ATI's Initial Responce to NVIDIA's Editors Day IQ Issues

[Arun]

Well, the phrase 'questionable optimisations' is not all that far away from saying cheating, there isn't really a huge distinction between the two in my mind beyond arguing the semantics.

I know this was in the beggining of the thread and we're at page 2 now but...
The correct semantic is, as used internally by NVIDIA for their "questionable optimisations", is: optimizations, that some have argued are overly-aggressive.

Damn, I *love* that 44.67 internal memo

---

Regarding that ATI statement...
Couldn't agree more with them on most points. And saying I thought NVIDIA couldn't become any lamer...



Uttar

So, Uttar...how's that editorial of yours coming along 8)

You guessed right, that memo is included in the editorial.
Still no response from "the one" AKA Voldemort. That means it could take either a few days, or many years
Evantually, I'm considering proof reading it myself and redoing some stuff BEFORE he gets it, so his job can be made more quickly - although I'm sure that even if I wrote something I'd consider "perfect", he'd manage to find a billion ways to make it better


Uttar

 

[Randell]

you gotta love the fanboys

nVidiot - nVidia didnt accuse anyone of cheating - I dare you to find where they said cheating..

FanATIc - even if the sites recognise the hypocrisy, I'm not forgiving them for their biased reporting earlier..

 

[micron]

Lars was answering questions at the THG forums...here's his post about ATi's possible optimizations..

Well, Refrast says the ATI image is too dark. The AM3 techdoc shows it in the way it´s on NV cards. Both, the massive overdraw test and the particles test work by blending alpha textures over each other. This is where the final brightness comes from. I did not finaly say that it is a cheat and that it´s possible that it´s a bug. But if it´s a bug, how can it be overseen by the driver devs?

Aside of many other "big sites" I´m talking about the 3DM issue and NVIDIA since a long time this year. I look very close on the NV drivers. But I also look very close to all others. ATI is very popular at the moment but does that mean that they´re untouchable. If Gabe talks about NV issues at an ATI event it´s OK to write about it. But if Mark Rein and Randy Pitchford speak about issues at ATI it´s not OK? And if I do then I´m biased? Did logic turn around into something new during the past months!?

I think we´ll know pretty soon what´s behind this thing in AM3. If it´s a bug then it will be fixed and in the end there´s something good coming out of it - like with the AM3 article showing NV driver issues (those lighting bugs were in the drivers from 44.03 to 51.75.... sometimes one has to write about to bring it to attention...)

Lars

http://www.community.tomshardware.c...517859&page=0&view=collapsed&sb=5&part=2&vc=1

 

[CorwinB]

This is where the final brightness comes from. I did not finaly say that it is a cheat and that it´s possible that it´s a bug. But if it´s a bug, how can it be overseen by the driver devs?

Sure. It could be a cheat that brings 0% performance increase ! Those ATI bastichs are so happy to cheat that they do it even when it doesn't make sense.

And we've seen much stranger bugs overseen by driver devs, like the static clip plane generating bug, for instance.

 

[micron]

It could be a cheat that brings 0% performance increase !

 

[Borsti]

There are a few things to say here.

Mark and Randy made comments on "issues" they have with the ATI drivers at a NVIDIA event, without going into details. They did it during the NV presentation from the background of the room (not during their presentations). When I talked about Gabes comments at an ATI event I see no reason not to talk about it this time. I make clear that I did not find anything in my own comparisons. But those are still comments from game developers.

Regarding AM3. There is potential to "cheat" at alpha blended textures. That´s what I say. I also say that it can also be a bug in the ATI drivers. It´s not my fault if people only watch the screenshots without reading the text and cry in some forums. Take a look into our AM3/NV drivers issues article to see how I handle real strange things. The ATI AM3 thing is one part of the NV38/36 article. The findings are correct. Even one week after the article ATI still does not know why this happens in AM3. They are sure that it´s a driver thing but they yet don´t know why it happens. They also confirm (to me) that it´s not a simple gamma thing. It might be a gamma issue but not an "overall scene gamma". People play with curves in Photoshop just to proove that you can change the look of an image. That´s groundbreaking news....

People liked the AM3 article because it showed incorrect rendering on NV cards. Now people cry that I write about incorrect renderings of ATI drivers. I can´t understand this. I did not call the results of ATI in AM3 incorrect as I did with the NV45.23/51.75 drivers in the AM3 article. I also did not use grey bars for ATI in the AM3 scores of the NV38/36 article.

It´s a fact that ATI is optimizing their drivers for applications (I say optimize - not cheat). It´s always possible that an optimization has impact on another application. ATI also explains to me that they have internal guidelines for optimizations and they now think about making them public.

Lars

 

[Mark]

I think the main problem people have is that it took you guys months to say anything about the stuff nVidia was doing (even with lots of obvious and clear evidence from many independant sources) but you almost instantly write up a pretty damning article about some stuff ATI may or may not be doing, which may or may not be a bug, of which you got most of the information from their chief competitor. I don't know about anyone else, but some journalistic integrity would be nice.

 

[THe_KELRaTH]

RE: Mark Rain - UT2003
I don't really see how anyone can say that there's a bug in the ATI drivers when so far no one has confirmed whether changing the DetailTexMipBias command settings in the UT2003.ini effects the texture detail quality on an NV FX card.

I've asked this a number of times on a number of forums and still have had no reply to it yet we know that other UT2003 requests are ignored.
I'm sure I don't need to say what it would mean if this were the case.

 

[Dave Baumann]

Lars - Did Gabe ever actually say "NVIDIA did this"? IIRC, I recall someone asking and his reply was "Well, I'm here at an ATI event...", but I don't ever reacll him saying NVIDIA driectly. However the issues that he spoke of had all been confirmed and verified by independant parties prior to his statements, he was just reiterating them - the only one that was questionalable was the "altering IQ in screenshots" but we know this is happeneing to a degree to capture FSAA images.

There is also a difference in how it was handled with all the sites - the next day NVIDIA released their statement which went into everyones articles, however here you haven't even given ATI the opportunity to make any statements before publishing.

Given Tom's little editorial in your 9800XT review just a few weeks before, this turning up in the 5950 review seems a little ironic dosn't it?

 

[CorwinB]

I make clear that I did not find anything in my own comparisons. But those are still comments from game developers.

Then surely you should say that what those game developers are saying is BS ? Because that's what your boss said in his little blurb regarding Gabe Newell, although the press was able to verify his data first hand...

That's a nice example of getting a point across ("ATI is cheating" in this case) while looking clean.

It´s not my fault if people only watch the screenshots without reading the text and cry in some forums.

That's the same trick Anand and his trusted sidekick Evan Lieb were trying to pull when they did the "we heard good HL2 performance numbers, take with a grain of salt" NV PR-stunt thingie.

As a matter or fact, you are responsible for what is done with your work. You know how the Internet work, and how rumors and quotes taken out of context fly on this medium. And if you don't, the nice people at Nvidia PR quarters giving you all those damning smoking guns certainly know how it works, and they are taking advantage of it, and of you.

Regarding journalistic integrity, at least the UT2003 part was totally debunked here (it's a game setting in the config file, and on ATI's part it's a 0.4% "cheat"), and of course no mention of it made its way to your article. What else could we hope for... I'm sure that following a pattern established by Kyle and Anand, you will post a retractation at the bottom end of a forum (where you can link to to show your "good faith"), with the unfounded accusations and suspicions staying the same in your article... I'm really in favor of passing laws forcing all media outlets (press, TV, Internet sites) to issue retractations at exactly the same place and for a longer time than the article that needed those retractations.

 

[nelg]

Lars, as Dave and Ratchet asked....
Why Is There a Double Standard.

 

[Bambers]

It´s a fact that ATI is optimizing their drivers for applications (I say optimize - not cheat). It´s always possible that an optimization has impact on another application. ATI also explains to me that they have internal guidelines for optimizations and they now think about making them public.

doesn't seem likely that the am3 alpha one is app sepcific. it happens even in much older drivers and on firingsquads IQ test recently the planes smoke trails in 3dmark03 were also darker on atis image than nvidia. Would be nice if someone can test this but it seems to me that this probably affects all d3d and maybe ogl ones too.

 

[Fred da Roza]

I think the main problem people have is that it took you guys months to say anything about the stuff nVidia was doing (even with lots of obvious and clear evidence from many independant sources) but you almost instantly write up a pretty damning article about some stuff ATI may or may not be doing, which may or may not be a bug, of which you got most of the information from their chief competitor. I don't know about anyone else, but some journalistic integrity would be nice.

A few months ago I would have, and did, argue differently. However with the last few rants from THG, they clearly have taken sides. I have to agree. THG's expeditious exposé of ATI’s issues just illustrates a lack of journalistic integrity in light of their tame and tardy response to nVidia’s “optimizationsâ€￾ in the face of the overwhelming evidence. What Tom so condescendingly chastised Kyle and Van for on Quack and Bapco, respectively, is what Tom and Lars clearly did with respect to their comments about ATI in their latest articles. What’s worse is that, at the very least, 2 of the accusations (UT2003 and Halo) are extremely thin on evidence. I guess Tom can join the club of those he condemns for showing a lack of integrity.

 

[Xmas]

RE: Mark Rain - UT2003
I don't really see how anyone can say that there's a bug in the ATI drivers when so far no one has confirmed whether changing the DetailTexMipBias command settings in the UT2003.ini effects the texture detail quality on an NV FX card.

I don't know how NVidia cards behave, and I can't test it currently.

However, I can say for sure that R300 based ATI products don't handle a positive LOD bias correctly when doing AF, neither in OpenGL nor in D3D. Whether this is a hardware or a software issue I don't know.

 

[THe_KELRaTH]

RE: Mark Rain - UT2003
I don't really see how anyone can say that there's a bug in the ATI drivers when so far no one has confirmed whether changing the DetailTexMipBias command settings in the UT2003.ini effects the texture detail quality on an NV FX card.

I don't know how NVidia cards behave, and I can't test it currently.

However, I can say for sure that R300 based ATI products don't handle a positive LOD bias correctly when doing AF, neither in OpenGL nor in D3D. Whether this is a hardware or a software issue I don't know.

Hi Xmas, What results would you expect if ATI's Lod Bias were working correctly?

From what I can tell setting Max Quality mipmap in the driver means the DetailTexMipBias has only one available setting of 0.40000. A low figure has no effect and 0.5 seems to drop mipmap to Performance level in drivers.
Setting mipmap quality to Performance in drivers then changing the DetailTexMipBias from 0.00000 to -0.30000 is the equivalent of changing the 4 steps in the driver so -0.30000 = Quality again.

The reason I'm interested to know if an FX card is effected by changing the DetailTexMipBias command is to confirm that the command is actually working properly before suggesting an ATI bug.

 

[sireric]

It´s a fact that ATI is optimizing their drivers for applications (I say optimize - not cheat). It´s always possible that an optimization has impact on another application. ATI also explains to me that they have internal guidelines for optimizations and they now think about making them public.

Lars

Lars, are you saying that ATI is doing app detection for its "optimizations"? If so, that's BS. Yes, we use applications to help us optimize (we need to optimize for things), but they are general purpose optimizations, not specific to any applications. Any application using a similar rendering path would get all benefits.

Just want to clear that up.

"No comment" on anything else.

 

[Xmas]

Hi Xmas, What results would you expect if ATI's Lod Bias were working correctly?

From what I can tell setting Max Quality mipmap in the driver means the DetailTexMipBias has only one available setting of 0.40000. A low figure has no effect and 0.5 seems to drop mipmap to Performance level in drivers.
Setting mipmap quality to Performance in drivers then changing the DetailTexMipBias from 0.00000 to -0.30000 is the equivalent of changing the 4 steps in the driver so -0.30000 = Quality again.

The reason I'm interested to know if an FX card is effected by changing the DetailTexMipBias command is to confirm that the command is actually working properly before suggesting an ATI bug.

I am not talking specifically about UT2k3. In fact I don't even have it.

When using a positive LOD bias in combination with AF on a R300 based card, it will behave as if you had set the minimum mip level. If you disable AF, it will work correctly. Follow the link in my sig and try it.

It seems like AF will clamp the LOD value to [0, ?] and only after that the LOD bias is applied. Therefore the LOD value can never be smaller than the LOD bias. Which is not what you'd expect from LOD bias.

 

[Doomtrooper]

Lars, are you saying that ATI is doing app detection for its "optimizations"? If so, that's BS. Yes, we use applications to help us optimize (we need to optimize for things), but they are general purpose optimizations, not specific to any applications. Any application using a similar rendering path would get all benefits.

Just want to clear that up.

"No comment" on anything else.

Thats is exactly the kind of comments we need from you, and other ATI representetives..I know Opengl_guy doesn't pull punches either.
One must be aware that Lars is part of the NVDC team...i.e Nvidia Damage Control Team

I only wish PR people would listen to the community, as obviousally the content that spews from some of the big 3 review sites are just 'out out lunch'.. obvious by the comments, obvious by their previous review content..like recomending a 5800 over a 9700 Pro, who in their right mind would do such things

ATI must play hardball, much like Nvidia does to sites that don't want to play 'the game'... i.e

http://www.beyond3d.com/forum/viewtopic.php?t=8744

 

[Doomtrooper]

It seems like AF will clamp the LOD value to [0, ?] and only after that the LOD bias is applied. Therefore the LOD value can never be smaller than the LOD bias. Which is not what you'd expect from LOD bias.

I have no interest in using postive LOD, if it was negetive I would be worried.

 

[Vince]

One must be aware that Lars is part of the NVDC team...i.e Nvidia Damage Control Team

This is a joke right? I just want to make sure as I'm having trouble believing your this far removed from sanity. Ohh, and the ATI pep-talk was charming, mix it in with a Bush-esque, "This world has only good and evil" comment and it's gold.