nVidia Application mode slow because of "debugging"
- Thread starter martrox
- Start date
Errr, what?
Any NV guys around want to offer an explaination?
Any NV guys around want to offer an explaination?
Don't tell anyone, but if you use Application mode, you'll get the exact same Pixel Shader architecture that on the NV35.
Damn, if debugging is really happening in Application mode, WTF does that thing do in release drivers, damnit?!
"The Gold Standard" - yes, maybe it's still slightly better in compatibility and in Linux, but damn, it's also the "Slowest Standard"...
Uttar
Damn, if debugging is really happening in Application mode, WTF does that thing do in release drivers, damnit?!
"The Gold Standard" - yes, maybe it's still slightly better in compatibility and in Linux, but damn, it's also the "Slowest Standard"...
Uttar
I doubt it's "debugging", but more like expensive corner case checking. The
OGL spec's (and probably D3D, but I don't know) have specified what should
happen if an application does something wrong. (one example that springs to
mind is that if any mipmap levels aren't defined then all texels are white)
Catching these error conditions can take quite a bit of time, and I suspect
that they don't do it in their "performance" modes. The really annoying side
effect is that there's a lot of apps out there that aren't written properly
becuase the developers don't realise their code has a problem.
grrrrrr.
OGL spec's (and probably D3D, but I don't know) have specified what should
happen if an application does something wrong. (one example that springs to
mind is that if any mipmap levels aren't defined then all texels are white)
Catching these error conditions can take quite a bit of time, and I suspect
that they don't do it in their "performance" modes. The really annoying side
effect is that there's a lot of apps out there that aren't written properly
becuase the developers don't realise their code has a problem.
grrrrrr.
Well, if this is the case then there's still questions - i.e. Why do it now, as it appears their previous generations don't have similar hits. Also, there's no speed differential between 'Application' and 'Balanced / Quality' on a GF4 last time I checked.
Uttar said:
I think that there was something on Anands (?) on this matter, saying that the aniso performance/quality will be fixed with newer drivers. And that those drivers wouldn't be released until the NV35 was released.
The question is then, why wait until the the release of the NV35 ?
Because of hardware bugs in the NV30 ?
Well, nothing changes in NV30 with the release of NV35 so the link seems tenuous.
BenSkywalker
Regular
There is with the 45.31s, although it is much less then what it was with the directly preceding drivers(it was ~50%, now its ~15%).
Is that with or without AF?
DaveBaumann said:
No, no, it makes sense.
If nVidia releases the AF fix before the NV35, they'll have to compare the NV35's fixed AF performance with the NV30's fixed AF performance.
If they release it at the same time, they can compare the NV35's fixed AF performance with the NV30's inefficient AF performance.
Don't you remember that nVidia's performance improving claims often compare the latest drivers at the release of the new product with older drivers of the old product? That's how their performance improvement claims alway look way better than they really are.
Heck, they want to claim *two times* the performance, doubling it, so such measures are really required in the NV35's case.
Uttar
Bjorn said:
nVidia has promised me "these" fixed drivers within a short period of time (as in days). Of course that won't be an official driver but I assume it'll get leaked to the web sooner rather than later.
So I think it's pretty safe to assume that these drivers will be available in one way or the other before the NV35 arrives.
BenSkywalker
Regular
Doesn't matter with the 45.31s, although I didn't pay attention in the prior revision to non AF scores(I just ran some quick Q3 benches when I read your post, will check some D3D now). It is ~10%-15% hit with or without AF or AA or with both.
Edit- Just ran some D3D benches, it shows up in some instances(without AF or AA) in the ~10%-15% range. These are all comparing Performance to 'Application'(which has been renamed to Quality, Balanced is now the middle setting).
Uttar said:
Exactly. And, if those 'hardware' bugs are fixed in the Nv31/34 then Nvidia might not want to compare the NV30 with the NV31 after the 'fix' is released. But, release the NV35 and it will cover up that problem a bit. And i'm guessing that the NV30 will be discontinued rather quickly also.
Ok, well, the above was all speculation of course. All Nvidia have to do is to release the drivers and prove me wrong
Bjorn/Uttar,
Bjorn makes the important link with NV31and that is key IMO. Sorry, I’m not sure I buy the idea that they would hobble NV31 reviews for the sake of making NV35 look better. NV30/NV35 are there to capture the mindshare, NV31/34 are there to capture the marketshare – an artificial poor showing (as we’ve seen so far) to make NV35 look better is just not at all sensible as they need NV31 to look favourable in the face of 9600 (which so far it seems that NV31 has gone down like a lead balloon in comparison).
The very fact that they have provided the ‘Balanced’ and ‘Aggressive’ modes and suggested reviewer use this is kinda proof that this isn’t the case.
Ben,
These stupid changes are getting right on my tits! So, you are compring the far left setting with the far right? I’m talking about the far left and middle settings. I tired it on either the 43.45’s or 42.69’s.
Bjorn makes the important link with NV31and that is key IMO. Sorry, I’m not sure I buy the idea that they would hobble NV31 reviews for the sake of making NV35 look better. NV30/NV35 are there to capture the mindshare, NV31/34 are there to capture the marketshare – an artificial poor showing (as we’ve seen so far) to make NV35 look better is just not at all sensible as they need NV31 to look favourable in the face of 9600 (which so far it seems that NV31 has gone down like a lead balloon in comparison).
The very fact that they have provided the ‘Balanced’ and ‘Aggressive’ modes and suggested reviewer use this is kinda proof that this isn’t the case.
Ben,
These stupid changes are getting right on my tits! So, you are compring the far left setting with the far right? I’m talking about the far left and middle settings. I tired it on either the 43.45’s or 42.69’s.
DaveBaumann said:
But what about the NV36? If that is destined to replace the NV31 is a fairly quick manner wouldn't it be best for nvidia to leave both the NV30 and NV31 hobbled so that the refresh parts look really good. It give the impression that they are progressing faster than ATi at scaling their architecture.
BenSkywalker
Regular
Yes, I was comparing the far left and far right settings(actually, I use the quick tweak menu but I digress). I'll go back and try it with the middle setting compared to the 'Application' setting and see if the results change. I'm assuming what you are looking for is non AA/AF benches?Without AA/AF there isn't any difference in performance. With AA/AF it's in the ~15% range.
BoardBonobo said:
But that doesn't explain the advocation of the use of 'Agressive' (fast) settings for NV31 (p)reviews.
DaveBaumann said:
Well, maybe the NV35\36 has fixed alot of the issues that plagued the NV30\31. It would look good for nvidia if their next refresh is the same speed or faster when using 'Quality' as the current gen is using 'Agressive'.
