Nice Q&A with matrox

[muted]

http://www.gamepc.com/reviews/hardware_review.asp?review=parheliap&page=7&mscssid=&tp=

Matrox's 16x Fragment Anti-Aliasing seems like an extremely smart and efficient way of giving users FSAA without a large performance hit. The technical specs say that only 5-10% of the screen's triangles will need to be anti-aliased, will we see little to no hit in performance with Fragment FSAA enabled?
That is correct Chris, however, don't confuse FSAA with FAA 16x. With FAA 16x we are only performing AA on the edge pixels, therefore not the whole scene. Edge pixels typically occupy only a minor percentage of your screen pixel coverage and therefore allow us to perform things like FAA-16x across three displays. Additionally, you don't get the texture smearing result that FSAA incurs because with FAA-16x you aren't even touching those internal textures. The result is 16x AA quality with internal texture precision of the original scene, and even better if you enable aniso, which our pipeline can do for free.

i like that , part.. aniso for free ..

 

[Kristof]

He is talking about 8 or 16 tap (tap as in sample) anisotropic being free, its all about samples. Their architecture can take 64 samples in total, thats 16 samples per pipe over 4 TMUs thats 4 samples (bilin) per TMU. This translates that 8 tap anisotropic is "free" if you only do dual texturing (8 samples consumes 2 TMUs) and 16 tap anisotropic is free if you are doing "single" texturing. An adapative technique would probably have been more interesting where you take samples as is needed...

K-

 

[PeterT]

But if you read this it seems that he means that aniso will only be "free" in 2-texture environments where the other 2 texture units could be used for better filtering. At least, that's what I percieve to be the case.

The thrust behind Parhelia-512 was to offer a very high quality graphics card with tons of features that will provide more than adequate performance when the quality is turned up. Where the competition will drop off considerably, the Parhelia-512 will keep chugging along happily due to our deep shader and texture pipeline. But we are still a four pixel architecture and as such when running simple dual-textured benchmarks and games without anisotropic filtering enabled, our card won't be working very hard (you get aniso for free basically with Parhelia).

[EDIT] Lol, Kristof was a bit faster [/EDIT]

 

[Hellbinder]

The problem i see is that

1. 8 Tap (sample) ansio is a JOKE. and has almost no impact in IQ

2. Almost NO modern games use only 2 texture layers. most games have 3 and 4 in some cases.

I just dont see where this is going to be that impressive. It will be a little faster than the GF4 (in some cases) Still no match for ATi's Rip-mapping technique. Even though it has a coupple holes, It does have a great overall effect on IQ and is VERY playable in 16x mode.

Just my opinion

 

[arjan de lumens]

Hmm - to me, it seems that the speed benefit of Parhelia over GF4 would be larger the more textures and the more aniso-taps per texture you make the pipeline work with, just because of all the additional texture units. If you use e.g. 4 textures with 16-tap anisotropic mapping on all of them, one Geforce4 pipeline would take 8 cycles while one Parhelia pipeline would take only 4. Whether or not the performance can beat whatever method ATI is using is not clear, though.

 

[nggalai]

It's sad how Sebastian reiterates the "FSAA smears textures" hogwash. As with the press material, Matrox seems determined to reduce the world into two camps:

Supersampling vs/ Quincunx-style Multisampling.

The free aniso bit is rather misleading, too--as Hellbinder mentioned, this won't be free at all if your engine employs more than 2 texture layers. In the light of the strong focus on "4 textures per pass" of the marketing papers, this Q&A reads rather ironic.

ta,
.rb

 

[Typedef Enum]

I find it rather laughable that anybody would judge this card at _all_ without having seen:

a. performance results
b. image quality results (non Matrox generated screenshots)
c. screenshots of leading games enabling IQ effects + performance results.

What's even funnier is trying to compare this product to the likes of R300/NV30...

How the hell do you compare something...which hasn't even been benchmarked...to something that hasn't even been launched?

 

[Kristof]

Typedef Enum, I agree but we do like to punch holes in Marketin statements and claims. The comparisons with "Future Unknown Product X" are indeed a bit of a waste of time... no product will win the battle against something people dream up ! Then again thats what NVIDIA/ATI are counting on... fan boys being convinced that whatever is next from them will beat this Matrox thing.

 

[Basic]

I think it's quite interesting that this super image quality card from Matrox doesn't even seem to be capable of doing more than an anisotropy of 4 with bilinear filtering and 2 with trilinear filtering. And that's when single textureing. When dual texturing, you'd have to divide those numbers by two.

Can anybody point to info at matrox.com that says something else?

 

[Humus]

]The problem i see is that

1. 8 Tap (sample) ansio is a JOKE. and has almost no impact in IQ

2. Almost NO modern games use only 2 texture layers. most games have 3 and 4 in some cases.

I just dont see where this is going to be that impressive. It will be a little faster than the GF4 (in some cases) Still no match for ATi's Rip-mapping technique. Even though it has a coupple holes, It does have a great overall effect on IQ and is VERY playable in 16x mode.

Just my opinion

Two comments:
I don't think there's a whole lot of games that uses more than 2 simultaneous textures.
ATi doesn't use ripmapping.

I'm with Typedef's on this one. I'm waiting until I've seen real reviews before I judge it.

 

[Nappe1]

I am "want some tests first" line, before saying how fast this thing really is. It is surely somewhere in GF4Ti performance and in most cases above that.

but Predicting how those features work in real world aplications, is just too hard, before we have some tests.

but to make some things clear, as ex-Matrox user, this chip has convienced me so well that I'm back to matrox. AIW Radeon is getting old and it looks like that Parhelia is best bet for me.

Still, I wouldn't say this as perfect chip, but it is enough for me.