John Carmack Responds to the ATI Humas tweek

Status
Not open for further replies.

BakedGoods

Limp Gawd
Joined
Jul 23, 2004
Messages
312
rancor said:
If he does it will probably be the next patch, but its highly unlikely.
Yeah, rancor, out of everyone here you seem to know a lot of technical mumbo-jumbo. Why, in your opinion, do you believe it is highly unlikely?
 

KSL

Weaksauce
Joined
Aug 7, 2002
Messages
123
I have tried the Humus tweak for Doom3, apparently no luck with it as my 9600pro wont give anything over 23fps during timedemo. Anyone know a way around this? Or is there something else I could do? BTW, I am currently using Ati 4.9b drivers.

Any help is appreciated

KSL
 

rancor

2[H]4U
Joined
Jul 22, 2004
Messages
2,230
KSL said:
I have tried the Humus tweak for Doom3, apparently no luck with it as my 9600pro wont give anything over 23fps during timedemo. Anyone know a way around this? Or is there something else I could do? BTW, I am currently using Ati 4.9b drivers.

Any help is appreciated

KSL
Whats you cpu?
 

rancor

2[H]4U
Joined
Jul 22, 2004
Messages
2,230
BakedGoods said:
Yeah, rancor, out of everyone here you seem to know a lot of technical mumbo-jumbo. Why, in your opinion, do you believe it is highly unlikely?

If this was found before the release there would have been alot more weight ATi could have pushed it on Id, but doing it after the release, its completely up to Carmack, One of his interviews he already stated he was working on his next engine.......

Its not something hard to do, I just don't see the reason behind it to do it, as someone mentioned early if he does that then consumers will expect him to make specific optimizations for nV cards too.
 

The Batman

[H]ard|Gawd
Joined
Jul 20, 2004
Messages
1,531
I'm hoping it's a retelling of D2. Just imagine Doom on Earth + Double-Barrel Shotgun. And if you can't imagine...just go watch the Halo 2 trailer.
 

R1ckCa1n

Supreme [H]ardness
Joined
Jul 22, 2004
Messages
4,355
eXzite said:
Wait, does he actually *work* for nvidia? Or by nvidia security guard, does that just imply he defends nvidia because he likes them.
LOL.. no, or at least I don't know that answer. It was my way of saying he defends Nvidia to the end even if it's not necessary.
 
Joined
Aug 28, 2003
Messages
567
rancor said:
Got to look up to the guy, makes a nice engine then goes ahead and makes another one lol
What else would he be doing? Going to work every day and playing soccer in his office?
 

rancor

2[H]4U
Joined
Jul 22, 2004
Messages
2,230
well his hobbies are buying farrari's and working on space ships lol nice life I'd say.
 

BoogerBomb

Supreme [H]ardness
Joined
Jan 10, 2003
Messages
6,470
burningrave101 said:
I was making mention of the reasons why i have no respect for ATI and not just Humus himself who has it out for nVidia because of the fact they wouldn't hire him lol.

Kyle = HardOCP and HardOCP = Kyle

Humus = ATI and ATI = Humus

If you work for a business, your actions reflect the business as a whole and that is why the owner of the business chooses his employees wisely.
What company do you work for? I will make sure I never buy anything from your company just because of your opinions and since you work for them you speak for your employer as well.
 

R1ckCa1n

Supreme [H]ardness
Joined
Jul 22, 2004
Messages
4,355
I hope this doesn't turn bad like the other forum threads. I have not had time to try these tweaks but do expect to tonite.

Anyone - post your results and make sure you test with AA and at least 8x AF. :p
 

BoogerBomb

Supreme [H]ardness
Joined
Jan 10, 2003
Messages
6,470
I treied version 2 (I think) on a 9800Pro and got a 6FPS increase. I have not tried the latest one but will and post my results.

Everyone that is testing it make sure you run the timedmoe TWICE. The second run will give you much more accurate results because the textures are cached.

Oh and I think its sad that Captain got banned for name calling when there are other posters who have broken the rules as well yet nothing happened to them. Too bad we can't go back and report posts from the other thread because a lot more would prolly get banned.

EDIT: Saw that the sticky post I read a few months back about the namecalling problem and the rule instituted afterwards is no longer stickied. Any chance of another one being made for this sub-forum since D3 will always incite flamer-wars between ATI and Nvidia owners?
 

^eMpTy^

2[H]4U
Joined
Jul 21, 2004
Messages
3,233
R1ckCa1n said:
I hope this doesn't turn bad like the other forum threads. I have not had time to try these tweaks but do expect to tonite.

Anyone - post your results and make sure you test with AA and at least 8x AF. :p
I read over on the b3d forums that the problem has everything to do with AF...so that's the only place you'll see an appreciable boost...
 

eXzite

Limp Gawd
Joined
Jul 22, 2004
Messages
144
^eMpTy^ said:
I read over on the b3d forums that the problem has everything to do with AF...so that's the only place you'll see an appreciable boost...
The problem is that the dependant texture used for the specular lookup is a 256x1 texture, and when you do aniso taps on that, you thrash your texture cache. IF you use Doom3's aniso setting in the doom3 control panel, the game knows not to do aniso on a dependant texture lookup, and therefore you get little perf benefit. If, however, you force aniso on everything either from the windows control panel or console, then you get a jump in perf because aniso is now on every single texture (this dependant one included), and you were doing multiple taps on an 256x1 texture before, thrashing your texture cache, where as now you are doing the standard specular calculation with a math operation.
 

PureBooYah

2[H]4U
Joined
Aug 26, 2002
Messages
3,363
well i just tried it out, i was staring down a hallway getting 30fps at 1600x1200 16xAF, high quality (used FRAPS) and wasn't moving at all. Put the tweak in and it went up to 45fps doing the exact same thing.

i'll take 15fps :D

edit: no AA btw..
 

burningrave101

[H]F Junkie
Joined
Sep 9, 2003
Messages
11,825
BoogerBomb said:
What company do you work for? I will make sure I never buy anything from your company just because of your opinions and since you work for them you speak for your employer as well.
Most of my opinions are facts and not just opinions so if you'd rather buy into lies then go right ahead.
 

R1ckCa1n

Supreme [H]ardness
Joined
Jul 22, 2004
Messages
4,355
burningrave101 said:
Most of my opinions are facts and not just opinions so if you'd rather buy into lies then go right ahead.
You didn't answer the question. I'm thinking about boycotting your company too since I don't like your opinions. ;)
 

R1ckCa1n

Supreme [H]ardness
Joined
Jul 22, 2004
Messages
4,355
^eMpTy^ said:
I read over on the b3d forums that the problem has everything to do with AF...so that's the only place you'll see an appreciable boost...
That could have been why I asked for AF.......

I'll take a 16X AF at the price of 8X everyday; or better yet a free 16X.
 

burningrave101

[H]F Junkie
Joined
Sep 9, 2003
Messages
11,825
R1ckCa1n said:
You didn't answer the question. I'm thinking about boycotting your company too since I don't like your opinions. ;)
Thats actually a good thing. Companies are better off not having people like yourself buy from them anyways because it just lowers the quality of the product. :)
 

^eMpTy^

2[H]4U
Joined
Jul 21, 2004
Messages
3,233
eXzite said:
The problem is that the dependant texture used for the specular lookup is a 256x1 texture, and when you do aniso taps on that, you thrash your texture cache. IF you use Doom3's aniso setting in the doom3 control panel, the game knows not to do aniso on a dependant texture lookup, and therefore you get little perf benefit. If, however, you force aniso on everything either from the windows control panel or console, then you get a jump in perf because aniso is now on every single texture (this dependant one included), and you were doing multiple taps on an 256x1 texture before, thrashing your texture cache, where as now you are doing the standard specular calculation with a math operation.
Ok so you only get the boost going from forced AF to in-game AF? So which was used in the [H] D3 benchmarks?
 

^eMpTy^

2[H]4U
Joined
Jul 21, 2004
Messages
3,233
burningrave101 said:
Thats actually a good thing. Companies are better off not having people like yourself buy from them anyways because it just gives them a bad name.
guys...seriously...is this necessary?
 

burningrave101

[H]F Junkie
Joined
Sep 9, 2003
Messages
11,825
^eMpTy^ said:
Ok so you only get the boost going from forced AF to in-game AF? So which was used in the [H] D3 benchmarks?
In-game AF was used. 8xAF is automatically enabled when you switch to High Quality mode.

ATI hardware also has a little lower performance through in-game AF but it is a little higher quality. Enabling it through the CP will give a little more performance but lower IQ.
 

BoogerBomb

Supreme [H]ardness
Joined
Jan 10, 2003
Messages
6,470
Sorry I only mentioned that to prove the point that my company does not own me even in my personal time if I should choose to do something as a fan and gamer that helps other gamers/fans out as well. Burninggrave has made up his mind and nothing will change it and I apologize for trying to get him to see our point of view that Humus did not do this as an ATI employee.
 

burningrave101

[H]F Junkie
Joined
Sep 9, 2003
Messages
11,825
R1ckCa1n said:
That could have been why I asked for AF.......

I'll take a 16X AF at the price of 8X everyday; or better yet a free 16X.
There is almost no performance difference between 8xAF and 16xAF on 6800's or X800's ither one.

The AF performance hits on the 6800's are at 4xAF primarily and another slighter hit at 8xAF.

The X800's lose very little performance from 2xAF-16xAF because they are more optimized using brilinear filtering. This also means the AF quality is slightly lower then the 6800's.
 

BoogerBomb

Supreme [H]ardness
Joined
Jan 10, 2003
Messages
6,470
burningrave101 said:
In-game AF was used. 8xAF is automatically enabled when you switch to High Quality mode.

ATI hardware also has a little lower performance through in-game AF but it is a little higher quality. Enabling it through the CP will give a little more performance but lower IQ.
What if you set the CP AF to 8X? Which will be used the in-game method or the drivers method?
 

eXzite

Limp Gawd
Joined
Jul 22, 2004
Messages
144
^eMpTy^ said:
Ok so you only get the boost going from forced AF to in-game AF? So which was used in the [H] D3 benchmarks?
No you get the 'real' boost from starting with forced AF ( which doesn't really make a difference anywhere else besides this shader ), and then changing the shader code to the math-op. If you change the shader code while using in-game, through normal settings panel AF (also turned on by high quality), there really isn't much of a boost at all, because AF is disabled for this dependant texture lookup to start with. On ATI hardware only, however, there's still a small jump in this case, just because the math-op is a slightly faster op on ATI's shader hw. This boost isn't as big as the 'real' boost everyone's so excited about however, and is usually like 2 or 3 %. So, the boost really affects people who forced AF on to start with the most.
 

Moofasa~

[H]ard|Gawd
Joined
Oct 13, 2003
Messages
1,571
burningrave101 said:
There is almost no performance difference between 8xAF and 16xAF on 6800's or X800's ither one.

The AF performance hits on the 6800's are at 4xAF primarily and another slighter hit at 8xAF.

The X800's lose very little performance from 2xAF-16xAF because they are more optimized using brilinear filtering. This also means the AF quality is slightly lower then the 6800's.
Show me images where it lowers quality...
 

R1ckCa1n

Supreme [H]ardness
Joined
Jul 22, 2004
Messages
4,355
burningrave101 said:
In-game AF was used. 8xAF is automatically enabled when you switch to High Quality mode.

ATI hardware also has a little lower performance through in-game AF but it is a little higher quality. Enabling it through the CP will give a little more performance but lower IQ.
Do you have problems understanding what you read? If you would take your hat off and read the B3D thread, you will soon understand it comes down to the hardware. ATI's cards where not optomized for the way Doom3 renders. The NV40 was made to run best on Doom3, which it does, where ATI is made to run all games consistant. What those guys at B3D figured out is how to tweak the shader to run it with more math, ATI's strong point. It's pretty much as simple as that. Once they figure out how to make the shaders mathmatically correct, it will render just as the artist entended it too.

Please don't make this into more than it really is, like some people.
 

burningrave101

[H]F Junkie
Joined
Sep 9, 2003
Messages
11,825
BoogerBomb said:
Sorry I only mentioned that to prove the point that my company does not own me even in my personal time if I should choose to do something as a fan and gamer that helps other gamers/fans out as well. Burninggrave has made up his mind and nothing will change it and I apologize for trying to get him to see our point of view that Humus did not do this as an ATI employee.
How can you say that Humus didn't do this as an ATI employee? lol

You have no idea at all whether or not Humus was influenced by ATI or not. And since it DOES work for ATI and since this DOES affect gaming performance for one of the biggest PC games in history, its hard to say what his reasoning was for certain.

It doesn't really matter though. If it gives you a performance increase and you dont notice IQ issues then good for you.

I just dont agree with the way Humus does things and i'm free to have that opinion and express it in any way i see fit.
 

^eMpTy^

2[H]4U
Joined
Jul 21, 2004
Messages
3,233
BoogerBomb said:
Sorry I only mentioned that to prove the point that my company does not own me even in my personal time if I should choose to do something as a fan and gamer that helps other gamers/fans out as well. Burninggrave has made up his mind and nothing will change it and I apologize for trying to get him to see our point of view that Humus did not do this as an ATI employee.
I think Kyle said it best:

Well I think it all depends on what arena you are standing in and what is being commented on. Let's say that SuperRob works for a company called "SuckitBitch Toilets," and decides to post an IQ hack for DOOM 3. I don't think your affiliation with SBT is going to be called into question. However, if you are suggesting mods for the latest low flow shitters that impact their ability to reduce skidder marks, you might be seen as a rep for your company.
 

R1ckCa1n

Supreme [H]ardness
Joined
Jul 22, 2004
Messages
4,355
burningrave101 said:
You have no idea at all whether or not Humus was influenced by ATI or not. And since it DOES work for ATI and since this DOES affect gaming performance for one of the biggest PC games in history, its hard to say what his reasoning was for certain.
Did you skip over his post saying how upset he was with the team responsible for id's relationship? Looks like ATI blew it by not finding this simple solution until after the game shipped.
 

Xtasy0

[H]ard|Gawd
Joined
Aug 25, 2002
Messages
1,782
Caffeinated said:
John Carmack Responds to the ATI Humas tweek

--------------------------------------------------------------------------------

http://www.beyond3d.com/forum/viewt...r=asc&start=220

I got a response from Carmack:

Quote:
Our specular function isn't a POW function, it matches the bias / scale calculations done on the NV10/NV20/R200 paths.

This of course explains why it doesn't look exactly the same. This is also a good thing. Now we don't even need to use a POW function, but we'll be fine with a MAD_SAT, which make it even faster. Got 21% boost in the timedemo with this. Now I still don't know exactly what exponent he's approximating, or if he's using different lookup tables for different materials, but I've tried this and it seems to look the same, but our fellow artifact hunters may want to verify that.

Code:
MAD_SAT R1, specular, 4.0, -3.0;

If that causes artifacts for someone, maybe approximating a lower exponent will do it:

Code:
MAD_SAT R1, specular, 3.0, -2.0;

*******************************************************************

He mentioned artifacting in the first post and made a reference to how to possibly fix it, and then you say that mentioning artifacting was unnecessary?

Here's your token sir, the clue bus is waiting for you at the stop...
he was directly quoting and replying to the user "@trapine", @trapine's post mentioned nothing about artifacting and his question was nothing that the batman was in a position to answer.
 

eXzite

Limp Gawd
Joined
Jul 22, 2004
Messages
144
R1ckCa1n said:
Do you have problems understanding what you read? If you would take your hat off and read the B3D thread, you will soon understand it comes down to the hardware. ATI's cards where not optomized for the way Doom3 renders. The NV40 was made to run best on Doom3, which it does, where ATI is made to run all games consistant. What those guys at B3D figured out is how to tweak the shader to run it with more math, ATI's strong point. It's pretty much as simple as that. Once they figure out how to make the shaders mathmatically correct, it will render just as the artist entended it too.

Please don't make this into more than it really is, like some people.
The thing is, Carmack wanted this dependant texture lookup in there to have control over the calculation. He wanted to have the game look the same on all rendering paths, and it wouldn't have if a math operation was used in each case. Therefore, they will never get it looking 'exactly' the same as before the change with just the math, but hey, it's pretty close.
 

^eMpTy^

2[H]4U
Joined
Jul 21, 2004
Messages
3,233
R1ckCa1n said:
Do you have problems understanding what you read? If you would take your hat off and read the B3D thread, you will soon understand it comes down to the hardware. ATI's cards where not optomized for the way Doom3 renders. The NV40 was made to run best on Doom3, which it does, where ATI is made to run all games consistant. What those guys at B3D figured out is how to tweak the shader to run it with more math, ATI's strong point. It's pretty much as simple as that. Once they figure out how to make the shaders mathmatically correct, it will render just as the artist entended it too.

Please don't make this into more than it really is, like some people.
Actually I read the thread and it's not at all like you're stating it. The problem isn't ATi being "better at math"...the problem is the x800s go all silly doing this texture lookup and do a bunch of anistropic filtering they aren't supposed to.

This isn't about fixing D3 to work better on ATi, this is about hacking D3 to work around a bug in ATi cards. It seems clear this will be fixed in a new driver release...
 
Status
Not open for further replies.
Top