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

Great job on the mouseovers, Apple740. I'm certainly sticking to it.

 

To be honest, I hoped that your CP AF also did look better, just as in Ati's case. If so, Ati would close (more or less) the gap in performance between NV and them.
But now that this is not the case Nv still "owns" this game. Congratulations.

 

Let me know if you can host the pics and I'll send you two PNG's with app and cp AF.

 

For evaluating App Con/Con Pan AF differences

What are the different texture layers applied in the game?
How many are color textures?
How are the non-color textures affected by AF, and how are they treated when the application has control?

From the notes in the interaction.vfp, per-surface textures include: a bump map, a diffuse map, and a specular map. Are the items noted a complete representation?

 

This is an issue that was noticed some time ago. I believe it's caused by anisotropic filtering set for the minification filter, and linear filtering set for the magnification filter. On ATI hardware, this causes aliasing similar to what you show in that screenshot. I don't think nVidia hardware has the same problem, due to more sane definitions of magnification and minification.

 

Less flexible, you mean.

-FUDie

 

Less flexible? What on Earth are you talking about?

 

Your comment about "sane definitions of magnification and minification". You act like what NVIDIA does is the only way to do things. Of course, I would expect nothing less from you.

-FUDie

 

can u plz answer my question:
1.in what page i can see the latest tweak? (not the one in page 1. i know there are new tweaks in other pages)
2.does DOOM 3 gonna have a 64 bit version?
3.i have
amd 64 3000
radeon 9800pro 128
1 gb corsair
how much fps shuled i get with 4 aa 8 af and high texture setting?
4.in page 15 i saw a picture of soemthing called "doom 3 settings" what is this thing?
i didnt notice it in my doom.

 

I had used Humus latest tweak and was getting about a 1 fps difference overall the game was running much smoother.

AMD 2800, 9800XT, 1.25 GB RAM, 1280x1024, 8x AF GP, Ultra Quality

Second run:
w/o Humus 28.9
with Humus 30.0

Today I decided to overclock my CPU to see if I could increase the performance a bit. I found that the overclocking I was doing was only giving me about .5 frames and causing some instability in the system. So I changed it back to normal and I noticed that I had AGP aperture set 64MB. I decided to change it to 256MB same as my card. I was very surprise at the difference.

Same settings as above but with 256MB AGP aperture 8x in CP
Second run:
w/o Humus 25.6
with Humus 32.4

So for the people that are getting just a frame or two with Humus tweak check your BIOS setting.

 

In OpenGL, you can't set anisotropic for minification and magnification independently. It is a setting "on top" of min and mag filters.


I can confirm there is a bug in the Catalyst drivers (at least in the latest Omega drivers based on Catalyst 4.7 with Mobility 9600) regarding application controlled AF and magnification filter in OpenGL.
This bug is most likely the reason for the issues in Doom3.
If you can read German, here is one instance where the problem occurred before.

You can also use my FilterTest app to reproduce the issue:
Set CP to app
start FilterTest
set checkerboard texture 0
set AF to 16x
set magnification to NEAREST, then back to LINEAR

It seems like after you set AF, you have to set the mag filter again to get anisotropic mag filter.

 

i am not sure HT looks inferior when you look the lower left floor

but i personnaly use the SUB_SAT .x mod from demirug which is much less different from the table lookup (photoshop diff shows a nearly black result)

 

It's much more conspicuous in ID's case, however, given the wholly misleading remarks Carmack made recently where he negatively disparaged the fact that ATi was doing trilinear optimization (while admitting it did not impact IQ in D3--which makes you wonder what his point in mentioning it in the first place actually was), while he completely failed to *mention* that nVidia is doing the same general type of trilinear optimization in its drivers by default.

Since there is no possibility that Carmack didn't know that nVidia was doing the same thing, only more egregiously (since when ON in the drivers by default nVidia's method does not switch off in the presence of colored mips or other conditions in which IQ is degraded by it), there's in no doubt whatever in my mind that he demonstrated a partisan point of view for an IHV, unrelated to software or hardware technology, in those statements. With regard to the trilinear optimization buzz, Carmack in his public statements made no effort to be even handed, equitable, or objective in his assessments.

So not only is ID software capable of seeing and commenting disparagingly on certain aspects of ATi technology, but incapable of drawing the same conclusions surrounding similar nVidia technologies (or even mentioning the fact that they exist in the first place), it now doesn't even appear as if ID is capable of spotting blatantly IHV-unoptimized conditions in its game engines--with the glaring exception of nVidia.

Last, to turn your "perhaps" speculation around, perhaps JC's already told ATi that the next patch would cover the problem, and so ATi hasn't commented directly because of that? That seems equally likely to me, in "perhaps" land...

[I see Humus has addressed your comments relating to his work so I won't comment on them.]

And "perhaps" the logic was simply to promote one IHV at the expense of another. When such promotion is based on objective reality, it doesn't bother me (for instance, no one need ever be ashamed of preferring nV25 to R200, imo, or R3x0 to nV3x, etc.) When it is based on a deliberately manipulated, tilted playing field, though, it bothers me. What your above "perhaps" musings do not explain, however, is why ID prior to shipping the game did not catch these things and fix them.

So, perhaps D3 shipped exactly as ID intended it to ship, whatever the configuration/rendering shortcomings were with respect to ATi's technology? That's a shame, really, as a great percentage of potential D3 customers own ATi hardware, and that's why only an idiot publisher takes part in partisan IHV marketing games which are based on fiction instead of objective reality. That's my opinion, anyway. Game publishers should be true to their customers first, themselves second, and IHVs third. I'm confident that with the proper level of effort expended by ID to *support potential D3 customers who own ATi hardware* that we would not be having these discussions at this time.

The "banding report" is a red herring resulting from gross abuse of the gamma/brightness controls in the game, and is equally applicable to all IHVs in all 3d games where such adjustments are possible.

I'll take you seriously when you apply this advice to yourself and all of the other people posting in this thread who have been saying either that Humus' tweaks don't do anything, or that there are bugs in the ATi drivers, or both--people who have been blaming everything *except* the new, yet-unpatched D3 engine's stock configuration. Rather, it seems to me my comments about Humus' tweak prove conclusively the "distinctively unoptimized" nature of the shipping D3 setup--else the minor changes Humus made would already have been done, and they'd not have resulted in a slow down for nV3x (which is precisely why, it appears, they were not done by Carmack and pals in the first place...)

 

Walt, the app AF issue is a bug in the driver, and Hanners' article shows that not everyone thinks it's that obvious.

Humus' changes result in a slowdown on NV3x hardware regardless of whether there are shader replacements going on or not.

 

What about X800 SE

Hey all,

Maybe I just missed it, but have any of you done any benchmarking with an X800 SE, that OEM-only board? I just ordered a Dell with one, and I was wondering if I should have taken the baseline X300 and hung on to my $120 bucks so I could buy NVidia later.

 

..while providing a control panel option to switch it off.

His statement was an objective truth nonetheless.

But it turns out it's not all that unoptimized and blatant if you don't use CP-AF.

 

What "AF issue" is that, exactly? (It's difficult to keep track of the different "AF issues" alleged)...

As to Humus' changes, that's why Humus' changes, or their equivalent, which simply optimize the D3 config settings to more efficiently support R3/4x0 hardware, weren't included in the shipping D3 config--because they *did* slow down nV4/3x, apparently. I guess, too, that it's simply a matter of the D3 engine not being very advanced with respect to it's general approach to fragment hardware--Carmack's own comments in the B3d interview as written by the Rev seem to support that perspective (As in, papraphrased, "If I'd known ahead of time how much time I'd have before the game shipped I might have approached this issue differently than I did," etc.)

 

Right, as the ATi optimization *switches itself off* in conditions where it would degrade IQ automatically, no user intervention required. Carmack *doesn't even mention* nVidia trilinear optimizations at all, in any capacity, is the point. From his comments it's not apparent that nVidia has *any* trilinear optimizations, when the fact is that they are on by default, and when on *do not* auto-switch off (as do the Catalysts.) Not only does Carmack not mention that you can turn off nVidia tri-opts, he doesn't even mention they are turned on in the nVx drivers by default... He just doesn't talk about them at all in the immediate context of disparaging ATi's tri-opts approach, which I find remarkable...

Parenthetically, the fact that a manual Cpanel switch for on-off trilinear optimizations is grossly impractical considering what trilinear optimizations are and what they do inside a game (not remotely in the same category as general AF or AA, for instance), unless you are someone who doesn't mind leaving the game and switching the Cpanel tri-opts on and off everytime you need to do so inside the same game... That would be very distracting to me--don't know about you, of course.

No it wasn't, because what he didn't say was more important than what he said, but you don't want to see it so there's no sense in repeating myself endlessly...

It's the opposite for me--I get better IQ with the Cpanel overrides in D3--but with all of my other games, including Q3--I'm better off using the in-game filtering controls and setting the Cpanel to App Preference so that the game engine tells the driver how to filter. D3 is the only exception I have to that rule, including Q3.

 

Nvidia's optimisations are optional and the driver even comes with a way to set it on a per game basis individually. ATI on the other hand does not provide a way to switch them off at all - you are completely reliant on the autodetect mechanism. You seem to think that's better, to me that's odd, but different strokes for different folks I guess.

What wasn't said was more important to *you*.

Nonetheless, his statement was accurate.

As you've said, this is v1 of the game engine - maybe ingame filtering on Ati will improve later, or just maybe it will take a driver update to fix this. I think we'll get more clarity on this issue soon enough.

 

Obviously, the single "app AF issue" you were talking about here: that Doom3 app AF causes artifacts on ATI hardware, but CP AF does not.
It is a driver bug, it can be reproduced with other apps than Doom3, and the reason you haven't seen it in another game is that there are few OpenGL games with app AF control.

So what do you think Carmack should have done?