The NV30 Chronicles: A Short story by Mr H. Binder

[BoardBonobo]

I agree with LeStoffer on that point, it seems strange that the driver set seems so unpolished. Soemthings just don't add up, maybe nV will release the card with a much better set? And what ever happend to Silent Running (TM), the dynamic control of the fan speed, has that just gone bye bye?

 

[epicstruggle]

You mean Hellbinder doesn't work for.

I thought he worked for their PR department.

Is there a chance of a refresh geforcefx before nv35 is released. where some of the problems of cooling, and fab issues are resolved? or will all this be addressed in nv35.

later,

 

[JF_Aidan_Pryde]

What's the final word in the NV30 primitive engine? Is it

1) Working?
2) In Silicon but broken?
3) Not in silicon?

 

[THe_KELRaTH]

Why do you assume that the drivers are so unpolished?
Isn't it just as feasable that the these drivers were supplied with the card with reviewers in mind and are designed to optimize 3Dmark and games like Quake 3, I doubt very much they expected any reviewer to test it on NFS-HP2.
I would think it far more likely that the final retail product will actually be slower due to more balanced drivers.
I expect we're going to see a magical fix for x2 Quincunx too but at a further performance hit.
My overall impression is that the card is shrouded by PR and missinformation. For example their terms for Anisotropic filtering, it seems the only way to use Trilinear is in Application Mode but at the same time it uses their version of Adaptive Texture Alogorithm that drops the quality down when ingame and benchmarks.
All that says to me is the card falls down hard with Trilinear Anisotropic filtering On and NV have deliberately made it near impossible to test.

If everything were above board there would be 3 modes plus an option that allowed dropping down a mode when required. (if you think agressive is worth anything).
Their overclocking solution is moronic. I doubt very much that your hardened overclocker would even opt for this design. I can just imagine the PR guys telling the engineers "We've found this - now use it and I don't wanna hear excuses".

To me this is just a GF4 with a faster clock speed and then overclocked to hell.

 

[demalion]

I tend to think the DX 9 shader functionality is quite unpolished, look at the "FX and PS 1.4, DX 9 tests" thread.

I don't expect the GF FX to automatically be a clear winner in complex vertex shading, but I also wouldn't expect it to perform relative to the R300 in such a way.

I don't presume other factors of its drivers are necessarily unoptimized, but I hardly expect them to get slower by release (I don't think the fog issues are going to result in a slowdown).

I also tend to believe it likely we won't get slowdowns from changes in Quincunx and 2x AA, but I do think the quality will be more disappointing than I thought (I had thought AA would be gamma-corrected and would tend to compare more favorably to the R300).

I also agree their "adaptive" aniso initiative is clouded by PR, and it is unclear where the "balanced" mode really stands in comparison to the 9700. Best case seems it is trilinear, will lose the "pleasing color mipmap" contest with the 9700, but might not be too bad in terms of actual texture filtering while in motion. Worst case is it is noticably worse than 9700 quality mode, and we'll have to resort to "GF 4 mode" versus "9700 Quality", which was not clear cut in terms of quality, but was in terms of performance...in which case the aniso performance of the card is not going to be pretty.

I disagree somewhat about their overclocking solution. The hardened overclocker would use a waterblock, or some other exotic cooling solution. What is disputably "moronic" is the volume and size of the cooler assembly (though I still see the likelihood of a Zalman heat pipe + quiet fan to address the noise part), not the 300 MHz/500 MHz thing (which I actually sort of like...or would if the cards can handle 400 MHz/500MHz well enough).
I think a marketing push for the Ultra as "pre-overclocked" is a valid strategy to address such people, and I wouldn't blame them for that as long as the MTBF compared to warranty time isn't too ugly or as long as image quality doesn't get worse in the process of driver performance increase in conjunction with that.

 

[3dcgi]

While the 'evolution' of the NV30 make sense, it doesn't make sense that the drivers should be so inmature at this point if they had dieffrent NV30-rev chips to work on for so long. Sure, they were not the same rev, but still. Think about it.

Although if those NV30 revs weren't stable they wouldn't have helped the drivers guys too much.