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

Can they be forced to run the ARB2 path? And if so, would it help then? Imagine an 8500 getting playable framerates in the ARB2 path!

 

I guess that's fair enough.

 

Why is it that so many people don't consider who they are talking to and what knowledge that person might actually have of the issues at hand before arbitrarily assuming that they know more and that the other person must be suffering from a reading comprehension disorder?

I am not sure who you are - since you haven't posted here before I wouldn't like to make assumptions. Perhaps you are in a good position to understand the issues at hand, but on this particular issue I would like to regard myself as at least as well informed as the average person.

And starting with an LOL or not my sense of humor managed to disappear within that one sentence of your post - something of a new record for me I think. Maybe I'm just being too touchy.

Prior to your post I knew nothing at all about the possibilities of using textures as lookup tables - I really am a bit of a newbie at this - thank you for enlightening me.

I certainly don't know anything about the performance characteristics of ATI hardware, and what runs fast and what doesn't. After all how could I possibly have detailed knowledge like that?

Sigh.

Do you actually know how complex the function being approximated is?

Would you like to hazard a guess as to whether I might possibly know how complex it is?

My oh my - you managed to miss the point completely - not a very good start for someone questioning someone else's reading comprehension.

You really should think about what you write and what issue was actually being discussed in the original message. The original question was one of precision, and what can be used as a reference, not whether we've magically discovered some new issue with texture lookups for the first time.

The question is - can a texture lookup for a function be tuned so that it can be regarded as the only appropriate reference implementation of the function, even when talking about very low-level differences in precision. The point that I was trying to make is that two differing implementations of hardware pipeline can easily produce relative errors when implementing the texture lookup that are just as large as the differences between the reference and an implementation that directly uses ALU instructions.

 

Couldn't have said it better.

The main point that I think needs to be made is that the hack does decrease IQ in some spots. Given that, what the hack allows us to do is perform AF correctly, which will increase the IQ and performance when playing with specific settings.

On a side note: I don't really see a point in needing to include this in any benchmarks for game or hardware reviews. The only article I'd like to see this hack included in along with benchmarks is an article with the sole purpose of detailing this hack and its implications. The article shouldn't be 33 pages long though, my eyes already hurt.

Great job to everyone who helped out with this, its been very much appreciated!

 

if you want to getting playable as an 8500,try to turn off the shadow effect, and you will get 10 or more fps improve and only a few quality loss

 

Re: Couldn't have said it better.

Decrease where? I thought that with the latest form of shader changes there were no IQ problems at all.

Exactly, the image problems are caused by AF not by the shader changes.

Sure there is a point! If users want to hear about a mode that improves IQ and helps performance, isn't that important? Shouldn't the performance improvements be shown and compared to other results?

-FUDie

 

As an Nvidia user I am glad that you guys have found this work around for what appears to be related to the ATI drivers. Heck even if it isn't related to the ATI drivers, I am still glad you guys have this tweak. I still feel pretty confident in my purchase of a 6800 ultra and am extremely happy with the way doom plays on my machine. I have nothing to complain about if you guys find a way to enhance your D3 experience.

The original post does come across a little harshly towards JC. I think that it is clear that John Carmack had a reason for doing what he did and to make a statement such as

is a little bold. Carmack clearly wasn’t writing the game to be optimized for the current generation hardware. He had a much lower standard set in his sights. The fact that some of these older cards can play D3 and look fantastic really leaves very little room for anyone to complain about his work.

Good work to the people involved in making this tweak. It is fantastic that you guys are able to find ways to enhance your D3 performance. I hope that ATI/id will be able to put something official together for their customers even if it means the 6800 ultra is knocked off the top of the benchmarks for D3.

To me D3 is all about re-living moments from the first real FPS I ever played. Doom was the game that got me into computer games. Quake and quake2 set the hook completely. What I’ve noticed is that if I keep the frame-rate counter on the screen it kills the atmosphere for me. With that in mind, I couldn’t care less about benchmarks because as long as my game plays smoothly I am personally satisfied.

Now about HL2… if my 6800 ultra doesn’t play smoothly in that game I am sending Humus a PM for help.

-Nathan

 

So, you're saying it would be wrong to run the Humus tweak on both nVidia and ATi hardware, or that it's wrong to run the Humus tweak on ATi but not on nVidia, or that it's wrong *not* to run the Humus tweak on ATi because it's unfair that although JC's generic approach is quite optimized for nV it isn't as optimized for R3/4x0?

Would it be correct, if you feel the Humus tweak is somehow immoral (although it actually improves both IQ and performance in D3), to disable all nVidia D3-specific driver optimizations on the grounds that they are also immoral for the same reasons?

I could easily see and agree with the point I *think* you are making, if it was true that the Humus tweak merely managed to increase performance while degrading IQ--but since it improves *both* performance and IQ in the game--I cannot see where using it would in any manner be unfair to anyone--unless you'd consider the only "fair" nV driver for D3 to be the one in which all D3-specific code optimizations have been disabled, as well.

As has been said so many times before, there are good optimizations and bad optimizations, the difference being that the good ones do not force you to pick between IQ and performance. To that end, I think your labeling of Humus' efforts as "tweaks" is entirely correct and proper--it *is* an actual tweak to the generic code of the game--and it's a tweak only made possible because Carmack programmed in such a fashion as to *allow* exactly this kind of tweak to take place in D3. Indeed, in the B3d interview text as compiled by the Rev and published, Carmack has nary a negative word to say about the Humus tweak, and even hints that he might have done much the same thing given more opportunity to do so.

So it seems to me that if Carmack himself sees no reason whatever to call Humus' tweak a "hack" or "improper" or "detrimental to D3's IQ"--it seems that the only actual immoral position there is with respect to this issue is a position that categorizes Humus' tweak as anything but desirable and perfectly acceptable.

Simply put, if a person believes that disabling shadows is a perfectly acceptable, moral way to play the game (as well as manipulating dozens of other settings, too), and he realizes that Carmack's own deliberate coding is that which made it possible to disable shadows in the first place, then he cannot find any fault whatever with Humus' tweak, since it is accomplished in exactly the same way that shadows are disabled, by the adjustment of settings relative to the game engine which Carmack deliberately and intentionally made accessible to user manipulation. Just my opinion, as usual...

 

Waltc i would have to say that right now it should only be benchmarked for the sake of exploring things.

IT shouldn't be used as a comparision for which card to buy.


I think a site should look into all of these optimizations and see what is going on.

I mean why not. It will only help those looking for more performance so that they will have (hopefully a non bias person) judging image quality and testing to see if performance gains are really to be had .

If we later find out that by some grace of god all of these optimizations end up with the x800s catching the 6800s and the r3x0 class out pacing the nv3x card . Then perhaps we should take a hard look at whats going on with id and nvidia .

 

I seem to have to use

MAD_SAT R1, specular, 3.0, -2.0;
rather than,
MAD_SAT R1, specular, 4.0, -3.0;

to avoid getting artifacts.

Since I am completely clueless about what's being discussed here, is altering this line from the default tweaked setting resulting in a degradation of the visuals in any way, or the performance? I meean will it look better and run faster with specular, 4.0, -3.0 rather than specular, 3.0, -2.0?

Now I never had any artifacts on my 9800 Pro before using this tweak, and was a bit surprised that it caused them. I'm confused as to why I'm getting the artifacts at the "MAD_SAT R1, specular, 4.0, -3.0;" setting and others don't (or at least aren't reporting it). Is it just random?

Are there other valid values to try for the specular settings?

No matter what I definitely see the performance benefit. So, thanks.

 

I get curruption at the outside spot in the beggining using the fix with my Radeon 8500:

With fix
http://home.comcast.net/~cc_silent/pics/shot00002.jpg
It puts like a dark black fog and I can't see at all. It's fine inside though.

Without
http://home.comcast.net/~cc_silent/pics/shot00003.jpg

 

Uhm, yes?







I meant that I thought it would be a good thing to add an additional bench of a run on the Humus' tweaked version in addition to a bench of the regular non-Humus' tweaked version of Doom3.

You can go ahead and bench it on anyone card you like that way.

I guess I do have a bit of a problem right now comparing a straight run of a 6800 on Doom3 compared to an X800 on the Humus' tweaked version, even though it is baseless and silly as you pointed out.

But I think it wouldn't sit right with me, not until the tweak is somehow officially endorsed by id.

 

First off, thanks for this tweak. This thread has certainly continued to grow. I think that is quite a testement to how helpful the information included is proving to diverse users.

Also, might I suggest you add a tiny "how to install this tweak" .txt file to your download. Perhaps people would quit asking the same question over and over again then. I guess they are too lazy to scan thirty pages. Perhaps some sort of version numbering would be helpful as well. That way there would be an easy method to determine if one is running the most recent version (i.e. the one without the IQ issues). Cheers.

 

Yah ....

The last Nvidia card I owned was a Geforce1 and I can't remember how many years ago that was. I owned an ATI Rage Fury or something, went to the Geforce1, then up to a Kyro2, then up to my Radeon 9500Pro. I wouldn't exactly call that a fan boy.

I just don't think any reviewers will be presenting this modified shader as the definitive Doom3 benchmark because it does not represent the performance an ATI user will get from Doom3 out of the box. If this shows up as a patch from Id, which I'm hoping will happen, then benchmark away. I guess if the benchmarks explicitly state that this tweak has been performed, how it is performed and how much performance is gained, then I wouldn't have a problem. As long as people understand that if they buy the game and install it, they won't be getting that performance out of the box. And just to be clear, I'm not talking about some guy on the internet posting benchmarks. I'm talking about internet published review benchmarks on proper hardware and gaming sites. I just don't see hardware reviewers benchmarking the game using this tweak unless it's included in an Id patch or something. ATI is in a tough spot, because they'll probably take flak for shader replacement and app detection if they try to do it in the drivers.

 

You people saying that ATI cards should use the tweak in benchmarks are idiots. He's approximating a function in a major shader in D3, which had a very specific purpose for being written the way it was, and now you people are saying, "Oh, it's magically delicious and we should benchmark our cards with it?" Hello? Did you think that NVIDIA was right in replacing major shaders in 3DM03 GT4 with approximations because they ran faster and gave almost identical IQ? Are you people totally insane?

*shakes head and remembers why he stopped dealing with this crap*

 

This tweak is meant for ARB2 so it shouldn't be working with the R8500.

 

I can't speak for Dave or any other B3D reviewer but I always benchmark a game as it is delivered (with patches applied, whenever possible of course) without *ANY* kind of "tweaking" in a product review or even in a next-gen first-view preview.

For the purpose of a, uh, "investigative" article, I wouldn't hesitate to experiment (like what Humus has done her) and provide any possible findings.

Has anyone here in this thread ever mentioned "What a can of worms Humus has opened up..." ?

 

maybe u didnt read, but didnt JC say in an interview that he wrote the code that way simply because it would allow support for older cards. so by changing the code it is simply optimized for newer ati cards.