nvidia 2x aniso artificial limit? (on NV17 etc.)

[Ante P]

I was just curious. Is the limit of 2x Aniso of GF2 and GF4MX etc. an artificial limit (like it is on the Parhelia) or is it simply there because of the performance degredation?

Also: did the GF (256) support Aniso as well?

And one last thing: max aniso level on the Kyro/KyroII?

 

[Thowllly]

I was just curious. Is the limit of 2x Aniso of GF2 and GF4MX etc. an artificial limit (like it is on the Parhelia) or is it simply there because of the performance degredation?

Also: did the GF (256) support Aniso as well?

And one last thing: max aniso level on the Kyro/KyroII?

AFAIK, it's a hardware limit, and the GF 256 could supported it too...

 

[Basic]

In what way is it an artificial limit for Parhelia. As I remember it from the launch, that looked like a rather hard limit.

 

[Ante P]

In what way is it an artificial limit for Parhelia. As I remember it from the launch, that looked like a rather hard limit.

Because according to Matrox own specs Parhelia supports 16x+Bilinear/8x+Trilinear.
Yet in the drivers you can only choose 2x. (+Bilinear/Trilinear)

 

[Hyp-X]

They (Matrox) probably have a hardware bug that prevents them to enable higher levels of aniso.

Btw, G400/G450 supports 4x aniso (and it is working).

 

[Ante P]

They (Matrox) probably have a hardware bug that prevents them to enable higher levels of aniso.

Btw, G400/G450 supports 4x aniso (and it is working).

that also applies for 5x0 too then I suppose?

 

[Hyp-X]

that also applies for 5x0 too then I suppose?

Probably... but I've never seen a G550.

 

[Basic]

I guess you refer to:

64 Super Sample Texture Filtering
* Highest quality trilinear and anisotropic filtering
* Sustained performance
* Dynamic allocation of texture units
* 8-sample anisotropic and trilinear filtering on 4 dual-textured pixels/clock
* 16-sample anisotropic filtering on 4 single-textured pixels/clock

Notice how they put 8-sample aniso and trilinear in the same sentence. They refer to a "sample" as one texture-sample (as opposed to one filtered texture sample). So their 8-sample aniso is likely what most would call 2x bilinear aniso. Whether 16-sample anisotropic is 2x trilinear aniso or 4x bilinear aniso is unclear, but it's most likely one of those two.

 

[Ante P]

I guess you refer to:

64 Super Sample Texture Filtering
* Highest quality trilinear and anisotropic filtering
* Sustained performance
* Dynamic allocation of texture units
* 8-sample anisotropic and trilinear filtering on 4 dual-textured pixels/clock
* 16-sample anisotropic filtering on 4 single-textured pixels/clock

Notice how they put 8-sample aniso and trilinear in the same sentence. They refer to a "sample" as one texture-sample (as opposed to one filtered texture sample). So their 8-sample aniso is likely what most would call 2x bilinear aniso. Whether 16-sample anisotropic is 2x trilinear aniso or 4x bilinear aniso is unclear, but it's most likely one of those two.

On the other hand that doesn't make sense if you consider the title.
8x+Tri is 64 samples
16x+bi is 64 samples
that's how I view it

reviews also mentioned this when the parhelia first arrived

 

[antlers]

On the other hand that doesn't make sense if you consider the title.

Maybe the "64 samples" from the title is the total over four pipes. Then your interpretation would fit.

 

[Basic]

Exactly antlers4. And that's the only way I can see where all numbers add up. 64 samples total over 4 pipes.

4pipes* 2 tex/pipe * 8 samples/tex = 64 samples (trilinear or 8-samp aniso)
4pipes* 1 tex/pipe * 16 samples/tex = 64 samples (16-samp aniso)

If "64 samples" was for each pipe, why can't they do 4 trilinear samples per clock and pipe. They do have four texture units, but can only do four texture samples per clock and pipe with bilinear filtering. Again an indication that it's 64 samples total.

4pipes* 4 tex/pipe * 4 samples/tex = 64 samples (bilinear)