OMG HARDOCP REVIEW OF UT2003 AND FILTERING

[digitalwanderer]

Lemme guess, Kyle didn't think quincux looked blurry either.... :bleh:

 

[WaltC]

Heh-Heh....Always nice to encounter folks more long winded than me, Dave H... There's too much for me to go about this point by point, so I'll just say I stick with my previous post to you on the subject and address a couple of things...

.....


Guess this is an agree to disagree kind of thing. But honestly, aren't you the least bit intrigued that they managed to find such a significant performance increase on the table without noticeably affecting output quality? If not, I'd suggest that you're more interested in the graphics market as morality play than for its technological content. (Not that there's anything wrong with that.)

I think you're for some reason missing the obvious... It was because people noticed a difference that they first started looking into whether or not the Dets were doing Trilinear in UT2K3. Your position seems to effectively be that they are doing trilinear filtering without doing trilinear filtering. They aren't, and the differences are noticeable, which is why this topic has come up. If there were no visible differences, the topic would most likely never have been raised in the first place, right?

You're reading way too much into [H]'s pronouncements that they couldn't tell much difference. As well, not even [H] says they can't tell any difference. As I said, I have no problem with nVidia offering a performance trilinear which is a mixture of tri and bilinear....none whatsoever. I understand what that is and am not curious about it (what's there to be curious about?) My problem is with the fact that they *substitute* it for full trilinear, even when the application asks for full trilinear. Very simple.

The reason you are "curious" is because nVidia, in this game, has substituted this performance bi/trilinear mode for full trilinear, and [H] editorializes that the substitution is A-OK with them because they "can't tell much difference" in the resulting IQ, although as I said even they don't deny differences exist. For some reason, you've concluded that this is a "great thing" having significant import. As pointed out by about everyone, the significant import is that in UT2K3 these Detonator users cannot get full trilinear filtering at all in the game. That's what I find significant and if I'm curious about anything it's about nVidia explaining why, in this one game, it substitutes its performance trilinear for its full trilinear without saying so much as a single word about it. Aren't you curious as to why initial information on this subject came from neither nVidia or [H]? nVidia still has yet to comment on it and [H] is only reacting to previous publicity generated by B3d. [H] at no point ever denies that nVidia is not doing full trilinear filtering in UT2K3--in fact, their entire article affirms and confirms it... Their only contribution otherwise is to state that they don't care, for one reason or another.

These sorts of hacks have always existed, are used by Nvidia, ATI and every other IHV [Q: then why doesn't Unwinder's anti-detect impact ATI performance on more games? A: we have no proof it is detecting all game-specific optimizations in either ATI's or Nvidia's drivers], and are 100% commonly accepted practice so long as output conformance is not broken. Here we have a situation where the output is not conformant, but the difference is apparently not subjectively noticeable to the end-user. But this is far from the only example of that sort of optimization either.

Missing the obvious here Dave H, again... First of all there are no similar driver hacks in the current Catalysts, are there? All one need do is to use the UT2K3.ini to turn on trilinear--and presto, the game is fully trilineared. Can't do that with the Dets, though. So let's dispense with lumping all of the IHV's together in a big pot and stirring it with the same stick. There are many, many obvious differences between the IHVs aside from the products they produce.

Did you write the Anti-Detector software, DaveH? The guy who wrote the software claims it does the same thing for the Catalysts and the Dets. Argue with him if you like...

Again you keep reaching an erroneous conclusion that the "output isn't subjectively noticeable"--of course that's simply not so. Were it so, no one would ever have been able to notice the difference, hence none of us would be talking about it right now.

You seem to have a problem with "almost as good" and "as good"....there is a distinct difference between these two states. Nobody, not even [H], claims that it's "as good." The entire [H] piece revolves around their subjective opinion that it's "almost as good" as far as they can see and where it isn't "as good" they flatly state they don't care. That's what subjective opinions do for you...

All those other optimizations are uncontroversial or at least unremarked upon. I'm not saying this sort of thing shouldn't be looked into. In fact that they should be looked into--and that's exactly what [H] has done. They've found that in this case the optimization doesn't impact subjective output quality, which, by the standards of their reviews, is all that matters.

Nope, that is not what the [H] article states at all. They've said that what differences they can see, and they do see them, are so "minor" in their opinions (because "all you do in the game is run around fragging people and so don't have time to look at the image quality" if I got that right) that they just don't care whether nVidia provides full trilinear like the Cats do or not. That's very, very different from your characterization of what they said.

Has it ever struck you that their article is so subjective it's worthless? Listen, opinions abound about IQ. Whereas I run with 2x FSAA enabled, 16x AF in my games by default--some people state they prefer 0x FSAA/8xAF for their own reasons. Which of us is "right?" Correct answer is "neither" because it's a matter of subjective preference only.

So.....what [H] views as "minor" IQ degradation because of the lack of full trilinear support in this game, someone else might view as a "major" IQ difference. The important point about this whole affair is this: nVidia has removed the option of full trilinear support from its drivers for UT2K3. Nothing else matters--at all. Had they not done this, there would be no issue whatever as no one would care what lesser IQ modes nVidia built into its driver support. There is no issue apart from this one in my view and as such [H]'s entire attempt at apology is a waste of epaper.

They're also calling on Nvidia to make this optimization selectable in the drivers, and indicated that Nvidia will be doing exactly that. The fact that it wasn't is bad on Nvidia's part, and [H] has criticized Nvidia for that.

They can "call on" nVidia all they like but until nVidia *does something* relative to the issue such statments are pompous and mean nothing, right? Heh...would have been nice if [H] had ONCE "called on" nVidia to stop cheating in its drivers relative to benchmarks...! What did [H] do instead? Tell everyone to dump their benchmarks, that's what [H] did... That's pretty funny, Dave H....

Until nVidia does restore full trilinear to its driver support for UT2K3 nothing anybody "says" means a bloomin' thing...

The only thing they haven't done is the only thing that would satisfy you, namely disqualify all Nvidia products from consideration because Nvidia has engaged in slimy behavior.

Don't know where this comes from....*chuckle* I can only count my lucky stars that I'm not a nVidia apologist. Sigh--what would satisfy me is simply [H] stopping its infantile behavior of apologizing for nVidia and plainly stating that you can't compare nVidia's faux-trilinear to the Catalysts' full trilinear in terms of performance because its not an apples-apples IQ comparison. A subjective opinion that something is "almost as good" for reasons I've already stated doesn't suffice, no.

That's the problem with viewing the realtime graphics ASIC industry as a morality play. It may be fun for a while. But the actors would rather view it as reality, and thus they're inevitably going to disappoint you.

Then your suggestion would be that we view it as an "immorality play" and that "anything goes" and we throw veracity out of the window? Excuse me--I don't want to personally verify the old saying that "a fool and his money are easily parted"....

 

[Dio]

I think Phineas T. Barnum's more appropriate quote would be "Never give a sucker an even break."

 

[nelg]

BTW Dave H. Kyle liked your post so much that he made a thread about it.
http://www.hardforum.com/showthread.php?s=b265cc63d4ed44bb6297dff3dd64e732&threadid=644239

 

[Doomtrooper]

Kyle just doesn't get it(does but doesn't want to lose some $$), and contradicts what their own reviewer discovered here from Brent, sad.


http://www.beyond3d.com/forum/viewtopic.php?t=4772&highlight=application

I'm sorry the story changes whever they like, and Nvidia IS dropping down to performance filtering (the old balanced Default Mode).

 

[Doomtrooper]

I would like to add the only way this stops is IHVs need to start policing these review sites. Anand had false information on his 5900 review for months, and still does.
Now Brent says he can't see any difference between the Peformance (Balanced) and Quality (Application) filtering. The lies have got to stop, yet every new release these sites get the cards 1st, even though posting garbage.

Hundreds of thousands of people read these, and if they are inaccurate impact sales. Doesn't mean the review has to be all roses, but ACCURATE.

 

[Mariner]

BTW Dave H. Kyle liked your post so much that he made a thread about it.
http://www.hardforum.com/showthread.php?s=b265cc63d4ed44bb6297dff3dd64e732&threadid=644239

Interesting.

Dave is not part of community here as he does not spend time here or post here. He only comes here to push his own agenda. He has his own forums for that.

Hands up everyone who thinks Dave should block Kyle's IP! Pot, kettle black.

 

[Dave H]

BTW Dave H. Kyle liked your post so much that he made a thread about it.
http://www.hardforum.com/showthread.php?s=b265cc63d4ed44bb6297dff3dd64e732&threadid=644239

Heh. Well, at least for the first time in my posting career I've got a .sig...

 

[WaltC]

BTW Dave H. Kyle liked your post so much that he made a thread about it.
http://www.hardforum.com/showthread.php?s=b265cc63d4ed44bb6297dff3dd64e732&threadid=644239

Of course, not surprising that he didn't provide a *link* to the B3d thread in which the post was made....*chuckle* Wouldn't do to have his audience corrupted by the alternative points of view in the thread they'd be exposed to by following the link and seeing the context, now would it?....

What a soap opera this is...

Edit: Ah, I see that "Lab Rat" thought to provide a link to the entire thread....

 

[Bouncing Zabaglione Bros.]

I suggest that B3D/ Dave H contact Kyle and let him know he is in breach of copyright by reposting selected parts of someone elses thread verbatim, and ask him to remove the post.

 

[Myrmecophagavir]

Again you keep reaching an erroneous conclusion that the "output isn't subjectively noticeable"--of course that's simply not so. Were it so, no one would ever have been able to notice the difference, hence none of us would be talking about it right now.

Might that not be only because of the mipmap level colouring option? Where was the first time this was noticed and what were the circumstances?

 

[Pete]

This was Kyle's response:

Dave is not part of community here as he does not spend time here or post here. He only comes here to push his own agenda. He has his own forums for that.

I have given him my phone number and invited him to call should he wish to discuss it. He has not done that.

Kyle forgot to add "exclusive" to describe his community. Assuming his current banning binge is not based on the banned's poor behavior (I didn't see any in the [H] threads I read, but I haven't read them all), it would seem anyone who doesn't fall in line with his opinion is pushing his own agenda, thus not useful. That sounds like suppressing free, and potentially enlightening, conversation. In other words, it sounds like he has something to hide or protect, and that isn't the Truth.

 

[WaltC]

This question was asked on the Hardforums:

I'm just wondering...
Why you guys banned Dave Baumann's IP from the [H]ardForums? Isn't it a bit childish to ban the owner of one of the most respected hardware sites out there over an argument?

This was Kyle's response:

Dave is not part of community here as he does not spend time here or post here. He only comes here to push his own agenda. He has his own forums for that.

I have given him my phone number and invited him to call should he wish to discuss it. He has not done that.

At least he finally admitted to banning him...

I am just amazed....Every time I think I've heard it all, then I read something like this....Incredible...

What century does Kyle inhabit? Has he heard of email? Is there some reason Dave B. should have to pick up the phone and call Kyle...? Heh...Wavey, I think you should email frgmstr (or pm him in his account here at B3d that apparently you haven't rescinded as of yet) and suggest that you would be glad to pick up the phone and call him collect and bend his ear for about an hour.... Maybe you could arrange to keep his line open for about 24 hours and run up a $500 collect call bill. Perhaps that would shock him into the 21st century. Imagine that--a webmaster suggesting that there's a better way to communicate something like this than emailing (or pm'ing through the B3d forums.) What a ridiculous suggestion--"I asked him to call me." Huh? Just couldn't "hash it out" over the Internet, right Kyle? Good grief--preposterous....

Ok, let's rehash...Kyle is invited on numerous occasions to take part in a discussion (any one of them) of his ideas within the B3d forums (I posted such an invitation more than once)--which he declines to do, even once. Dave B. visits [H] and weighs in with information and is banned from the site and accused of having an "agenda" by Kyle. Amazing contrast....truly remarkable. Unreal.

Well, I suppose this is Kyle's way of admitting that he himself follows an "agenda," and so its natural for him to suspect everybody else of the same thing. Wow. Right, Kyle--it's not the information people have that's important--it's the agendas that underlie that information, which of course you telepathically divine (since Dave B. has never done anything I might think of as following an "agenda".) Uh, in much the same way you were able to divine ET's unspoken motives in your "Two days after the D3 preview..." character assassination story which impugned motives which you never attempted to prove existed. (Even funnier, today I read D3 won't ship until 2004! So much for the relevancy of the vaunted "Doom 3 Preview"...)

The question I have is why some people become so threatened by simple information--which is neither good or bad in itself...? What is it they are afraid of, and why? If I wanted to make a pessimistic wager I might say that Kyle is scared to death of what he imagines to be "his community" defecting to another one. More's the pity...the Internet itself is a big community and the people in it are beholding to no man--or web site. What--Kyle imagines that the people who contribute to [H] forums never read and contribute anywhere else? Very difficult nut to crack here...

 

[swanlee]

I think it's pretty simple Kyle is a paid shill. He's been targeted by Nvidia because of the following he has built up, that following is his life blood because if that falters his checks from Nvidia will stop. Nvidia doesn't care one flip about Kyle all they care about is the fact they can speak through him and use him without reflecting anything on the nvidia PR department. Kyle is trying desperately to try and maintain his audience but he has freakin paniced and has completely lost his mind.

Like any great dictator near the end of his run every day that passes by means he has to resort to more and more outlandish measures to try and maintain his control.

His kingdom is crumbling around him and at the rate he is banning people from his forums pretty soon he will only be talking to himself there.

 

[WaltC]

Might that not be only because of the mipmap level colouring option? Where was the first time this was noticed and what were the circumstances?

That might be a valid point of view, except for a couple of things:

1) The color-code filtering option is built into the UT2K3 engine for the express purpose of allowing people to look at filtering that may or may not be obvious in a certain scene. It is the same method used in Q3, I believe, for the same purpose. These tools have routinely been used by hardware reviewers for a long time to illustrate differences in filtering techniques.

2) What often happens is that people who use these tools then go back to the scenes and view them again and notice things they didn't see before. It's a tool designed specifically for the purpose of evaluating filtering. That's all it is.

What's being implied is that somebody did something "wrong" by using the engine's color-code filtering tool to check on the quality and type of filtering used in the game by different hardware and different driver sets. Of course, that's nonsense. It's standard hardware-review fare and has been done routinely by a variety of hardware review sites over the years--including [H.]

[H] has completely mischaraterized the issue in its zeal to apologize for seemingly anything nVidia does these days. Nobody's criticizing nVidia's performance trilinear filtering option--I'm certainly not. What I'm criticizing, and what I think other people object to, is the fact that nVidia told no one that it was substituting this filtering mode in place of full trilinear in UT2K3, even when the application instructs the drivers to provide full trilnear support. In another UT2k3-engine game, Unreal 2, nVidia's drivers provide full trilinear when asked to do so. So this is something nVidia's not even doing for a specific game engine--but for a specific game.

Now, if nVidia saw it as a "feature" and an "advantage"--why not publicize it ahead of time--or make it available as a driver-selectable option....? Why not make some "hay" on it, if nVidia thinks it's such a great thing? That's the key question, I think.

The answer I think follows a predictible pattern: knowing that many hardware review sites use UT2K3 and its associated timedemos for hardware benchmark comparisons, nVidia gambled on slipping this change in for a performance boost and hoping no one would ever be the wiser. That's IMO, of course.

So to answer your question as to whether it makes a difference whether the filtering tool first discovered the problem--I think not. If that was the case than we'd never have needed a color-coded filtering tool in the past, right? The fact is that differences in filtering techniques, depending on a variety of in-scene factors, are not always immediately evident. This what what the color-code approach was implemented to detect *years* ago--making it easy to spot the differences between bilinear and trilinear filtering, etc. It's the height of hypocrisy for a hardware review site to suddenly pipe up and declare that tools such as these are "nerdy and useless" simply because its favorite IHV got caught misrepresenting something again. Were nVidia and its dupes able to "get away" with this kind of thing it would set the 3D industry back several years, IMO.

 

[whql]

I was under the impression Kyle didn't like "Leeching" off of other sites for fear of copyright issues. Perhaps he meant that he only doesn't agree to leeching when it disagree's with his agenda, but its fine to leech when it agree's...

 

[Bolloxoid]

I suggest that B3D/ Dave H contact Kyle and let him know he is in breach of copyright by reposting selected parts of someone elses thread verbatim, and ask him to remove the post.

The fact that Kyle is censoring opinions opposite to his own does not mean we should start doing so. And anyway, threats like that would only give him more ammunition.

Besides, that really is not a breach of copyright, just selective quoting.

 

[Dave H]

the points used were only at the spawn points, so it will depend on the number of detail textures, and the types of textures at that point only. the level I used was well away from a spawn point becuase there weren't any detail textures close to the spawn point.

Ok, fair enough. Clearly in order for the mipmap transitions to be visible we need an area with cooperating geometry (long flat areas extending in the z-direction) that is detail textured, and decent lighting. I'd appreciate some vague approximation of how often such areas arise in practice. Obviously they exist--you have pictures to prove it. But would you say that the fact that apparently Brett and Kyle didn't come across any such areas while running around to assess the issue in motion means that they didn't do a very thorough job, or are they actually relatively rare?

An answer (from anyone) in the form of "in my estimation, mipmap transitions would be visible in motion to someone actively interested in general IQ but not specifically looking for them across x% of the overall UT2003 game world" would be highly appreciated.

Where did Nvidia claim they were doing full trilinear? All they promised was "quality image quality"!

Dave, I went over to Santa Clara for a days worth of meetings on the day of the 5900 launch - I sat through at least an hour presentation going over the IQ changes of 44.03 and how it matches ATI, and how the "Quality" mode gives Trilinear but without the "Debugging" - I think I even have this recored on my PDA still! (Also, given that they are doing these types of things in the quality mode this does in fact bring into question whether there really was any debug stuff going on). NVIDIA also guied reviewers to use particular tools to highlight the IQ output under the various modes and then, consequently that is removed in UT2003 at least.

Understood. I didn't realize quite how explicitly the claim had been made that 'quality' meant trilinear. I understand why you'd be pissed about this.

But to me this is still different than if the setting was labelled "trilinear filtering" in the drivers themselves. As it is, Nvidia lied (or perhaps, if you could examine an exact transcript of the meeting you'd find in hindsight that they'd merely carefully mislead, as they did with the whole 8x1/4x2 mess) to reviewers, and thus facilitated a round of incomplete if not partially misleading reviews. That's quite bad. But at least it's been discovered and disseminated, and any future review failing to discuss the issue will have only the reviewer's incompetence to blame. Only that fraction of consumers who pays attention to the review sites were ever affected, and presumably most are now aware that the issue exists (although for some too late to affect their purchase decision). Those who don't read the review sites were never under any impression that 'quality' necessarily meant trilinear in the first place, only that it offers sufficiently higher quality than 'performance', which it does.

That's different than if Nvidia had actively mislead their users by misrepresenting what the slider does in the drivers. I suppose it's a subtle difference, but it's worth something IMO.

think about it for a second Dave - what does Trilinear do over Bilinear? Take more samples; by reducing the the level of Trilinear you are reducing the number of samples taken per pixel, and what does undersampling result in? Aliasing.

Right, that's why I asked. But it's not as open and shut as you make it seem: you have to be "undersampling" not in some general sense but relative to the Nyquist limit, and the mere fact of having extra samples won't necessarily help if they don't provide the right sort of extra data. It was my impression that by keeping the mipmap LOD within .5 of the ideal LOD, bilinear was enough to prevent texture aliasing (assuming the LOD bias wasn't screwed with to make nicer-looking screenshots). Now that I think about it, though, I'm not at all sure that that impression is correct.

Nor am I sure that trilinear should necessarily help things. Yes it takes more samples, but it also samples from a more detailed mipmap half the time. I would guess the blend with the less detailed mipmap should handle any aliasing, but I can't convince myself one way or another without quite a bit more thought or even an overdue purchase of a graphics textbook. Instead I'd rather one of the many forum members who've already read graphics textbooks just explain it to me.

In any case, it doesn't help matters that the purpose behind trilinear is universally given as "getting rid of those mipmap transition lines you get with bilinear".

I mentioned before the Antalus level that was used (this is the grass covered level thats used in the standard UT2003 benchmark, along with another level) does feature a detail map acorss the entire grass surface. The detail map is there to generate the grass detail - now, in this level it is actually more difficult to see the mipmap transistions introduced by the lowering of the filtering, however because of the nature of the textures you are more likely to increase the amount of texture alaising noticed when in motion.

Perfect! This is exactly the sort of thing I'm looking for. If you (or someone else with access to an NV3x) would compare texture aliasing in that detail textured grass, under normal "quality" filtering and with the anti-detect to enable true trilinear (and perhaps a comparison to an R3x0 for fun), and post your subjective findings, I'd be entirely satisfied. In fact, if it turns out that Nvidia's faux-trilinear leads to noticeable texture aliasing, I'll even register at [H] and post in the Dave H Lovefest thread about it!

As it is I can't tell if you've actually seen such issues or if you're just noting that they theoretically might exist. For [H] to be wrong on their own terms (as a site reviewing only on the basis of actual in-game experience), the quality differences have to be noticeable in-game.

EDIT: BTW, Dave B., I wanted to add that I've just now (after I posted this) noticed the reply you made to me several pages and several posts ago. I'd still like more detail, but it does answer many of the questions I keep asking. In particular, the comment that once you notice the difference you always notice it indicates to me that perhaps [H] came to the wrong conclusions after all.

So if it seems I've been ignoring that post, sorry, but I'd just somehow missed it until now.

 

[Dave H]

I suggest that B3D/ Dave H contact Kyle and let him know he is in breach of copyright by reposting selected parts of someone elses thread verbatim, and ask him to remove the post.

The fact that Kyle is censoring opinions opposite to his own does not mean we should start doing so. And anyway, threats like that would only give him more ammunition.

Besides, that really is not a breach of copyright, just selective quoting.

Actually it is technically a breach of copyright; he published my entire post (slightly incorrectly, as it apparently couldn't reproduce the nested quotes), instead of selectively quoting it which would fall under fair use. But as I don't much agree with that aspect of copyright law, I'm not going to ask him to change the post. As you say, just because Kyle has been a dick to a bunch of people doesn't mean I should be a dick to Kyle. And just because I don't often agree with Kyle doesn't mean he can't think he agrees with me.

(Although I did notice he didn't quote my earlier post where I referred to his "ridiculous editorials, double-standards, [and] forum fascism". )

 

[Blackwind]

All in all I find the process of finding issues a great idea. Wanting better tool function within the drivers great. Better control from the user side ideal.

Wasting breathe Riding Moral High horses? A big fat waste of time. Great techincal posts, keep em coming. Great information with out any vendor worshiping pomposity would be Eden.