Check out this thread

[Xentropy]

Originally posted by: Curley
Yes, I am pushing the conspiracy theory.

. . .

Setting optimizations and controlling benchmarks to make one piece of hardware or another look better
is what I found irritating.

Maybe the man behind the grassy knoll (an nVidia employee, no doubt) called his alien friends to control Carmack's mind and cause him to write one questionable shader. :roll:

-----

Pete,

Basically, ATI tweaks good, nVidia tweaks bad? Or is mathematically unequivalent tweaking of real games ok but not the same of benchmarks? Or is it okay for an employee to be "unsanctioned" and do the tweak "on his own time" (wink wink) as long as nothing's "official"? If you don't see the hypocrisy others see here, you're turning a blind eye to it.

 

[nemesismk2]

Can anyone confirm that with the tweak there is NO texture filtering at all?

 

[gururu]

i've never written any sort of code, so I'm left to wonder...
is this the kind of thing that say, JC, would not have known about? Or would he more than likely be aware of it but picked the option he thought was best, the lesser of two evils?

 

[Shagga]

Originally posted by: Schadenfroh
most of the people in that thread report gains of like 4 FPS, that doesnt even come close to 40%, well, maybe if you have less than 10 FPS

Just tried it myself.

1024 x 768, High Quality 44.90fps

With patch...

1024 x 768, High Quality 47.80fps - Increase of 6.5%. I didn't notice an IQ problem.

 

[Pete]

Originally posted by: Xentropy
Pete,

Basically, ATI tweaks good, nVidia tweaks bad?

No.

Or is mathematically unequivalent tweaking of real games ok but not the same of benchmarks?

As long as it's not slipped into the drivers or presented as official/company policy, but presented as an enthusiast's tweak that serves to improve gameplay rather than marketing numbers (because 3DM03 scores are ultimately nothing more than marketing numbers, you realize), then yes.

Or is it okay for an employee to be "unsanctioned" and do the tweak "on his own time" (wink wink) as long as nothing's "official"?

Without the wink wink, it's OK, as I thought I sufficiently explained in my post, but elaborated on above. With the wink wink--if this is in fact ATi masquerading behind Humus--well, who gives a flying fig? Websites aren't slipping this into their benchmarks yet, are they? And don't you think they'd note the currently reduced/affected IQ? This is more like the option nV gives you to disable trilinear optimizations than the "option" they gave you when they switched in their 3DM03 shader replacements. You may note a quote from JC himself that was noted in Humus' B3D thread:

It is going to require fairly deep, non-backwards-compatible modifications to
an engine to take real advantage of the new features, but working with
ARB_fragment_program is really a lot of fun, so I have added a few little
tweaks to the current codebase on the ARB2 path:

[...]

Light and view vectors normalized with math, rather than a cube map. On
future hardware this will likely be a performance improvement due to the
decrease in bandwidth, but current hardware has the computation and bandwidth
balanced such that it is pretty much a wash. What it does (in conjunction
with floating point math) give you is a perfectly smooth specular highlight,
instead of the pixelish blob that we get on older generations of cards.

I think this is directly related to Humus' tweak, relpacing cubemaps with math.

If you don't see the hypocrisy others see here, you're turning a blind eye to it.

No, a more discriminating (in the sense of enlightened, not biased) one.

 

[darkswordsman17]

Gururu, what JC did is just write code that worked the best for a wide range of hardware, instead of writing optimized code for specifice cards. This would have been the best path because it runs well on a wide range of cards, and they could get the game done earlier.

The problem here is that it works for ATi cards, but not nVidia, or so it seems to me. Whenever trying to play online this is probably gonna cause problems due to the differences, but I don't know this for sure.

Its nice to see the ATi camp get some good news, but there's still some problems, like ATi's filtering optimizations that cause a small, yet noticeable difference in the visuals. There's a thread about this, so I'm not going to go into detail about it.

Also, I think that the people who have called JC's integrity to question, you should send him an email apologizing. How many of you fanboys were ready to burn him at the stake because you were disappointed in your performance and then found a somewhat solution and thought to immediately accuse him of purposefully sabotaging his game that he slaved to work on for several years? Does it make any sense for him or iD to do this? There's no amount of money that nVidia could pay them to do such a thing, especially since they would have to know that they would be under the greatest scrutiny and would have their heads on the chopping blocks if they allowed such a thing to transpire. It is an illegal business practice that would be punishable by law.

 

[clicknext]

I was playing around with this yesterday, and it honestly works quite well, but only on the right settings. Here, I recorded the settings where it made the biggest difference, and thus I will use this setting to play from now on.

---
System:
Athlon XP 2.2Ghz
ATI Radeon 9600 Pro
Asus A7N8X-E Deluxe
---
Tests:
800x600, High Quality, 16XAF, No AA
Humus Shader: 40 FPS
Normal Shader: 28 FPS
---

Pretty damn close to 40%, wouldn't you say. ^_^ What I found is that the higher anisotropy is set, the more performance. Furthermore, it performs faster AND looks better on high quality. Here are a complete set of my results, not really well organized, but the results are certainly interesting.

---
Normal shader, no AF, medium:38 FPS
Normal shader, 8XAF, medium:34 FPS
Normal shader, no AF, high:29 FPS
Normal shader, 16XAF, high: 28 FPS
---
Humus/Demirug shader, no AF, medium: 37 FPS
Humus/Demirug shader, 8XAF, medium: 38 FPS
Humus/Demirug shader, 16XAF, medium: 40 FPS
Humus/Demirug shader, 8XAF, high: 39 FPS
Humus/Demirug shader, 16XAF, high: 40 FPS
---

I'd like to add that I do not notice one bit of image quality difference. At first, when I added the "PARAM specExp = { 16, 0, 0, 0 };" line, there were the white spots but I took that chunk out because it doesn't give me any extra performance, at least with the MAD_SAT line.

 

[PhoenixOrion]

Bump here and PM to you clicknext.

 

[Childs]

Originally posted by: clicknext
I was playing around with this yesterday, and it honestly works quite well, but only on the right settings. Here, I recorded the settings where it made the biggest difference, and thus I will use this setting to play from now on.

---
System:
Athlon XP 2.2Ghz
ATI Radeon 9600 Pro
Asus A7N8X-E Deluxe
---
Tests:
800x600, High Quality, 16XAF, No AA
Humus Shader: 40 FPS
Normal Shader: 28 FPS
---

Pretty damn close to 40%, wouldn't you say. ^_^ What I found is that the higher anisotropy is set, the more performance. Furthermore, it performs faster AND looks better on high quality. Here are a complete set of my results, not really well organized, but the results are certainly interesting.

---
Normal shader, no AF, medium:38 FPS
Normal shader, 8XAF, medium:34 FPS
Normal shader, no AF, high:29 FPS
Normal shader, 16XAF, high: 28 FPS
---
Humus/Demirug shader, no AF, medium: 37 FPS
Humus/Demirug shader, 8XAF, medium: 38 FPS
Humus/Demirug shader, 16XAF, medium: 40 FPS
Humus/Demirug shader, 8XAF, high: 39 FPS
Humus/Demirug shader, 16XAF, high: 40 FPS
---

I'd like to add that I do not notice one bit of image quality difference. At first, when I added the "PARAM specExp = { 16, 0, 0, 0 };" line, there were the white spots but I took that chunk out because it doesn't give me any extra performance, at least with the MAD_SAT line.

Anyone have an explanation as to why its faster using 16xAF vs 8xAF? Thats never happened before and seems counter-intuitive.

 

[clicknext]

I think I remember Humus saying something about that in his thread, but either I don't remember it or it was too high level for me to understand.

http://www.beyond3d.com/forum/viewtopic.php?t=14874&postdays=0&postorder=asc&start=0

Here's the link if you've got the patience, but no guarantees...

 

[Genx87]

You might want to refine your sense of irony. When nV substituted shaders in 3DM03, they did so without telling anyone until they were caught, they used their whole company to discredit 3DM03, and they never once apologized for cheating in a benchmark.

Not quite. I am sure if we go back a year or so we will see the same people who praise this were bashing Nvidia for doing the same thing. So I ask what is the difference? Both are hacks, both do not give the same exact pic quality.

Why are you pandering to the lowest common denominator in this discussion? nV never "gave out" driver tweaks, it basically slipped them into its drivers without telling anyone. nV certainly was cheating, though the devil comment is hyperbolic fanboy talk.

Hyperbolic fanboy talk also results in people claiming "ID is working against ATI." Why do you bother listening to people ignorant or biased enough to make such statements? Do you enjoy bringing this forum down to the level of soccer hooligans? Who or what exactly is the "it" that is proclaiming that "ID is working against ATI," BTW? Surely you can't mean ATI itself, as it hasn't commented on this--but then, who?

Did you even bother to read the original posters post?

 

[GeneralGrievous]

Not quite. I am sure if we go back a year or so we will see the same people who praise this were bashing Nvidia for doing the same thing. So I ask what is the difference? Both are hacks, both do not give the same exact pic quality.

One was an official hack released in drivers. The other was an unofficial, voluntary "hack" released on a forum.

 

[Keysplayr]

I cant wait until it's time to hack apart HL2.

 

[clicknext]

This tweak was not created by ATI. It was created by someone who's only interest is to improve performance, not look endlessly for graphical imperfections. Regardless of who he works for, he represents himself, and only himself.

Now, can we PLEASE talk about the tweak itself instead of waste summer time arguing about nothing?

 

[Childs]

madshi:

(1) Doom3 does a lot of work through texture lookups, which could also be done by math instead. On older graphics cards texture lookups are faster.

(2) When Doom3 asks for AF, it only asks for those texture layers and accesses, which need AF. So the AF cost is not that big.

(3) When you force AF through ATI's control panel, all texture layers get AF, and also the texture accesses described in (1) get AF - although it makes no sense there! The result is a *hefty* drop in performance on ATI cards.

(4) Humus hack replaces the texture lookups in (1) by math. As a result the problem described in (3) goes away, since math doesn't get AF.

He seemed to sum up what some of the others thought was happening with AF.

 

[clicknext]

That sounds like it makes sense, but it doesn't exactly explain my performance INCREASE when using higher levels of AF.

 

[Childs]

I don't quite get it myself. You would think that 16xAF would be the same as 0xAF not faster than doing no filtering at all!

I would think that ATI's CP is doing something like brilinear instead of full trilinear. The last part of madshi's post came to that conclusion, but no one else in the 27 pages thought that was the case so I left it out. The precision somewhere has to be dropping in order to get an increase over no filtering.

 

[Pete]

Originally posted by: Genx87
Did you even bother to read the original posters post?

I ignored the obvious flamebait (nobody tells JC what to do). You're generally above that, though, and you saw the OP for the flamebait that it was. Yet you went on to toss out some flamebait of your own, so I called you on it. I wouldn't have wasted my time if you didn't post intelligently at least some of the time. ;P The fact that keysplayer immediately echoed your post made me want to correct your false links even more, before more people joined your rather uninformed bandwagon.

 

[clicknext]

Originally posted by: Childs
I don't quite get it myself. You would think that 16xAF would be the same as 0xAF not faster than doing no filtering at all!

I would think that ATI's CP is doing something like brilinear instead of full trilinear. The last part of madshi's post came to that conclusion, but no one else in the 27 pages thought that was the case so I left it out. The precision somewhere has to be dropping in order to get an increase over no filtering.

Hmm, I'm using performance 16XAF as I always do, not quality 16XAF. I heard somewhere that performance is bilinear and quality is trilinear, not sure of it's true or not. However, I did all my tests on performance and 16X was faster than 8X.

 

[Childs]

I have the feeling something sneaky is going on, as there should be no performance increase over no filtering, all things being equal. It can conceivably be the same, but not faster.

clicknext, which drivers are you using? The 4.9betas?

 

[gururu]

Darkswordsman17 wrote:

Gururu, what JC did is just write code that worked the best for a wide range of hardware, instead of writing optimized code for specifice cards. This would have been the best path because it runs well on a wide range of cards, and they could get the game done earlier.

thanks!
I feel the same way. We'll see if JC or ATI publicly respond.

 

[clicknext]

Originally posted by: Childs
I have the feeling something sneaky is going on, as there should be no performance increase over no filtering, all things being equal. It can conceivably be the same, but not faster.

clicknext, which drivers are you using? The 4.9betas?

4.7's.

 

[Noid]

Originally posted by: Kobra
I pointed out a flaw with 9800's months and months ago that can cause 9800 overheating a large portion of their cards... Theres a metal retaining border around the GPU, during manufacturing, this can be placed slighting crooked on the board - which seems to happen a good majority of the time.. When this happens, the heatsink and fan don't seat tightly enough on the 9800's, and causes overheating because 9800's already run hot.

This problem is compounded because of the poor heatsink and fan on the 9800. Result is artifacts, tearing, overheating, crashing and lockups. I harped on this problem with ATI for months, RMA'd 2 cards, and both had the same problem.. I'm RMA'ing my last 9800Pro to them this week, then throwing it up on Ebay to try and recover some of my losses from those bastards.

You have to be kidding ... right..?
(No ,, I guess your not ... lol )

This isnt news by any means. This was ocurring 1st on the 9700pros, (they used the 'shims' 1st)
(I replaced my HSF unit as soon as I got it)
And,,, if you been building systems awhile,,, you know it's NOT just Ati with poor attachment of HSF units.

I ALWAYS check under ALL my new components with HSFunits for proper cooling.

 

[StrangerGuy]

Well, I applied the "hack" and the results are pretty shocking to me.

Stock 9800 Pro / Official Catalyst 4.7

10x7 / High Quality / Timedemo demo1
Before: 28fps
After: 40fps

That's is a 40% fps increase, and I didn't noticed any form of image degradation. All for free.

 

[clicknext]

Originally posted by: StrangerGuy
Well, I applied the "hack" and the results are pretty shocking to me.

Stock 9800 Pro / Official Catalyst 4.7

10x7 / High Quality / Timedemo demo1
Before: 28fps
After: 40fps

That's is a 40% fps increase, and I didn't noticed any form of image degradation. All for free.

Exact same settings and improvement as me... except I'm at 800x600 because of my 9600Pro.

All kinds of people (mainly nvidia users) are saying that we are sacrificing image quality for FPS, and it may be true, but if it's image quality that we can't even tell is gone, then there's no drawback!