Parhelia's LOD and anisoquality

[nggalai]

Hi there,

Originally posted by mikeul
Notice they only wrote "consider".
I believe it was an attempt of Matrox to trick reviewers that their anisotropic filtering is as fast as competitors [/B]

Unfortunately, the highest setting on the Parhelia-512 GPU (trilinear 8° anisotropic filtering) needs all four TMU of all pipes to render ONE pixel. Bye, bye, fillrate.

It still would have rocked to see how it copes with 4°AF, though . . .

ta,
-Sascha.rb

[nggalai]

Addition to the anisotropic filtering debate:

I double-checked with Leo, the reviewer who wrote the Parhelia first-look for 3DC. Matrox Germany told him that support for higher degrees of anisotropic filtering was disabled due to a bug, and not ommitted for performance reasons . . . this seems quite a different reply than Tech-Report got from their Matrox contact.

Matrox, just a hint--even multinationals can benefit from a unified PR strategy . . .

ta,
-Sascha.rb

[Marshmallowman]

The Card been released for hours and your hammering matrox about fixing bugs....Jeeze can you lend them your time machine

[nggalai]

Hi Marshmallowman,

I am not "hammering" them about fixing bugs. But the anisotropic debate is crucial for me. Tech-Report's Matrox contact claims that any form of AF > 2° was disabled because it was running too slowly, and that Matrox is CONSIDERING enabling it in future drivers.

Our Matrox contact claims disabling 4° and 8° anisotropic filtering is a driver bug which will be fixed in future drivers.

I am a sucker for texture quality; should the Parhelia not be able to deliver at least 4°AF, the board is completely uninteresting for me. Hence I want to know what Matrox contact was talking out of his/her ass, and what's the status quo regarding anisotropic filtering on the Parhelia.

Sorry if I came over all whiny and bitchy. I am not slapping Matrox for the current driver state. Actually, the reviews were everything I hoped for from Matrox, and the Parhelia is still occupying first slot on my shopping list. Driver issues at the launch of a first-generation board are completely normal and to be expected.

ta,
-Sascha.rb

[mikeul]

"Unfortunately, the highest setting on the Parhelia-512 GPU (trilinear 8° anisotropic filtering) needs all four TMU of all pipes to render ONE pixel. Bye, bye, fillrate. "

Yes, but I thought the specs of the cards (high bandwith, etc.) were enough to cope with this, as GF4 can.

[EvilDonnyboy]

Bandwidth isn't the performance problem with Parhelia. It's it's low clockspeed and lack of double texturing fillrate.