NVIDIA 60.72 Fiasco

[Ardrid]

Looks like the drivers are in the clear after all:

http://www.techreport.com/onearticle.x/6668

 

[Joe DeFuria]

Well, most of us gathered something like that was the case. Good to hear a more official explanation though. Thanks for the link.

 

[trinibwoy]

"The DX9 reference rasterizer does not produce an ideal result for level-of-detail computation for isotropic filtering, the algorithm used in NV40 produces a higher quality result.

It was still funny to see all those headhunters go for Nvidia's throat at the whisper of a possible controversy.....

Don't give up fellas...Nvidia might have some cheats up it's sleeve still for you to scream about.

 

[Pete]

This was a fiasco, now? Did I miss the original ruckus, or was the brou-haha only in a few hyperbolic minds?

I call shenanigans!

 

[Ante P]

I don't really get how harsher mip map boundaries spells higher image quality though..?

 

[Ardrid]

This was a fiasco, now? Did I miss the original ruckus, or was the brou-haha only in certain people's hyperbolic minds?

I call shenanigans!

It was the only word I could think of at the time

 

[Eronarn]

Don't give up fellas...Nvidia might have some cheats up it's sleeve still for you to scream about.

61.11, eh? 8)

 

[trinibwoy]

61.11, eh? 8)

i'm not sure but has nvidia announced the driver version that will be made availabe upon launch?

 

[KimB]

i'm not sure but has nvidia announced the driver version that will be made availabe upon launch?

They never have and never will. These things are pretty fluid.

 

[Ante P]

61.11, eh? 8)

i'm not sure but has nvidia announced the driver version that will be made availabe upon launch?

I just saw a notice about how 60.8x was being WHQL'ed for the 6800 release

 

[Miksu]

Anyone know if 60.72 still has PS2.0-test optimizations enabled for 6800U, in 3DMark'03? I know that Futuremark has approved the drivers for 6800U but that doesn't mean that there aren't any optimizations in this one test, right?

 

[Ante P]

Anyone know if 60.72 still has PS2.0-test optimizations enabled for 6800U, in 3DMark'03? I know that Futuremark has approved the drivers for 6800U but that doesn't mean that there aren't any optimizations in this one test, right?

futuremark seems to only validate drivers for the "game tests" yeah

 

[Evildeus]

the algorithm used in NV40 produces a higher quality result.

Now, that's really great. Are we going to see some good article from DH saying that the NV40 produce the best output

 

[vb]

are you sure their name isn't 64.20?

Maybe 64.23 with WHQL.

 

[bloodbob]

Ehh hmm I wonder if Nvidia will give out this algorithim or if it will remain a trade seceret? frankly it looks like just a LOD bais adjustment not some new way to calculate which level of detail texture to use.

 

[Bambers]

I don't really get how harsher mip map boundaries spells higher image quality though..?

Me neither. I've been wondering about this for a while. What nvidia has done is changed the filtering so it is closer to a perfect circle on the AF tunnel test.

What im wondering is whether a circle is actually ideal as fiddling around on that test shoes that aliasing appears closer on vertical/horizontal surfaces than it does on 45° ones, the more usual method seems to fit pretty well actually.

 

[KimB]

Ehh hmm I wonder if Nvidia will give out this algorithim or if it will remain a trade seceret? frankly it looks like just a LOD bais adjustment not some new way to calculate which level of detail texture to use.

There's nothing "new" or "secret" about it. It just uses a more correct algorithm (for the non-anisotropic tests, anyway). Just look at the tunnel tests:

http://www20.tomshardware.com/graphic/20040414/geforce_6800-44.html

You'll notice that it's more like a circle, as it should be.

There is a well-defined function that is to be used for MIP mapping, and that's the best approximation of that function we've yet seen.

 

[Xmas]

I don't really get how harsher mip map boundaries spells higher image quality though..?

Harsher?

A comment about correctness: The OpenGL spec contains a formula to calculate texture LOD, and an "approximation". Demirug implemented both as PS in his D3D AF tester, and noticed that the correct formula results in circles, while the approximation looks like NV3x/R3x0.

However, I'm not entirely convinced this formula is really correct.

 

[Snarfy]

Remember quack 3?

If you liked that, you're gunna LOVE this

renaming farcry.exe to fartcry.exe caused a huge performance drop in 61.11 drivers...

 

[KimB]

Complain if it's there when the boards are released.