I may be wrong, but I understand D. Perez's answer as saying "the compiler is finished" as saying "it's a working product and not a work in progress". For example, Photoshop 7 is a "finished product", that doesn't mean Adobe is not working on Photoshop 8 (or 9, I can't remember what the Photoshop of the year is called).
Futuremark Announces Patch for 3DMark03
- Thread starter Nick[FM]
- Start date
I may be wrong, but I understand D. Perez's answer as saying "the compiler is finished" as saying "it's a working product and not a work in progress". For example, Photoshop 7 is a "finished product", that doesn't mean Adobe is not working on Photoshop 8 (or 9, I can't remember what the Photoshop of the year is called).
Magic-Sim said:Hanners said:Not as ridiculous as the complete communications breakdown that's been exposed at nVidia.
I know Uttar's editorial pointed towards it, but it almost feels like nVidia read it and said 'Hey, that's not true - It's actually much worse than that! Lets go and prove it to discredit him!'
Hey, just imagine what Uttar would feel like if it was proven that you touched the truth
Actually, don't quote me on this, but the pseudo-official NVIDIA stand is that they don't know what ULE stands for.
Although if they really did try to discredit me by proving they're worse than I said... I'd say I'd have to give credit to certain people who encouraged me to make the editorial more bitter and aggressive.
NVIDIA communication system is both too naive and too complex. It simply doesn't work.
NVIDIA needs serious consulting from people who actually know WTF they're talking about when it comes to GPUs, finance AND other misc stuff.
I say they should just hire me, but hey, I'm biased!
EDIT: Imagine that'd happen though, I know a few people who'd simply give up and quit the company on the spot - in fact, some of the ones that should either change or be fired anyway - so that'd be a helluva easy job, too
I hope you understood I was kidding, too
Uttar
Uttar said:Magic-Sim said:Hanners said:Not as ridiculous as the complete communications breakdown that's been exposed at nVidia.
I know Uttar's editorial pointed towards it, but it almost feels like nVidia read it and said 'Hey, that's not true - It's actually much worse than that! Lets go and prove it to discredit him!'
Hey, just imagine what Uttar would feel like if it was proven that you touched the truth
Actually, don't quote me on this, but the pseudo-official NVIDIA stand is that they don't know what ULE stands for.
Although if they really did try to discredit me by proving they're worse than I said... I'd say I'd have to give credit to certain people who encouraged me to make the editorial more bitter and aggressive.
NVIDIA communication system is both too naive and too complex. It simply doesn't work.
NVIDIA needs serious consulting from people who actually know WTF they're talking about when it comes to GPUs, finance AND other misc stuff.
I say they should just hire me, but hey, I'm biased!
EDIT: Imagine that'd happen though, I know a few people who'd simply give up and quit the company on the spot - in fact, some of the ones that should either change or be fired anyway - so that'd be a helluva easy job, too
I hope you understood I was kidding, too
Uttar
If you were hired and given the power to fire whoever you want to place someone more able in exchange, who woul'd you fire in first, and replace with ?
The whole hoopla and pointless division between synthetic benchmarks and games is getting more and more hopeless. What makes a triangle drawn by a "synthetic" benchmark any different than a triangle drawn by a "natural" game? If card A draws "synthetic" triangles twice as fast as card B, what is the chance that card B draws similar "natural" triangles as fast or faster than card A? If I want to know what card performs better I'd like to have even more synthetic benchmarks that factor out all other variables. Simple synthetic benchmarks (basically something like the 3DMark fillrate test, not the "game" tests) stress the hardware performance much more and takes away the chances driver can play dirty with. This gives a better picture of the actual capabilities of the card and projected future performance which can't be faked with any driver tricks.
Obviously game benchmarks have their place as well, but I wouldn't base my purchasing decision on games I don't play anyway.
Obviously game benchmarks have their place as well, but I wouldn't base my purchasing decision on games I don't play anyway.
Magic-Sim said:If you were hired and given the power to fire whoever you want to place someone more able in exchange, who woul'd you fire in first, and replace with ?
I don't think NVIDIA really needs a lot of terminations. That also has the risk of reducing employee moral; just not a good idea IMO.
Although a few "example" terminations could be justifiable. Dan Vivoli's would be the first on my list, although I'm sure there are other people meriting that treatment even more.
My Top 3 list of "has to be done". These are often relatively quick things and ideas, which do NOT include terminations. They only resolve part of the problem; certain other things, which are more more precise and, I'd be tempted to say, "microarchitectural" are necessary. For example, neither of these 3 points will help engineering much, but by making team-to-team communication easier.
1a) Reorganize communication. This would be done through people who got good technical knowledge ( although doesn't need to be great either ) and some managing experience.
These people would have the power to halt ANY public releases or actions related to the departments they deal with. And they should have high influence when it comes to suggesting to change plans and projects. Other misc communication changes are also required of course, as this would in no way be sufficent, but it'd already help a fair bit.
I also believe more quickly written internal guidelines are required; minimizing the time to takes from concept to internal publishing is important, and that means hiring two or three more writers would be a good idea.
1b) Simplify organization for engineers. It's just a tad too complex right now. Without very detailed information about how it currently works though, it's hard to think about any quick fixes, though.
2) Have databases (MySQL?) of released public statements. Making DR, PR and IR not contradict themselves is crucial IMO. This would also be useful for managers to realize if some incorrect things have been said in the past, possibly on purpose or not.
Have a simple and to the point system for engineers being able to comment on those public statements, anonymously if they wish so - this would prevent crap from being released to the public without anyone in the company realizing it.
Of course, should engineers begin to do nothing but comment on public statements, something would have to be done to minimize how much they can comment and read them
3) Get 2-3 people maximum who's *only* job is to analyze the public ( both forums and "common", Joe Consumer public ). Clearly make the difference between the different market segments and what type of PR/Marketing is positive/negative to get the sales.
NVIDIA doesn't seem to have a clue what the public wants and thinks anymore, and such a move is required to minimize unjustified debacles IMO.
That's just IMO. I don't pretend to know everything I should know, and that's because I don't. I don't pretend any of this could help NVIDIA, as I've got no consulting experience, and I'm not ready to claim I'm really 40 and worked as an industry analyst and consultant for the last 10 years. Because that's simply not true.
Uttar
I didn't notice this being mentioned:
It would be interesting to examine the details of the PS 2.0 test of 3dmark 03. Register usage characteristics, the role of SINCOS, and the significance of the role vertex processing plays...all would be interesting things to go over at some point.
It would be interesting to examine the details of the PS 2.0 test of 3dmark 03. Register usage characteristics, the role of SINCOS, and the significance of the role vertex processing plays...all would be interesting things to go over at some point.
jpaana said:
I have to agree with you about that. I play mostly online games like everquest, Dark Age of Camelot ad so on and theses games are never benchmarked as i dont even think it s possible to do.
When i read people posting about real games will tell the truth, they are just speaking about some FPS or fligtsim, but what about all other games ?
Uttar said:Magic-Sim said:If you were hired and given the power to fire whoever you want to place someone more able in exchange, who woul'd you fire in first, and replace with ?
I don't think NVIDIA really needs a lot of terminations. That also has the risk of reducing employee moral; just not a good idea IMO.
Although a few "example" terminations could be justifiable. Dan Vivoli's would be the first on my list, although I'm sure there are other people meriting that treatment even more.
My Top 3 list of "has to be done". These are often relatively quick things and ideas, which do NOT include terminations. They only resolve part of the problem; certain other things, which are more more precise and, I'd be tempted to say, "microarchitectural" are necessary. For example, neither of these 3 points will help engineering much, but by making team-to-team communication easier.
1a) Reorganize communication. This would be done through people who got good technical knowledge ( although doesn't need to be great either ) and some managing experience.
These people would have the power to halt ANY public releases or actions related to the departments they deal with. And they should have high influence when it comes to suggesting to change plans and projects. Other misc communication changes are also required of course, as this would in no way be sufficent, but it'd already help a fair bit.
I also believe more quickly written internal guidelines are required; minimizing the time to takes from concept to internal publishing is important, and that means hiring two or three more writers would be a good idea.
1b) Simplify organization for engineers. It's just a tad too complex right now. Without very detailed information about how it currently works though, it's hard to think about any quick fixes, though.
2) Have databases (MySQL?) of released public statements. Making DR, PR and IR not contradict themselves is crucial IMO. This would also be useful for managers to realize if some incorrect things have been said in the past, possibly on purpose or not.
Have a simple and to the point system for engineers being able to comment on those public statements, anonymously if they wish so - this would prevent crap from being released to the public without anyone in the company realizing it.
Of course, should engineers begin to do nothing but comment on public statements, something would have to be done to minimize how much they can comment and read them
3) Get 2-3 people maximum who's *only* job is to analyze the public ( both forums and "common", Joe Consumer public ). Clearly make the difference between the different market segments and what type of PR/Marketing is positive/negative to get the sales.
NVIDIA doesn't seem to have a clue what the public wants and thinks anymore, and such a move is required to minimize unjustified debacles IMO.
That's just IMO. I don't pretend to know everything I should know, and that's because I don't. I don't pretend any of this could help NVIDIA, as I've got no consulting experience, and I'm not ready to claim I'm really 40 and worked as an industry analyst and consultant for the last 10 years. Because that's simply not true.
Uttar
Well I didn't ask a flawless strategy, I just asked what You'd do in your own opinion, and you answered fully my question, and, even more. Thanks.
It is just sad to see that they are cutting themselves from the real world, hey, maybe Matrix dissuaded them to take the red ang go down with the white rabbit to wonderland.... Who knows, whith them, we can make tons and tons of hypothesis
PatrickL said:
Flight Sim's ? Even those are rarely used. And I cannot remember the surprise I had when Marc from Hardware.fr tested the 5600 Ultra on car sims !
http://www.hardware.fr/articles/468/page12.html
Ouch you said ?
If there is something like that in 52.16 driver, compiler perhaps doesn't work in 3DMark03 v340 :
If 3DMark03
Then activate 3DMark03 shader remplacement
and disable compiler (because useless)
Since 52.16 with 3DMark03 340 have much better performance that 44.03 with 3DMark03 330 (25.1 vs 19 fps), i doubt this is the case, but it's possible ... isn't it ?
If 3DMark03
Then activate 3DMark03 shader remplacement
and disable compiler (because useless)
Since 52.16 with 3DMark03 340 have much better performance that 44.03 with 3DMark03 330 (25.1 vs 19 fps), i doubt this is the case, but it's possible ... isn't it ?
Magic-Sim said:Who knows, whith them, we can make tons and tons of hypothesis
I had a fun, non-serious one I posted somewhere else...
Heck, you could even say NVIDIA's wisest business decisions were made at the time Donald Rumsfeld, currently part of the Bush Administration, was a business consultant for the company. ( Yes, that's true he worked for NVIDIA as a business consultant for a while, around 1998 I think ). But without any tangible proofs, I'd have a hard time defending that point of view
Donald Rumsfeld sold million of dollars worth of NVIDIA stock when he became part of the Bush Administration, as he could not keep it, since it could make him "subjective".
Uttar
Magic-Sim said:ROFLAMO, or he felt that nVIDIA would go down the hill at light speed ?
Hehe. You'd have to get into conspiracy theories about how he also sold other stocks to cover his tracks, then
Plus, he didn't quite have the choice to keep his NVIDIA stock, anyway.
Uttar
Uttar said:
Okay, I've tried to search out an answer, but I can't seem to find one with regards to the register usage problem.
Why? Why is there such a significant penalty? It can't be something silly like that there's a problem sharing The One Register
.Why is it still extant in the next generation GPU? What, are they literally missing dozens of registers that the competition has and can't remember where they are?
Does competing hardware have the same problem, albeit with greater numbers of registers? Why does the NV3x architecture show such pronounced and seemingly anomalous behavior so easily? What did they put in place of having additional registers?
I'm not singling out anyone in particular for answers, I'm just hoping somebody knows.
I'm afraid that I'm beginning to obsess over it, which doesn't appear healthy.
Magic-Sim said:A conscience case, maybe ? Or like you told so, a vague fear of some lawsuit
No, just something they have to do. "Conflict of interest" and what-what, you know? (Not that it really stops with the selling of stock anywho.)
Derek Perez response to Dave Bauman
[/quote] At one we asked Derek how this sat with the optimisations guidelines that were given to press by NVIDIA, specifically the guideline that suggests "An optimization must accelerate more than just a benchmark" To which Derek's reply was "But 3DMark03 is only a benchmark" -- we suggested that this particular guideline should read "An optimization must accelerate more than just a benchmark unless the application is just a benchmark"! [/quote]
I took from Derek Perez's response to Dave Bauman that it was okay to inflate any non game benchmark to reflect "actual game performance". If that is accurate how will the other benchmark makers respond?
[/quote] At one we asked Derek how this sat with the optimisations guidelines that were given to press by NVIDIA, specifically the guideline that suggests "An optimization must accelerate more than just a benchmark" To which Derek's reply was "But 3DMark03 is only a benchmark" -- we suggested that this particular guideline should read "An optimization must accelerate more than just a benchmark unless the application is just a benchmark"! [/quote]
I took from Derek Perez's response to Dave Bauman that it was okay to inflate any non game benchmark to reflect "actual game performance". If that is accurate how will the other benchmark makers respond?
Is it possible that the source of the "they disabled the compiler" story is that nv was doing a two step id on 3dmark --step one detected the app, and told the compiler that when 3dmark was detected to not use the compiler optimizer, but instead id by the digital fingerprints and replace the shaders with the hand-tuned ones? So that when the digital fingerprints were no longer matching, *no* optimization happened, not even that that the compiler could have legitimately supplied? This would also explain why running both the 330 and 340 shader outside of 3dmark would return the same results.
Or am I trying too hard to give them even the tiniest bit of logic in the way they view the world?
Or am I trying too hard to give them even the tiniest bit of logic in the way they view the world?
Similar threads
