[Murakami]

From a tecnhnical point of view, is it better to have 4 pixel pipes each with one texture units or 2 pixel pipes each with two texture units (assuming the same texel rate, for example 500 megatexel/sec)? And why? Second question: is there a list of games that support bump mapping? In your opinion, which is the game with the best use of DOT3 bump mapping? Thanks all.

[Kristof]

In terms of efficiency 4x1 is better than 2x2, since in the 2x2 design if a polygon is single texture (or tripple, etc) then one texture unit will be unused and wasted. In the 4x1 design all units will always be active no matter the contents of the 3D scene (single, dual, tripple, ...) assuming that multitexture can be used, if the software need to go to multipass geometry then efficiency might fail due to AGP or other geometry related bottlenecks. But generally X by 1 is more efficienct than Y by 2. Things get worse with more textur eunits.

From a silicon area point of view 2 by 2 is better since adding a texture unit is cheaper than adding a full pipeline.

DOT3 Bumpmapping... list... not that I know off. Best game... hard to say... I would say that DoomIII will probabaly be the first game to "really" use per pixel DOT3 effects to the max. Matrox on the other hand probabaly has a list of games using EMBM bumpmapping.

K-

[Murakami]

Originally Posted by Kristof

In terms of efficiency 4x1 is better than 2x2, since in the 2x2 design if a polygon is single texture (or tripple, etc) then one texture unit will be unused and wasted. In the 4x1 design all units will always be active no matter the contents of the 3D scene (single, dual, tripple, ...) assuming that multitexture can be used, if the software need to go to multipass geometry then efficiency might fail due to AGP or other geometry related bottlenecks. But generally X by 1 is more efficienct than Y by 2. Things get worse with more textur eunits.

From a silicon area point of view 2 by 2 is better since adding a texture unit is cheaper than adding a full pipeline.

DOT3 Bumpmapping... list... not that I know off. Best game... hard to say... I would say that DoomIII will probabaly be the first game to "really" use per pixel DOT3 effects to the max. Matrox on the other hand probabaly has a list of games using EMBM bumpmapping.

K-

Thanks Kristof, in fact a GeForce2 MX is only marginally faster than a GeForce 1 SDR, even if the texel rates are not marginally different (700 vs. 480). Just a curiosity: if multitexturing is used, are always there some poligons single textured? Thanks all.

[Murakami]

Originally Posted by Murakami

Thanks Kristof, in fact a GeForce2 MX is only marginally faster than a GeForce 1 SDR, even if the texel rates are not marginally different (700 vs. 480). Just a curiosity: if multitexturing is used, are always there some poligons single textured? Thanks all.

..hmm.. even if the performance of two chips are limited by the bandwidth of the SDR memory.. invalid comparison?

[Kristof]

Yes, not every effect requires multiple texture layer, there will almost almost be some objects or effects which only require a single texture layer. For example text, HUD, explosions, alpha effects, etc.

Memory bandwidth always has an impact; if you want to take this into account then outputting more pixels is probably more expensive than grabbing some more texels (texels can usually benefit from a good cache) using a second texture. But its hard to say since bandwidth is really quite complex to predict or analyze.

K-

[3dcgi]

There are also many cases where no textures are used. Rendering shadow volumes is one example. In this case a 4x1 architecture is much more efficient.

[Maverick]

And then again, if you're using multiple textures, but doing some pixel shader calculations in between, a second TMU isn't of much benefit, because by the time the next texture fetch is needed, the first TMU will be free...

[Murakami]

Well.. it looks like GS (PS2) architecture is the best..

[Glonk]

Originally Posted by Murakami

Well.. it looks like GS (PS2) architecture is the best..

[3dcgi]

Originally Posted by Murakami

Well.. it looks like GS (PS2) architecture is the best..

It seems to me that the Radeon's 8x1 is pretty darn good. The PS2 a 4x1?

[Murakami]

Originally Posted by 3dcgi

It seems to me that the Radeon's 8x1 is pretty darn good. The PS2 a 4x1?

PS2 is 16x1

[alexsok]

PS2 is 16x1

What is better: 16x1 or 8x2?

[Thowllly]

Originally Posted by alexsok

What is better: 16x1 or 8x2?

16x1 should be better if it can do loopback or pipecombining, something the PS2 can't do. And the PS2 is actually 16x0/8x1 (16 pixel per clock with no texture or 8 pixel per clock with 1 texture)

[alexsok]

16x1 should be better if it can do loopback or pipecombining, something the PS2 can't do. And the PS2 is actually 16x0/8x1 (16 pixel per clock with no texture or 8 pixel per clock with 1 texture)

Thx m8!

[ram]

Originally Posted by Kristof

In terms of efficiency 4x1 is better than 2x2, since in the 2x2 design if a polygon is single texture (or tripple, etc) then one texture unit will be unused and wasted.

Agreed, but on the other hand, an 8 pipeline architecture can work less efficient for very small triangles in some situations than a 4 pipeline architecture. Assuming a bunch of semi-transparent triangles of size 3 pixels coming behind each other at almost the same location on screen (used for a neat explosion effect or whatever). This situation stalls several pipelines in the 8x architecture.

[Chalnoth]

Originally Posted by Murakami

Well.. it looks like GS (PS2) architecture is the best..

That taking it way out of context.

If you're attempting to make a comparison between, say, 8x2 and 16x1, and you want to ask which is better, you have to take it in the context that everything else is the same.

Yes, everything else remaining the same, the 16x1 architecture will be more powerful.

But the fact is that the PS2 architecture is incredibly limited in other areas that really prevent it from competing with even GeForce3-level hardware, if not earlier (theoretically, anyway...it's not like you can directly compare them...).

[Murakami]

Originally Posted by Chalnoth

That taking it way out of context.

If you're attempting to make a comparison between, say, 8x2 and 16x1, and you want to ask which is better, you have to take it in the context that everything else is the same.

Yes, everything else remaining the same, the 16x1 architecture will be more powerful.

But the fact is that the PS2 architecture is incredibly limited in other areas that really prevent it from competing with even GeForce3-level hardware, if not earlier (theoretically, anyway...it's not like you can directly compare them...).

I agree with you, i said "architecture" only referring to pixel pipes, not entire machine..thanks.

[Murakami]

Originally Posted by Thowllly

16x1 should be better if it can do loopback or pipecombining, something the PS2 can't do. And the PS2 is actually 16x0/8x1 (16 pixel per clock with no texture or 8 pixel per clock with 1 texture)

Are you sure that PS2 -as many times many people said- can't do loopback? Is there, on the entire web, a single article, in depth view, technical brief or product overview about GS, in any kind comparable to the article about EE that appears some time ago in "Ars Technica"? Thanks all.

[alexsok]

the article about EE that appears some time ago in "Ars Technica"?

Do you have a link to that EE article or did it appear in a magazine?

[Murakami]

Originally Posted by alexsok

Do you have a link to that EE article or did it appear in a magazine?

http://arstechnica.com/paedia/3dtech.html

[alexsok]

Thx m8!

I realized I read it already a long time ago!