL, why is it that so many people can write but not read? :?

JC has stated that the function realized via the lookup is not a plain power function and it it would take several shader instructions to calculate it exactly. That's the beauty of lookup tables - you can stuff arbitrarily complex functions in there and still get the same constant-time response. JC has also stated why they continued using this function instead of re-tweaking everything in order to accommodate a simpler calculated function that might give a marginal improvement on ATI's newest cards.

The problem that Humus' tweak addresses is a driver issue; to wit, that the lookup is slowed down if filtering is forced on via the control panel. The workaround devised by Humus avoids this performance hit at the cost of using a slightly different function, which results in occasional small IQ degradations (but which you'd probably have to search out specifically to notice). That is why you can see such a huge performance gain on some cards, and this has been explained several times in this thread.

And the reason why people force AF through the control panel is another ATI driver issue - namely that the driver will not filter correctly if the AF is requested by the application and this causes image degradation that is noticable and jarring (at least if you're as spoiled as I am ). This has also been already explained in this thread and demonstrated by way of screenies.

I find these driver problems especially odd as image quality is the only advantage of the new generation of cards over the old and the only advantage of the X800 over the GF6800. Just look at the benchmarks - on the whole it is only at high-resolutions in eye-candy mode that the X800XTPE leaves the GF6800U completely in the dust, and without AA/AF the previous generation of cards (especially the 9800XT) are more than sufficient for the majority of games. If you take away the quality then you take a way the superior advantage of the X800XTPE.

Be that as it may, it would be nice if some version of this fix made it into a D3 point release or something, because it would be perfect for deathmatch where a dependably high frame rate is more important than perfect visuals (as long as there is no shimmering or aliasing, of course, so AA and AF are still a must).

My oh my, I think you've uncovered a real problem here. People like JC have been using texture lookups for years and now we find out that it cannot work! Send an email to id, quick!

 

So I'm guessing this only works for dx9 supporting cards, and geforce 3/4 and 8500 users are left out?

 

Since they won't be running via the ARB2 path, Humus' tweak won't affect them....I think. Too tired to check!