nVidias rules of optimizing a driver

A

Ante P

Veteran
take a look at [H] three pics from an nvidia document that speaks about optimizations
pretty comical that they don't follow their own recommendations (ie don't lower quality and don't optimize solely for synthetic standalone benches and don't optimize for fly paths etc.)

perhaps the new 44.90 driver is the first fruit of these rules it doesn't seem to have any of the bad optimizations in it anymore
 
This is what I said in an other Thread about it :
=============================
And the newly leaked 44.90 beta doesn’t seem to contain any optimizations, as we obtained the same fps values on the same card at the same resolutions, as 44.03 and 44.65 versions had provided with Anti-Detect.
Click to expand...

Ahh...

Nvidia is playing "nice guy"

And so they will "prove" that the encrypted drivers have no (obvious) "optimisations" to lure the consumer / reviewers / critics / hw geeks away from the "optimisations" which are in the drivers now.

Afterwards, when no one will and can check the drivers due to the encryption the optimisations will be put back slowly over time and Nvidia is (relatively) save cause no one can check if the drivers behave correct anymore.

Nice try. IMHO this will even work. :cry:

So over the next few weeks we will see reviews from the big sites stating that the older drivers had "bugs" and the new driver 44.90 works correct now and so everything is fine and will be fine in the future too. Maybe even Nvidia will bring an new label like "approved drivers" to turn the debakel into an positive marketing stunt.

Everyone will be satisfied and goes back to normal. Then we are back to the old mode with normal time-demo's, easy benchmarking and cheating all over the place. Will be nice to watch.


=============================
In short: I believe it when I have seen nonoptimised drivers for (at least) one year.
 
44.90 drivers are BULL... I have them installed.

Hey what a surprise, same exact scores as with the 44.67 and 44.71 drivers!!! Wow!!! I guess they aren't cheating then!

Give me a break. Who does nVidia think they are fooling? The only people that read these websites and see that statement are "in the loop" so to say and know what the deal is with this cheating crap... They're just digging their hole deeper.
 
Everyone will be satisfied and goes back to normal. Then we are back to the old mode with normal time-demo's, easy benchmarking and cheating all over the place. Will be nice to watch.
Click to expand...
I wonder if this will happen with ATI if NV40 turns out to be better than expected. And, of course, it will happen if R420 is better than expected.
 
The Baron said:
Everyone will be satisfied and goes back to normal. Then we are back to the old mode with normal time-demo's, easy benchmarking and cheating all over the place. Will be nice to watch.
Click to expand...
I wonder if this will happen with ATI if NV40 turns out to be better than expected. And, of course, it will happen if R420 is better than expected.
Click to expand...

A fair question. I tend to think not, because of ATI's track record (including Quack) of responding to this issue. However, I also tend to think that it is important to keep looking and determing for ourselves based on reasoned observations.

What disgusts me at the moment is simply the selectivity of some sites in describing what we do find from reasoned observations, and how far down the list of priorities accurate consumer information seems to be in general. :-?
 
links:

http://www.hardocp.com/image.html?image=MTA1ODUxMjczMW9qOHdzb29SdU9fMV8zX2wuZ2lm

1058512731oj8wsooRuO_1_2_l.gif


1058512731oj8wsooRuO_1_3_l.gif


What year will these new ideals be implemented?
 
I love the way their internal dev process is a closed loop. Everything gets rejected or retarde... sorry, retargeted.

MuFu.
 
Here's what I see:

1. They hung somebody out to dry, at least rhetorically. They are clearly pointing at worker-bee individuals (not even mid-level management of the driver team) as the culprit. Over zealous employees. Management had no idea, etc. So, did they fire anybody, eh?

2. They hung mid-level management out to dry moderately, at least rhetorically. This would be the driver team management that didn't have the right process in place to catch the "over zealous" etc from point 1. Upper management (Yay for Upper Management! Hearts of gold, etc) had to get involved and re-engineer their process. Again, anybody get fired at this level?

3. The crux of their Guidelines is "must accelerate more than just a benchmark" and "must not contain a pre-computed state". My question for our local gurus is: Assume for the moment that nV is really sincere here (yes, I know how hard this will be --do it anyway), would this be enuf to solve the issue if (again, big IF) actually enforced? My impression from previous threads is that the answer may be "yes".

4. If the answer is actually "yes", then this is a useful document because it gives us a standard that nV has agreed to in advance, at their own initiative, with which to judge them in the future.

5. None of which matters all that much if the big sites won't call a rose a rose when they see one in the future any better than many of them are now. Ah well, sometimes all you can do is kick the can a little farther down the road.
 
You guys are going way overboard and past fair criticism. I see nothing controversial about those slides.

Since there is no formal specification as to what a "correct" image is (what is correct AA? Correct filtering? there's no standard, is ATI's aniso which fails at certain critical angles "correct"? It's subjective), this is of course nothing more than a guideline to the driver developers.

Since this is a powerpoint, we do not know what the speaker said at the driver team meeting, perhaps he elaborated and said "don't drop geometry, don't drop textures, etc"

It seems right now, NVidia could put out a presentation saying something innoculous like "Roses are red" and hundreds of fanboys would write critical messages saying "How dare they! LOL, there are yellow roses as well! Who are they kidding!!" There's a limit to how much you can try to read sinister motives into everything.
 
The fact that you don't trust them still doesn't mean you can twist any statement they make into a falsehood. If Nvidia says "2+2=4" on one of their slides, you gonna criticize it just because you don't like them?

There was a time when no one trusted ATI and people generately regarded their drivers and products as shite (pre-Radeon days), eventually they saw the light. If they are making a general effort to right past wrongs, do you want to encourage them not to, or help them?

A constructive critique of their guidelines would be much better.
 
i think those slides set some great guidlines.
FOR THE FUTURE
But, by their own admission, they WERE cheating (optimizations cannot change IQ) so how can nVidia get around that? It just seems that these slides labelt hem exactly what they did not want to be labeled...
 
I agree, however, saying optimizations can't change IQ is a slippery slope. We know precalculated data is bad, and we know that dropping textures or geometry is bad, but on the criteria that you can never change IQ, compression would be ruled out, as would be some AF techniques. Change is the wrong word, since you could improve the IQ and that would be a change.

I think a more legitimate statement would be to say that drivers are not allowed to disobey what the application, and ultimately, the user requests So if the application says "I want trilinear filtering on X", you are not allowed to drop back to bilinear unless the result would be indistinguishable.

Perhaps the principle should be driver optimizations should not cause regressions in image quality over previous versions
 
DemoCoder said:
Perhaps the principle should be driver optimizations should not cause regressions in image quality over previous versions
Click to expand...
Well, i agree for the most part - the only issue there is that IQ is subjective, leading to huge demalion-esque arguments :)

drivers are not allowed to disobey what the application, and ultimately, the user requests
Click to expand...
This is another good guide, but leads to other problems - mostly dealing with labeling things well enough so that the users understand what they are choosing. If you just have a button that says "aniso" and one that says "speed", its not clear that "speed" might effect the quality or level of "aniso". But thats a much greyer area.
Also, the drivers need to be able to disobey the users occasionally. IMO, nVidias solution to rendering errors with Splinter cell AA is a good one - just turn it off, no matter what the user wants. The only problem is then one of notifying the user. If the driver can detect a game being run, why cant it pop up a box that says "this game doesnt work right with our AA solution, so AA has been disabled in this game. Check this box to turn off further notifications for this game"?
 
You guys are going way overboard and past fair criticism. I see nothing controversial about those slides.
Click to expand...

Although the content of the slides seems very fair and reasoned, what is controversial is the source from which they came.

It's like seeing slides of how to run a fair and ethical business model from Enron, or how to properly and respectfully treat women from Mike Tyson.
 
You must log in or register to reply here.
Back
Top