P. anisotropic filtering

[drslump]

from http://www.matrox.com/mga/products/p...upersample.cfm

Parhelia-512 integrates the world's most advanced texture filtering units allowing for the dynamic allocation of up to 64 texture samples per clock—double the number available on competing GPUs. These samples can be flexibly allocated to provide higher quality texture filtering with minimal performance impact. For example, the 64 samples could be arranged to deliver:


Dual-textured pixels with trilinear filtering at the same performance as dual-textured pixels with bilinear filtering
Dual-textured pixels with 16-sample anisotropic filtering at comparable performance to dual-textured pixels with trilinear filtering on competing GPUs
With 64 Super Sample Texture Filtering, Parhelia-512 is able to offer a higher quality 3D experience without the performance penalty suffered on other GPUs.

does it mean that parhelia can do 16x aniso filtering at good frame-rates? or are they speaking about 2x anisotropic?

I know the drivers are limited to 2x, is there any hack to change that?

cheers, Ivan

[TdB]

I always thought it meant 16 samples for bilinear filtering and 8 samples with trillinear filtering.

don´t know about the "good framerates" though

it would be interesting to see in action, but matrox decided to disable it in the drivers completely, and i don´t think a simple regestry-hack is enough.

why can´t we (the users) decide when we should use it!

[Chrono_Wanderer]

i wonder if there is actually some hardware anisotropic filtering issue in the core...

[[GDI]Raptor]

They have said that they will fix it in a new driver release, but it wouldn't suprise me if it's a hardware problem that gives them problems runing anistropic filtering in high levels....

The bigest problem with the Parhelia preformance wise is a to low core speed holding it back.... Imagine what a Parhelia @ 325 - 350 Mhz core could do....

[P5ycho]

Originally posted by [GDI]Raptor
Imagine what a Parhelia @ 325 - 350 Mhz core could do....

I've been imagining for more than 3 years now... time to move on.

[drslump]

I've been imagining for more than 3 years now... time to move on.

I agree.

They have a buggy product so the normal thing should be to warn their costumers. They could just send a newsletter or put a "knowledge database" or something like that.
The last driver release didn't even include a whatsnew.txt!
Hell, I don't even know what are the real caps of this card.
At least the latest drivers seems to fix some bugs (not the important one anyway ) but their releases are becoming more and more slower.
Actually I wouldn't be surprised if matrox droped their consumer graphics division and concentrate in the medical image/video editing markets.
They lost the battle for the fastest GPU and now their are losing the one for DX9 compatibility. If I'm not mistaken ATI and nVidea have already DX9 drivers. Not that anyone needs them but it'll be a nice extra logo to put in the specs.

[K6-III]

The release notes were posted a bit after the drivers were released.

Apparantly the next driver release should give us DX9 compatability.

Seeing as Matrox apparantly gives us drivers on Fridays, those are the days to watch for them....

[chhfchhf]

i like super sample than multy sample
multy sample is too cost resource to operation
super sample need large fill rate,Parhelia can operate is properly

[Indiana]

This thread is about anisotropic filtering, not FSAA (those different techniques like multisampling and supersampling you mentioned refer to FSAA).
And Matrox doesn't need multisampling FSAA since they have the IMO superior (from a sheer technical and bandwidth/fillrate aware standpoint) FAA. Unfortunately at least the current algorithm has some "peculiarities".

To the topic: I think they have hardware-related issues with higher anisotropic levels as well - otherwise, they would've put it into the drivers (as they earlier suggested) by now.
Sometimes you just have to make some compromises in development, you just don't have infinite transistors, etc. that will fit on a chip in the respective manufacturing process - at least at a reasonable cost.
Just think of the GF4 that is slower in anisotropic filtering (slower in the meaning of takes an even higher performance-hit) than the GF3, apparently due to limitations of the hardware. The GF4 also seems to lack large parts of the earlyZ removal that already were present in the GF3 and are present in the GF4MX (can be proven by easy front-to-back and back-to-front rendering benches).
Or think at the Truform hardware support that obviously has been left out of the R300, resulting in it giving lower fps than the R200 in quite some cases when TruForm is activated.

So what use would it be if you could enable higher anisotropic levels on the Parhelia but the speed would slow down to a slideshow? Only result of this would be lots of complaints about bad performance, so I think from a company standpoint it's wise to leave it out. Be sure if it was possible at decent performance levels it would be integrated in the drivers (those driver and hardware folks are not totally stupid, you know).

[3dfx]

I completely agree with Indiana's statemeant almost took the words right out of my mouth ..lol I mean the Parhelia is a great card, with lots of inovinative features like FAA 16x and triple head, but the anistropic is imo abysmal. That is one of the main reasons I bought a Radeon 9700 for gaming. I like the Parhelia and it's great features but I really like the filtering methods of the Radeon. I mean the is practically no performance hit and it goes all the way up to 16x filtering and the parhelia on does 2x which is the equivalent of an original Geforce whihc is nothing to be proud of as it still takes a massive hit in performance. I hope Matrox can fix it