Matrox Parhelia

[GS-1719]

John Carmack's latest plan:

The Matrox Parhelia Report:

The executive summary is that the Parhelia will run Doom, but it is not performance competitive with Nvidia or ATI.

Driver issue remain, so it is not perfect yet, but I am confident that Matrox will resolve them.

The performance was really disappointing for the first 256 bit DDR card. I tried to set up a "poster child" case that would stress the memory subsystem above and beyond any driver or triangle level inefficiencies, but I was unable to get it to ever approach the performance of a GF4.

The basic hardware support is good, with fragment flexibility better than GF4 (but not as good as ATI 8500), but it just doesn't keep up in raw performance. With a die shrink, this chip could probably be a contender, but there are probably going to be other chips out by then that will completely eclipse this generation of products.

None of the special features will be really useful for Doom:

The 10 bit color framebuffer is nice, but Doom needs more than 2 bits of destination alpha when a card only has four texture units, so we can't use it.

Anti aliasing features are nice, but it isn't all that fast in minimum feature mode, so nobody is going to be turning on AA. The same goes for "surround gaming". While the framerate wouldn't be 1/3 the base, it would still probably be cut in half.

Displacement mapping. Sigh. I am disappointed that the industry is still pursuing any quad based approaches. Haven't we learned from the stellar success of 3DO, Saturn, and NV1 that quads really suck? In any case, we can't use any geometry amplification scheme (including ATI's truform) in conjunction with stencil shadow volumes.

[Lord FlatHead]

Cool. Next time though you might want to mention the source.

[Unbreakable]

Lord FlatHead said:

Cool. Next time though you might want to mention the source.

Bluesnews, voodooextreme and shacknews didnt put the source down, as its e-mailed to all those sites from id.

So there is no actual source, only id.

[Naked Snake]

/me shrugs
Lucky me, I have a GF4 ^_^

[DooMBoy]

Is any of this stuff posted by Carmack supposed to make sense?

[Livo]

Basically, the Parhelia's hyped and allegedly "innovative" features are of no use to him in Doom 3, nor does there seem to be much prospect of their use in future engines. He also is disgusted with the performance compared to other cards like the Geforce Ti 4600 and Radeon 8500, especially when it was supposed to be a revolutionary card.

[Dima]

Livo said:

Basically, the Parhelia's hyped and allegedly "innovative" features are of no use to him in Doom 3, nor does there seem to be much prospect of their use in future engines. He also is disgusted with the performance compared to other cards like the Geforce Ti 4600 and Radeon 8500, especially when it was supposed to be a revolutionary card.

It never was supposed to be a revolutionary card.

[Naked Snake]

DooMBoy said:

Is any of this stuff posted by Carmack supposed to make sense?

It does if you read it hard and guess what he means by certain things.

[gatewatcher]

Carmack is saying that if you are buying a new card for Doom3, don't buy this one.

[EsH]

Personally, I think this card has a lot of interesting features, the foremost being displacement mapping. Carmack is a brilliant person, but I really don't agree with his rather over-dedication to squashing hardware accelerated parametric surfaces.

Parametric surfaces are the proper evolution of traditional polygonal geometric art-- his observation that you "can't tell the difference" really doesn't hold water for anything but fps games where you don't have close, sustained proximity to characters and other objects. For games that rely more on drama and human interaction, this subtle increase in quality might be crucial. It's just silly to have high res and low res models just to avoid angular artifacts, when what we really want is smooth, general curves where the explicitly specified geometric detail gives out. No, this isn't a perfect solution, but it's MUCH cleaner and belivable looking that seeing sharp angles caused by roughly approximated curves.

In fact, Carmacks bump-mapping technique is really in concept a type of parametric surface-- a lot of the bumpmapping just goes into making the low poly models appear less angular. But his method fails when you get really close, or view a polygon from a shallow angle. It just looks flat and the realism is ruined.

His other argument that you could have more detail if you went in and explicitly specified the geometry in the model is also flawed. Yes, you could, but in a huge number of cases, this is just waste. Not only do you have the memory overhead of explicitly specifying stuff that is really just slightly displaced tesselation of a curve, but the card can't perform any optimizations that might result from generating the vertices itself. With paramtric surfaces, we'll eventually get per-fragment geometry, just like in off-line rendered stuff. You CAN'T do per-framgment polygons with explicitly specified geometry, because no matter how much detail you put in you can always move the view closer to the object. One might argue that there's really no need for getting within a certain distance of most objects, but when you consider things like leaves, grass, flowers-- these objects can fill a fields where the models are both less than the size of a pixel, and one petal fills the screen. With characters, we might want to zoom in on the characters mouth or eyes-- somthing you just can't do naturally well, even with Doom3 technology unless you create excessivley complicated, special models. With things like this, the explicit detail is less important than the un-obtrusiveness of the primitives that make up the model.

We need paramtric surfaces, with displacement mapping, and it really, really bothers me that Carmack is trying to help kill this sensible technology off.

[GiXeLz]

Hrm..

Parhelia's displacement mapping is just a hardware driven geometrical mipmapping setup. It actually adds the detail to the geometry.

The point of Dot3 bumpmapping is to emulate the addition of detail, because it's cheaper to do that than to actually add the detail. Much cheaper.

"In any case, we can't use any geometry amplification scheme (including ATI's truform) in conjunction with stencil shadow volumes."

[Dima]

"In any case, we can't use any geometry amplification scheme (including ATI's truform) in conjunction with stencil shadow volumes."

Exactly!

I fully agree with Carmack on this one, but EsH also brings some very valueable points...

I think Carmack doesn't need diaplacement mapping cause it would be unneeded for FPS games like DOOM III where most of the action (if not all) is happening inside and not outside (where this technology would unleash it's full power as far as I know).

Perhaps it could be useful for certain things, but it's definetly not needed for DOOM III and I can see why.

[Ultimate Demon]

Why buy a Geforce4 when ATI is soon gonna release it's new card?

[Wobbo]

Its an example of FOCUSING on what he can do and what the game will be. Doom will not be outdoors or well lit, and even though we will be right up close to some textures for a while, its still at heart an action game.

hes made about faces before, meybe next engine

[EsH]

I think Carmack doesn't need diaplacement mapping cause it would be unneeded for FPS games like DOOM III where most of the action (if not all) is happening inside and not outside (where this technology would unleash it's full power as far as I know).

Perhaps it could be useful for certain things, but it's definetly not needed for DOOM III and I can see why.

Carmack designed Doom3's graphics engine to what he felt is the optimal performance/quality spot for the mid-line of hardware. New features and performance optimizations come out so fast now that it's impossible to use the newest cards that come out to there fullest ability. It's not "needed" for Doom3 simply because the game was coded around an earlier technology base. It doesn't mean the new technology is conceptually wrong.

Outside vs. inside environments isn't the issue I'm talking about. When I mean tesselation and displacement of curved surfaces, I mean down to sub-fragment levels for an overal increase in the quality of all geometry (not just landscapes). This is how geometry is handled in offline CG productions, for the most part. Parahelia uses landscapes as an example, but this is because the technology is still quite primitive and can't produce really hardcore tesselation densities. This doesn't mean it's a useless technology, only that it needs to be nurtured (much like the crappy embossed bump-mapping developed into the dot3 bumpmapping Carmack uses).

Again, sorry for the rant, but this is a really sore point for me.

[Wobbo]

by outside areas she means places where many objects will be in view at once, i dont think dima was only talking about landscapes, just where dynamic tesselation would be useful in general - in very large open spaces

as you zoom very far away, the technology would be quite useful, but in dooms lower light and lower distance environments its not a big deal

[EsH]

Oh, cool! Carmack updated his .plan file, and I totally agree with him now.

BTW, open areas would be much less useful for hardware tessealtion than the closed areas of Doom3. The whole point here is to eliminate the angular edge of polygons. In open areas, most of the geometry is distant, and the original polygons converge to pixel-sized dimentions anyway, totally making tesseslation of those polygons pointless. But in the closed environment, such as Doom3, the polygons are really close, and each take up hundreds of pixels on screen. It's here that fine hardware tesseslation would shine.

Once again, I'm sorry about the rant, and promise never to use the words "hardware tessellations" unless asked to. For the rest of the day.

[Zoost]

Esh, do you have the stutter, and want this to reflect in the digital world?

[EsH]

Zoost said:

Esh, do you have the stutter, and want this to reflect in the digital world?

?

[Alientank]

The answer to your video card needs are here...http://www.3dcenter.de/artikel/2002/05-22_a.php

[Darkhound]

Ultimate Demon said:

Why buy a Geforce4 when ATI is soon gonna release it's new card?

Why buy the new ATi card when nVIDIA will probably release geforce 5 soon etc etc, with that attitude you would never upgrade plus the gf4ti's are excellent cards and if the radeon 8500 is any indication, i wouldnt be to excited, it just cant keep up with the ti4600