nVidia's Problems

[CaiNaM]

Originally posted by: VIAN
The 5800 Ultra was a pretty good card. I don't think it was as quick as the 9700 Pro, but still very fast. And the cooling looked pretty cool. The only problems it had were bad drivers and loud inefficient cooling. The 5900 hardware tweak was barely an upgrade at all. The 5900 gets bested by the 5800 in most tests. But the 5900 Ultra bests the 5800 Ultra in most if not all tests. There is very little difference in performance. Makes you wonder why they even did the tweak. The only section the 5900 excels over the 5800 is in AA and some shadow calculations.

? the 5800's used a 128-bit memory interface, and it's performance was atrocious compared to ati. even nvidia fanboys quickly change the subject when someone bring up the 5800

 

[CaiNaM]

Originally posted by: SilverTrine

Originally posted by: CaiNaM
an no bfg, your're rationalizing the degrees of "cheating" and that's BS. cheating is cheating, and to claim or assume one who has been caught cheating multiple time is now "not cheating" simply because they haven't gotten caught is nieve at best. however, this while line of thining it out of context for this thread anyways, so if you really want to continue on this path start another thread, i'll be happy to join in, but it doesn't belong here.

I hope ATi sues you for defamation, trust me 3dmark2003 is not detected by ATi drivers and is by Nvidia. We can say 100% that Nvidia cheats and ATi doesnt.

lmao...

 

[Insomniak]

Originally posted by: SilverTrine

I hope ATi sues you for defamation, trust me 3dmark2003 is not detected by ATi drivers and is by Nvidia. We can say 100% that Nvidia cheats and ATi doesnt.

lmao....fanboy much?

 

[VIAN]

? the 5800's used a 128-bit memory interface, and it's performance was atrocious compared to ati. even nvidia fanboys quickly change the subject when someone bring up the 5800

This is with information from VGA III. Check it out for yourself. I don't think it was a bad card. Just disappointing. Too much hype before release.

BFG ignore previous posts about cheating. I have already agreed with you on the rules here.

 

[Megatomic]

Originally posted by: Insomniak

Originally posted by: SilverTrine

I hope ATi sues you for defamation, trust me 3dmark2003 is not detected by ATi drivers and is by Nvidia. We can say 100% that Nvidia cheats and ATi doesnt.

lmao....fanboy much?

Apparently not, his fanboy-fu is weak.

 

[BenSkywalker]

Then again I have to ask why even list it in this thread as a "problem". Again I'll ask: (1) how many of those bullet points do you actually agree with and (2) what is the point of this thread?

I don't think it's wrong, I think it is a problem. In the US we have(realisticly) only two political parties. I don't think this is wrong, however it is a problem. I agree with all of the bullet points I posted, have no doubt about that.

ATI doesn't do application detection. However it appears to be company policy for nVidia
That is utter rubbish and you know it. I've admitted many times that not all performance gains on nVidia's part are cheats and I've even listed several of their actions that I class as genuine optimizations.

Not trying to patronize at all. Say you take a look at the review of the new FW 54.xx drivers and you see performance up by 20% in one game bench, what is the first thing that is going to go through your mind? We both know.

So renaming the executable string produced different results? Or are you saying that you tricked the game into rendering a different way?

I used an anti detect app to trick the game, worked for the most part.

In any case, a lot of the benchmark games were having issues on nVidia boards too (including AquaMark, ironically even while nVidia was pointing the finger at ATi).

He commented on the Halo issue and their inability to find any differences. You may recall that I was believing the varrious sites that stated such things for quite some time, they were wrong. There are very clear differences between ATi's part and nVidia's part even now after ATi has improved their drivers for the game.

 

[BFG10K]

I don't think it's wrong, I think it is a problem.

But if it causes a problem doesn't that make it wrong? If nVidia only performed genuine optimizations then there'd be no problem.

Say you take a look at the review of the new FW 54.xx drivers and you see performance up by 20% in one game bench, what is the first thing that is going to go through your mind? We both know.

What do you expect me to think when a reviewer then renames the executable, runs anti-detect, notices image quality problems that weren't there in previous drivers (which just happened to be slower too) and/or runs a custom timedemo and notices the performance plummet?

I'll gladly accept any benchmark gain that doesn't suffer any of the above effects when exhaustively scrutinized, regardless of the vendor.

Of that list two of them are cheats and the custom benchmark is also highly likely to be exposing cheating too. The IQ issue could be a bug but at the end of the day it still invalidates the performance gains.

The days of simply looking at benchmark graphs and taking them at face value are long gone.

There are very clear differences between ATi's part and nVidia's part even now after ATi has improved their drivers for the game.

I don't deny that ATi might've had bugs in Halo. What I take issue with is the accusation that they were cheating.

 

[BenSkywalker]

But if it causes a problem doesn't that make it wrong?

No, it means there is a problem with it. Free speech brings with it all sorts of problems, certainly doesn't make it wrong

If nVidia only performed genuine optimizations then there'd be no problem.

The genuine optimization part is the problem.

I'll gladly accept any benchmark gain that doesn't suffer any of the above effects when exhaustively scrutinized, regardless of the vendor.

This is the kneejerk reaction I was talking about. You have to wait and see that the above instances don't impact anything first, then you can accept them. You auto assume a cheat, that is a problem.

The days of simply looking at benchmark graphs and taking them at face value are long gone.

For you, for nVidia parts. This is why their optimizations are a problem.

I don't deny that ATi might've had bugs in Halo. What I take issue with is the accusation that they were cheating.

I can change several rendering elements of the game considerably by fooling the app into thinking it is running a FX part. If the reverse were true of a nVidia part you would without a doubt be saying they were cheating.

 

[BFG10K]

No, it means there is a problem with it. Free speech brings with it all sorts of problems, certainly doesn't make it wrong

Well let's put it another way: if nVidia performed genuine optimizations would you list it in this thread as a problem?

The genuine optimization part is the problem.

No, cheating is the problem. Again if nVidia only performed genuine optimizations then you wouldn't list that bullet point in this thread.

This is the kneejerk reaction I was talking about. You have to wait and see that the above instances don't impact anything first, then you can accept them.

If you burn your hand on an active stove a few times do you keep sticking your hand on it before checking the hot light first? No? Then why - in the face of overwhelming cheating evidence against nVidia in the past - should I suddenly start ignoring it and start taking their benchmark results at face value without checking them first?

If I did I'd be nothing more than a blind zealot, willing to accept anything that their pretty graphs show without forming my own thoughts about it.

You auto assume a cheat, that is a problem

And you assume that nothing of nVidia's is a cheat and that nVidia have never cheated in the past (except for the static clip planes). That is a problem.

For you, for nVidia parts.

Yes, just like after burning my hand on a hot-plate a few times I always check whether the light is out first before touching it. Correct me if I'm wrong but I'd consider that fairly normal behaviour, wouldn't you?

I can change several rendering elements of the game considerably by fooling the app into thinking it is running a FX part. If the reverse were true of a nVidia part you would without a doubt be saying they were cheating.

I think not. Changing how the application behaves isn't changing how the drivers behave. An application can do what it likes; there's no concept of a cheating application (except maybe if it was entered into a programming contest and broke some rules there). The problem is when generic shared code (such as drivers or the API) starts getting massaged in such a way to artificially inflate performance in a non-realistic way.

Hard-coding generic drivers to applications is as ridiculous as Microsoft coming along and changing the Win32 API to run Quake III faster, or Intel including SSE3 instructions that only accelerate Halo. It's totally and utterly ridiculous and no neutral programmer would ever consider that a valid optimization.

 

[JJN]

ATI has pulled alot of bullshit in the past, lets not put our fan boy goggles on too tight now, geeesh.

 

[Gagabiji]

If you burn your hand on an active stove a few times do you keep sticking your hand on it before checking the hot light first?

I've done that before, nothing new.

 

[Corporate Thug]

Originally posted by: Insomniak

Originally posted by: shady06
LOL that very first line is quite funny. its like saying honda made a crappy accord (which btw will never happen), its accords fault, not hondas

No, you apparently are incapable of associating meaning with words. What I said was like saying "The accord is crappy, not Honda."

You see, my uber-literate friend, NV3x has problems, But GeForces 1 - 4, all the Quadros, and the Nforce chipsets have been nothing but badass. Nvidia doesn't have problems - NV3x does. It's a shoddy architecture, and in no way representative of the company as a whole.

I never said anything about fault. You stuck that in on your own. If my statement is analogous to "It's accord's fault, not Honda's" then Skywalker's original post is like saying "The problem with Honda's board of directors are loose timing belts in the 1996 model year Accord."

Point being, Nvidia is a well diversified company that has put out a lot of great shit and has one bad product line on the market right now.

Please remove head from anus before posting in the future. Thank you in advance.

"The accord is crappy, not Honda" would be equal to saying NV3X is crappy not nvidia which i agree with you, but you clearly said it isnt nvidia problem, so NV3X, a product made by nvidia isnt there problem (ie a crappy accord made by honda isnt honda's problem)? the first statement does not make any sense to me. how is it that its not a companies problem if they make a "bad" product

just saying releasing a bad product is not the companies fault who releases shows a complete and utter lack of common sense

 

[BenSkywalker]

Well let's put it another way: if nVidia performed genuine optimizations would you list it in this thread as a problem?

That's exactly what I did.

Again if nVidia only performed genuine optimizations then you wouldn't list that bullet point in this thread.

The reason it is a problem is because it leads people to the wrong conclusion.

Then why - in the face of overwhelming cheating evidence against nVidia in the past - should I suddenly start ignoring it and start taking their benchmark results at face value without checking them first?

Overwhelming being the static clip planes in 3DM'03.

And you assume that nothing of nVidia's is a cheat and that nVidia have never cheated in the past (except for the static clip planes). That is a problem.

We have gone point by point and I've shown you where ATi has done equal or worse for the rest of the 'cheats' you claim(UT2K3 'brilinear' as compared to ATi's tri/bi as an example) and yet you ignore those points or try to marginalize them. I've stated numerous times I think the whole 'cheat' thing is a joke for both companies.

I think not. Changing how the application behaves isn't changing how the drivers behave.

And what do you think Futuremark is doing? I am so glad I finally got you to say this

 

[BFG10K]

That's exactly what I did.

So if you thought nVidia cheated you wouldn't list it as a problem?

Come on man, this game you're playing is certainly humorous but it's also quite silly.

The reason it is a problem is because it leads people to the wrong conclusion.

If you incorrectly believe it's a genuine optimization then it would also appear to you as leading people to the wrong conclusion.

Overwhelming being the static clip planes in 3DM'03.

Ask any neutral software engineer what he/she thinks of the nVidia actions that I've singled out and they will most certainly agree with my stance. Heck, even Carmack himself was saying that optimizing based on application name was wrong because it subverts the purpose of benchmarking it, and that's not too different to what Gabe was saying. Carmack must be on the ATi payroll too then, hmmm?

We have gone point by point and I've shown you where ATi has done equal or worse for the rest of the 'cheats' you claim(UT2K3 'brilinear' as compared to ATi's tri/bi as an example) and yet you ignore those points or try to marginalize them.

Not really; I've agreed that ATi's AF control panel issues somewhat resemble nVidia's though nVidia is worse because they overrule both the game and the user. As for the rest of the evidence (easily a dozen examples), well you've simply brushed it off using a wide range of invalid tactics.

And what do you think Futuremark is doing?

Absolutely nothing for non-cheating vendors. Remember nVidia is the odd one out here; their tactics not normal by any stretch of the imagination. Yet you insist on blaming FutureMark and claiming that nVidia is following standard optimization procedures instead of just admitting what the real problem is here.

 

[BenSkywalker]

So if you thought nVidia cheated you wouldn't list it as a problem?

I listed that their optimizations creates confusion which it obviously does, simply look how confused you are.

Absolutely nothing for non-cheating vendors. Remember nVidia is the odd one out here; their tactics not normal by any stretch of the imagination. Yet you insist on blaming FutureMark and claiming that nVidia is following standard optimization procedures instead of just admitting what the real problem is here.

As long as you are on board with ATi cheating in Halo and KoTOR then you have a point. If you aren't, then you are still very, very confused.

 

[reever]

Originally posted by: BenSkywalker

So if you thought nVidia cheated you wouldn't list it as a problem?

I listed that their optimizations creates confusion which it obviously does, simply look how confused you are.

Absolutely nothing for non-cheating vendors. Remember nVidia is the odd one out here; their tactics not normal by any stretch of the imagination. Yet you insist on blaming FutureMark and claiming that nVidia is following standard optimization procedures instead of just admitting what the real problem is here.

As long as you are on board with ATi cheating in Halo and KoTOR then you have a point. If you aren't, then you are still very, very confused.

As long as you think whatever you think is right and everybody else is either totally wrong or "confused" it's ok i guess.

 

[BenSkywalker]

Changing how the application behaves isn't changing how the drivers behave.

This is exactly what Futuremark did with build 340 in order to break optimizations that were API wide, general optimizations they couldn't break any other way. Now, if Futuremark deciding to change the application around and ending up with different results means nVidia was cheating, then by that exact same standard ATi is cheating in numerous games currently. Of course both of those statements are utterly absurd, but if you support one you must support the other or you are confused.

 

[BFG10K]

I listed that their optimizations creates confusion which it obviously does, simply look how confused you are.

I've been programming in some form or another for over twelve years. I also have a software engineering degree and three year's experience programming in the industry.

So do not presume to patronize me and call me confused when making calls about valid or invalid software engineering practices, OK?

As long as you are on board with ATi cheating in Halo and KoTOR then you have a point.

How did ATI cheat in Halo and Kotor?

This is exactly what Futuremark did with build 340 in order to break optimizations that were API wide, general optimizations they couldn't break any other way.

Then all vendors would have fallen over. Now take a look at the before and after graphs and you'll notice only one vendor sticking out like a sore thumb.

Their trivial changes were no different to renaming the executable name and causing their fragile optmizations to fall over. Is renaming the application string also the fault of the program?

It's interesting: first you claim that nVidia's application optimizations are valid and cause no problems, yet as soon as they're defeated in a trivial fashion (which no other vendor experiences I might add) you then start blaming the application for causing it. Again, nVidia can do no wrong; it's always someone else's fault.

 

[BenSkywalker]

I've been programming in some form or another for over twelve years. I also have a software engineering degree and three year's experience programming in the industry.

Then why do you fail to understand this issue? You keep ignoring one of the things FM did- they reordered instructions. I can 'break' SSE by reordering instructions using earlier compilers, so was Intel cheating? Can do the same for 3DNow!, was AMD cheating?

How did ATI cheat in Halo and Kotor?

The software vendor changed code that only impacted their parts and it alters output. The same thing that nV/FM are going through.

 

[BFG10K]

You keep ignoring one of the things FM did- they reordered instructions.

Correction: they reordered instructions in 3DMark. They didn't just take some random program, reorder its instructions and then feed that to nVidia's compiler. If they did I'm certain it wouldn't have made a shred of difference to the before and after results because nVidia's optimizations relied on detecting 3DMark in order to work, and that makes them invalid.

They are fragile, easily broken and rely on prior knowledge to work. Remember, no other vendor (ATi, Matrox and a few others) had problems with these reordered instructions. I mean this is one of the rules nVidia posted themselves that they wouldn't violate for heaven's sake, yet you still turn around and blame FutureMark for this issue instead of blaming nVidia who is at fault.

Carmack himself was saying application detection and optimization fragility was bad and his comments are in line with what Gabe was saying. What more evidence do you want? What more will it take to convince you that nVidia is cheating?

I can 'break' SSE by reordering instructions using earlier compilers, so was Intel cheating?

If their compiler was checking for a specific application being built on the basis of performing those optimizations (which otherwise couldn't exist) then yes. Of course they weren't so no, it was falling over because the compiler's optimizing technology wasn't as good as it could have been.

 

[BenSkywalker]

Correction: they re-ordered instructions in 3DMark. They didn't just take some random program, and re-order its instructions and then feed that to nVidia's compiler. If they did I'm certain it wouldn't have made a shred of difference to the before and after results because nVidia's optmizations relied on detecting 3DMark in oder to work, and that makes them invalid optimizations.

They are fragile, easily broken and rely on prior knowledge to work. Remember, no other vendor (ATi, Matrox and a few others) had problems with these re-ordered instructions.

Because you say it doesn't make it so. There is absolutely no evidence to support this at all, in fact their is very strong evidence to indicate that this is in fact completely wrong as when FM 'broke' the last set of nV optimizations they failed to 'break' their optimizations for the dedicated pixel shader test. If FM released a patch to break app detection, which isn't what they did, it would have broken all of the optimizations.

FM has stated what they did moving from build 330 to 340. This is the confusion I'm talking about. Because they do use app detection some times, people like you assume that they are doing it all the time and what's more, you assume that the manner in which they are doing it is a cheat.

 

[BFG10K]

There is absolutely no evidence

No evidence? So tell me, how did the other vendors react to these reordered instructions? That's all the evidence you need.

This is ridiculous. It's like renaming an executable string, having one vendor fall over as a result and then turning around and blaming the application on the grounds that its name has changed and it has "singled out" the vendor that fell over.

FM has stated what they did moving from build 330 to 340.

Yeah, they made trivial changes like reordering instructions and changing variable names. Trivial changes that no non-cheating vendor was affected by.

Because they do use app detection some times,

Sometimes? Sometimes is right, as in "if it's likely to be benchmarked then we'll do it, otherwise we won't bother". How do you explain the forty+ application detection strings Unwinder found in nVidia's drivers? Wait, don't tell me: he's on ATi's payroll too isn't he? Or is he Gabe's evil assistant? Or he's delusional and he didn't really see what he thought he saw?

 

[BenSkywalker]

Yeah, they made trivial changes like reordering instructions and changing variable names.

You've been coding for so long you should very easily grasp the direct comparison to the earlier SSE compilers. You could reorder instructions and Intel's performance dropped off by a decent amount while AMD's didn't move when you benched them. You are saying that Intel was cheating then, right? I'm sure you must have dealt with the earlier SIMD compilers.

Sometimes is right, as in "if it's likely to be benchmarked then we'll do it, otherwise we won't bother".

One of the strings they were using in the 53.03s was for MOH:AA, I don't recall having seen that game be benched, anywhere actually(not saying it wasn't, but I don't recall seeing it). Again I will remind you that I never had a problem with this back when PVR was doing it with virtually every app and you didn't seem to have a problem with it then either. Some people have a problem with it, such as yourself, which means it is a problem for nVidia. I don't think there is anything 'wrong' with it and I wouldn't have a problem if ATi did something comparable so I could run Halo as it should be without having to use 3DAnalyzer.

 

[BFG10K]

You've been coding for so long you should very easily grasp the direct comparison to the earlier SSE compilers.

It would be a direct comparison if both were checking for specific applications or neither of them were. However nVidia is checking for applications (by your own admission) to perform optimizations that othewise couldn't exist while Intel/AMD are not. As a result one is exposing flaws in the optimizations while the other is exposing flaws in the compiler.

Again, no other vendor had problems with FutureMark's changes. Not just nVidia vs ATi like Intel vs AMD mind you, there were several other vendors who all kept consistent levels of performance like ATi did. When you've got multiple vendors doing that while one vendor is consistently falling over it's almost impossible that say the other vendors are the ones who are abnormal. And if FutureMark was really doing something screwy then some of the other vendors would have almost certainly been affected.

One of the strings they were using in the 53.03s was for MOH:AA,

MOHAA games have problems with fog and haze on those drivers. If they're really checking for that game then they're doing one heck of a poor job.

 

[BenSkywalker]

However nVidia is checking for applications (by your own admission) to perform optimizations that othewise couldn't exist while Intel/AMD are not.

When did I say anything like they are performing optimizations that couldn't otherwise exist? We have had this discussion already and I'm not going through it again, I've explained to you mutliple different reasons why it can be done for beneficial reasons.

The rest of you posturing revolves around the fact that they are using optimizations that rely on app detection to increase benchmark scores. FM's latest patch lowered their score while failing to move the dedicated pixel shader test. Now FM is saying the pixel shader test isn't valid, so what's going on there? Either nV was using app detection and FM broke it(which means FM is lieing about the pixel shader tests), or they had specific compiler level optimizations for some of the in game shader used that FM changed around to break, but they missed compiler optimizations for the pixel shader specific tests. FM has come out and said they reordered instructions, which directly supports the second theory. The second theory also explains why certain tests dropped to what FM says is valid while others did not that FM says are invalid. Everything points against app detection being the issue in terms of 3DM2K3 as of now. I'm not saying nVidia doesn't do it, but all evidence now, including FM's own stances and statements, indicate that app detection is not the issue with 3DM2K3 nor has it been for a while now.

Again, no other vendor had problems with FutureMark's changes. Not just nVidia vs ATi like Intel vs AMD mind you, there were several other vendors who all kept consistent levels of performance like ATi did.

Which vendors? Point me to a vendor that comes close to rendering 3DM2K3 correctly with any of the builds besides ATi or nVidia.

MOHAA games have problems with fog and haze on those drivers. If they're really checking for that game then they're doing one heck of a poor job.

Agreed, and I listed that as one of their problems too