Excellent FX Review

[LeStoffer]

I don't know why any hardware that is designed to be quick at PS2.0 should be too slow on PS1.4 to gain an advantage from using it over multipassing earlier shader models.

I'm with you, but nonetheless GF FX - at least with the current drivers - just loves PS 1.1 and dislikes PS 1.4.

Right now the register combiners path are plenty fast, but things are still twichy in the PS 1.4 - PS 2.0+ pixel shader/fragment program path. So the question is whether GF FX was designed to be quick at PS 2.0 without the functional help of register combiners?

Edit: Damn, I didn't realize that Andy is from ATI.

 

[THe_KELRaTH]

NVIDIA 6xSAA vs ATi 6xAA + Anistropic

This result was perhaps the biggest surprise and may just show the advantage ATi have with their reduced colour precision and increased memory bandwidth

I didnt realise that 12bit Integer was now higher than 24bit fp - is this another case of Nvidia rewriting the standards

The F1 2002 and CFS 3 AF comparision images really do show how bad Nvidia's AF implementation is. Although I knew it was far from equal I hadn't realised just how bad Nvidia's AF was till I saw those pics, even Balanced doesn't compete with ATI's performance mode.

Any ideas as to why the 9700Pro takes such a high fps hit with x2AA x2AF in F1-2002? 86 down to 28 seems like rather a large performance hit!

 

[Nagorak]

This was an "ok" review, but it really needed at least a cursory editing run. In addition to some very poor wording in places, I counted several run-on sentences throughout the review. I know, grammar isn't everything, and the author does manage to get the point across, but these problems still manage to distract from the quality of the review. A review doesn't have to be Shakespere, but having obvious errors like that makes it seem like either the review was rushed, or the author doesn't take pride in his work (even if it is just a hobby).

 

[Xmas]

Is that "best performance" LOD-setting really default in latest drivers? Has never been. If not, why did they even use it? "Best quality" has been standard in every Detonator I've used so far, and there's no reason to change that setting.

 

[Reverend]

I thought the review was a nice read overall although I almost stopped reading after the first page (the tables comparing the R300 and NV30 wrt PS/VS... incorrect info there, which means he should've done some investigating instead of using what NV gave them).

 

[andypski]

I don't know why any hardware that is designed to be quick at PS2.0 should be too slow on PS1.4 to gain an advantage from using it over multipassing earlier shader models.

I'm with you, but nonetheless GF FX - at least with the current drivers - just loves PS 1.1 and dislikes PS 1.4.

Right now the register combiners path are plenty fast, but things are still twichy in the PS 1.4 - PS 2.0+ pixel shader/fragment program path. So the question is whether GF FX was designed to be quick at PS 2.0 without the functional help of register combiners?

Edit: Damn, I didn't realize that Andy is from ATI.

Sorry - I thought everyone knew that...

Anyway, the point that I was trying to get at is that 1.4 and 2.0 should be sufficiently similar in performance characteristics that an architecture designed to be fast at 2.0 should be fast at 1.4. The only thing I can think of that might impede performance is emulating some 1.4 modifier operations which don't exist in 2.0

nVidia may now be doing some optimisations to get parallelism out of their register combiners (which as you note seem to be fast) to improve speed while still running the 1.4 path. Only they know for sure.

- Andy.

 

[rhink]

One thing that struck me about this review.... why the heck was he screwing around with the mipmap LOD? That's all the "mipmap detail" setting he was talking about does.... no shit, if you crank the LOD to +2 it's gonna look awful. I haven't found many settings where it affects performance a ton anyways.

 

[tEd]

I would like to know if balanced does alter the mipmap details. Dave?

 

[Pete]

Thanks for bringing this great review to our attention, Typedef.

tEd, Dave's day-old FX:U preview doesn't answer your question with its Serious Sam screenshots?

 

[tEd]

Thanks for bringing this great review to our attention, Typedef.

tEd, Dave's day-old FX:U preview doesn't answer your question with its Serious Sam screenshots?

Before i read the 3dvelocity review i thought the difference between "application,balanced and aggressive" setting lies solely in the AF algorithm but i'm not so sure anymore , that's why i ask.

 

[Xmas]

Before i read the 3dvelocity review i thought the difference between "application,balanced and aggressive" setting lies solely in the AF algorithm but i'm not so sure anymore , that's why i ask.

It affects the AF algorithm as well as how trilinear filtering is performed (aggressive is 'almost bilinear'). Changing the AF algorithm implicitly means changing the LOD selection. However, it is not like a global LOD bias/shift.