19167 3dmark03 Score on the Hall of Fame (FM)

Personally I would think such a switch would be nice to analysing driver/shader performance for reviewers like those here on Beyond3D.

 

How so? What would you use it for?

 

Maybe it saves them the time of having to write thier own routines to disable the Pixel Shaders....if the drivers can do it for them then thats time saved (however small) and time that can be spent on other more important issues..

 

You don't have to re-write any routines from my understanding, just re-compile the code with a switch.

 

Well, since it isn't a "meaningful optimization," and it is only able to "render garbage" while spitting out perfectly legible benchmark scores at the the same time, scores based on the speed of rendering garbage instead of the benchmark being run--what's your assessment as to the purpose of this "option," then? If not a device designed specifically to output bogus benchmarking scores based on garbage rendering speeds--what is its purpose?

In all instances I'm aware of, the certainty as to the things you allege came directly from ATi, in the form of public answers to questions asked. Hence, where is the intent to "hide" anything?

nVidia, however, when asked similar questions, often does not answer at all.

Also, unlike nVidia's claimed trilinear optimizations, ATi's turn themselves on and OFF automatically, depending on software conditions. Turned on, nVidia's *never* turn themselves off under any software conditions. As well, what proof do you have that when you use the CP to turn nVidia's Trilinear optimizations off--that they are indeed turned off entirely for all 3d games?

ATi's texture stage optimizations, as you put it, apply only to forcing settings through the CPanel, and not to setting things up from within the games themselves. Hence it is not accurate to state that texture stage treatments are ever forced *except* when using the CPanel to force them (in the event a game has no internal settings controls.) Oddly enough, as well, I note that your assessment of texture-stage optimizations omits the fact that nVidia does the exact same thing with respect to texture-stage treatment when forced through the CPanel (last I read.)

Last, and most importantly, the kinds and types of all these optimizations as executed by the two companies are entirely different in nature, scope, and result. So any direct comparison between them which is based only the general descriptive *type*--such as "Trilinear optimization," etc.--is entirely an apples to oranges comparison. In other words, talking about ATi's "trilinear optimizations" neither justifies or defines or describes nVidia's "trilinear optimizations," and vice-versa.

 

Regardless.
It's still more than needs to be done if the drivers can do it for you.

 

But, unless, as has been pointed out, developers can be assured that every person running their software will have access to drivers that offer such an option, then developers will still have to do it themselves internal to their own software, which means the option in the drivers as such will have no direct relevance for them in making their coding any easier.

 

WaltC, I think your point is understood. But could you please come up with a scenario in which Nvidia can leverage this option to their advantage outside of an in-house "yaaay we got 23K with no shaders enabled" chant by the water cooler?

 

OK, guys - this is not the place to get personal. You got a problem with someone, take it to PM's or email.

 

Excuse me if I am unable to construe your point... Isn't posting bogus scores to the 3dMk "Hall of Shame" precisely what prompted this discussion? If so, then you have your answer... Also, it wasn't *that* long ago, really, when we were all marvelling at nVidia cheating 3dMk and then quitting the 3dMk program after being caught... So, I find it amazing that anyone should even question that nVidia might willfully do something like this relative to inflating benchmark scores, since nVidia's done it before, and in many various ways. Pardon me if I misconstrued your question, though, but nVidia has always put a ridiculously high premium on benchmark scores--often to its PR detriment, I might add.

 

If you read the option. It's to improve performance on areas that are pixel shader bound. The option cant be enabled unless you install a coolbits registry hack. It's not enabled by default. The idea of the feature is to discover whether your GPU is bottlenecking you.. It's just like Nvidias LoD manipulation. Changing it to +3.0 will improve bandwith/texture bound situation. It's a test purpose function that happens to improve frame rates, They also have a function that nullls T&L.

This feature has been in Nvidia drivers for a long time now. But just because the feature is there doesnt mean its single purpose is to inflate benchmarks. I dont see how this is any different than Nvidia giving users high control over LoD.

 

Ummmm did Nvidia post the scores? I've never seen a car manufacturer get indicted in a hit and run, have you?

Nvidia has cheated in the past but in all those cases it was a cheat/optimization that was enabled by default in the driver and attempted to increase performance by a slight reduction in IQ. They were present in all the reviews / articles posted online and in magazines. How does one idiot posting a bogus score on Futuremark using a driver switch that generates garbage IQ equate to those past scenarios? Yes people put a premium on benchmark scores but I'm quite sure 100% of people look to reviews for those scores and not the Futuremark rankings. Especially since those Futuremark scores are obtained with ridiculously souped-up hyper-chilled overclocked systems.

Do you construe now?

 

Cheating is cheating...... but in this case, it's not nVidia that's cheating. However, nVidia should really be trying to avoid this kind of thing, especally considering how they have acted over the last 2+ years. Of course, FM should be held somewhat accountable for not pulling this score as soon as the cheat was discovered.......

 

It's not a way for NVIDIA to cheat in benchmarks and go "OMG OMG LOOKZIE WE HAV TEH HIGETS THREEDEEMARCS!!!1" It's a way for the silly little fanboys to do that. It's just like LOD hacks, and texture quality sliders, et cetera, et cetera...

 

I think you are somewhat missing the point here. If you want to talk about a company being elusive and arguably deceptive, then you really need to talk about both NVIDIA and ATI. For example, how about the instructions that ATI sent out to reviewers urging them to disable all optimizations on the NV 6800 cards, even when they were secretly using tri and anisotropic filtering optimizations on by default (with no good way to disable them) on the X800 cards!

EDIT: I see that you pointed out in a later post that both companies have been deceptive. No argument from me on that point.

NV will obviously not divulge totally complete details on their architecture, to protect their intellectual property and to protect their own interests. This is the same reason why ATI does not divulge all details on their optimized algorithms. To single out one company makes little sense. We all know that they are obligated first and foremost to their shareholders.

 

Ok so firstly Nvidia was cheating. And now they are not policing their code well enough so that dishonest people won't manipulate it so they can boast about their 3dmark score? hahahahaha. you're kidding right? The only way this will affect anybody's opinion of gpu performance is if all review sites were in a conspiracy together to enable the option.

Like Baron said, all this provides is a way for clowns to have some fun. It's not Nvidia's responsibility. If anything Futuremark should just delete the entry and put in a check for that flag. Sheesh.

 

just to clear something up...when i said "it's not a bug, it's a cheat"...i did not mean that nvidia was the one cheating, i meant the person benchmarking (with the intention of gaining a higher score and not for the impacts of the PS) did...it was not my intention to label nvidia as cheaters and start a flame war, although i do see how it could be taken as such and for that i apologize for not clarifying

 

Well one thing is certain, I still apparently got the ability to piss off a whole lot of people pretty easily.

 

I dont think anyone here is angry. But seriously questioning your logic on this matter.

 

I really think you should quit while you're only a tad behind.... (Yes, auto manufacturers are routinely sued when it's determined fatal accidents occurred because of intentional defects in manufacturing they knew beforehand could cause accidents, but included nonetheless.)

Back to 3d IHVs, though...nVidia did not post the original cheating 3dMK "hall of shame" scores, either--which were posted by nVidia users before knowledge of the original cheats came to light. The people posting the scores did not write the drivers, and so in that sense are not the guilty parties--how you might think they could be is beyond me...

The problem is that there at least 4-6 separate methods nVidia used to cheat at that time, not just one, and when one method was exposed they'd quickly find another way to cheat.... Their methods ran the gamut from reducing the off-track workload running 3dMK03, to permanently disabling detail-texture trilinear in UT2K3, to later disabling it for all 3d games, to so-called "compiler optimizations," to FSAA fudging with ordered grids and/or post-filter blending, to "broken" trilinear optimization defeat switches, etc., ad infinitum--whenever one method was exposed nVidia would quickly find another. The goal was always the same: raising benchmark scores to levels exceeding that normally seen in running 3d games with their products.

Easy answer: it inflates benchmark scores artificially, which is the common thread relative to all such behavior (You seem determined to keep missing the point.) The important point is not the action of "one idiot", it's the fact that an option which fantastically inflates benchmark scores while rendering garbage without *crashing* those benchmarks exists (obviously not useful to anyone running and playing 3d games), and that such an "option" finds its way into a set of IHV drivers in the first place.

Why does anyone need a "render garbage at fantastic framerates" option in his CPanel? Heh...

No, what I said was that *nVidia* places a ridiculous premium on benchmark scores, a premium totally out of proportion to any benefit to end users or nVidia alike. End users don't benefit from inflated benchmark scores since that performance is not supported in the 3d games they run, and nVidia doesn't benefit when it is found out and exposed. It's a "one-step-forward, two-steps back," kind of thing. For every person fooled by such things, two persons are turned off to nVidia, etc.