RSX: Vertex input limited? *FKATCT

[Neb]

You sure? I read a spanish interview, and I also read a translation of said interview, was that what you read? or was it something else?

LINK:
spanish interview thread

"el nivel del agua con el jetski es de los mas grandes y tiene unos 6 millones de poligonos mas el reflejo en tiempo real de los modelos en el agua con bump mapping, refraccion, jitter, sombras, etc........aparte de personajes, fisicas, AI, explosiones volumetricas, bla bla bla smile_.......END"

I translated it!

And the level with water and jetski is one of the biggest levels and contains 6 million polygons (me: means the whole level otherwise it is called per frame) , real-time reflections of the models with bump mapping on the water, refractions, jitter, shadows, etc.... apart from personalities, physics, AI, explosions, volumetrics, bla bla bla smile_.......END

 

[....]

I translated it!

I'll accept that interpretation, but without the clarification that sentence is a bit unclear, imo.

That's because I believe 'has' could also be used instead of 'contains', in the translation, giving:

And the level with water and jetski is one of the biggest levels and has 6 million polygons , real-time reflections of the models with bump mapping on the water, refractions, jitter, shadows, etc.... apart from personalities, physics, AI, explosions, volumetrics, bla bla bla smile.......END

 

[Geo]

AAAIIIEEEE ZOMBIE THREAD OF THE UNDEAD.

Really, anybody got any new FACTS here?

 

[Betanumerical]

AAAIIIEEEE ZOMBIE THREAD OF THE UNDEAD.

Really, anybody got any new FACTS here?

Nope, but I do have a question

Any know the performance of the SPU's at doing vertex stuff (skinning, etc) ?.

 

[Frank]

Vertex stuff is easy, but textures (skinning) require you to think the other way around. If you do, both can be very fast.

 

[nAo]

Vertex stuff is easy, but textures (skinning) require you to think the other way around. If you do, both can be very fast.

Why does skinning require textures?

 

[assen]

Nope, but I do have a question

Any know the performance of the SPU's at doing vertex stuff (skinning, etc) ?.

The one estimate I've heard is by IronPeter, one of the guys behind the RSX unlock (which has sadly, but expectedly, been patched with the latest firmware). He estimated one SPU to slightly exceed the vertex performance of a 6600 for a workload consisting of skinning and backface culling. A 660 has 3 vertex units, RSX/7800s have 8. This is in line with the number thrown earlier in this thread ("2 or 3 SPUs dedicated to geometry processing").

 

[nAo]

The one estimate I've heard is by IronPeter, one of the guys behind the RSX unlock (which has sadly, but expectedly, been patched with the latest firmware). He estimated one SPU to slightly exceed the vertex performance of a 6600 for a workload consisting of skinning and backface culling. A 660 has 3 vertex units, RSX/7800s have 8. This is in line with the number thrown earlier in this thread ("2 or 3 SPUs dedicated to geometry processing").

And it's not even a completely fair comparison as vertex shaders don't perform any culling.
In this case is fairly correct to simply expect SPUs to match clock per clock modern vertex shader implementations.
As we know SPUs run at 3.2 Ghz..

 

[Betanumerical]

And it's not even a completely fair comparison as vertex shaders don't perform any culling.
In this case is fairly correct to simply expect SPUs to match clock per clock modern vertex shader implementations.
As we know SPUs run at 3.2 Ghz..

So then, wouldn't throwing 3 SPU's at this problem practically double your vertex power?, and you still have plenty left to work with to.

 

[Heinrich4]

Someday im read pdf of PSEdge talking about SPUs can process (40/60% culling?) 800/750 thousand verts/sec at 60fps:

http://hellome.gametrailers.com/gamepad/

http://www.watch.impress.co.jp/game/docs/20070316/pe.htm

http://cgi1.watch.impress.co.jp/cgi...h.impress.co.jp/game/docs/20070316/PSEdge.zip

The Gethaway
http://generationdreamteam.free.fr/afrika/thegateway/gatewaymars6.jpg
http://generationdreamteam.free.fr/afrika/thegateway/gatewaymars7.jpg
http://generationdreamteam.free.fr/afrika/thegateway/gate3.gif http://generationdreamteam.free.fr/afrika/thegateway/gate2.gif

 

[Betanumerical]

Someday im read pdf of PSEdge talking about SPUs can process (40/60% culling?) 800/750 thousand verts/sec at 60fps:

http://hellome.gametrailers.com/gamepad/

http://www.watch.impress.co.jp/game/docs/20070316/pe.htm

http://cgi1.watch.impress.co.jp/cgi...h.impress.co.jp/game/docs/20070316/PSEdge.zip

The Gethaway
http://generationdreamteam.free.fr/afrika/thegateway/gatewaymars6.jpg
http://generationdreamteam.free.fr/afrika/thegateway/gatewaymars7.jpg
http://generationdreamteam.free.fr/afrika/thegateway/gate3.gif http://generationdreamteam.free.fr/afrika/thegateway/gate2.gif

I found proof of it in text. I was looking for the slides, and found all of them but the one you were looking for.

Test case examples:
* In general 1 SPU can process 750k triangles per frame at 60FPS while hopefully culling 60% of the triangles.

http://www.gfdata.de/archiv03-2007-gamefront/2881.html

Rest of the slides

A slide related to culling..

 

[Neb]

I'll accept that interpretation, but without the clarification that sentence is a bit unclear, imo.

That's because I believe 'has' could also be used instead of 'contains', in the translation, giving:

"Tiene unos" would still be "has some" or "contains" or "has roughly".... 6 million polygons. Tiene -> has unos -> some

So it still hints completly to the whole level. :smile:

"And the level with water and jetski is one of the biggest levels and contains/has roughly/has some 6 million polygons"

So he is talking about how the level is one of the biggest and contains lots of polygons for a FPS level.

AAAIIIEEEE ZOMBIE THREAD OF THE UNDEAD.

Really, anybody got any new FACTS here?

I'll crawl back now into my cave...

 

[archangelmorph]

I found proof of it in text. I was looking for the slides, and found all of them but the one you were looking for.



http://www.gfdata.de/archiv03-2007-gamefront/2881.html

Rest of the slides

A slide related to culling..

So that's 750k "tris" not "verts"?

 

[betan]

And it's not even a completely fair comparison as vertex shaders don't perform any culling.
In this case is fairly correct to simply expect SPUs to match clock per clock modern vertex shader implementations.
As we know SPUs run at 3.2 Ghz..

What is the real catch though?
I mean everyone seems to agree that it should be relatively trivial, and c/c++ shaders aren't new nor "unportable". Why don't Sony provide a library or sample implementation for SPU vertex shading? After all a couple of SPUs are better than five of them, costwise.
I'd think it's much beneficial than pixel visibility culling under MSAA.

Or the "lazy ports" weren't really suffering from "limited vertex shading capabilities of RSX" as we were let to believe?

 

[Betanumerical]

So that's 750k "tris" not "verts"?

It would appear from the table and quote that it is indeed triangles and not vertices. Does this make much difference? I'm not well versed in the differences between them.

 

[AlNom]

It would appear from the table and quote that it is indeed triangles and not vertices. Does this make much difference? I'm not well versed in the differences between them.

Something to chew on:

1 individual triangle = 3 vertices
2 triangles back-to-back may be 4 vertices
3 triangles can be 5 vertices

 

[pjbliverpool]

Is it possible to view the breakdown of work between vertex shaders and pixel shaders in a PC GPU? If it is, why doesn't someone just run CoD4 on a G71 at 600p/2xAA and see if its vertex or pixel bound.

That should give a pretty good indication.

Testing other games aswell would obviously give an even more rounded picture.

 

[AlNom]

Is it possible to view the breakdown of work between vertex shaders and pixel shaders in a PC GPU? If it is, why doesn't someone just run CoD4 on a G71 at 600p/2xAA and see if its vertex or pixel bound.

That should give a pretty good indication.

Testing other games aswell would obviously give an even more rounded picture.

It is possible to get a snapshot of pixel shader and vertex shader work during a frame, but I believe such programs are only in the hands of developers (or someone who knows how to make the program).

 

[Shifty Geezer]

Something to chew on:

1 individual triangle = 3 vertices
2 triangles back-to-back may be 4 vertices
3 triangles can be 5 vertices

In an optimized mesh, you get pretty close to one vertex per triangle. For each triangle, one new vertex is added to the two vertices forming the edge of the triangle to which the new triangle is attached.

 

[betan]

Is it possible to view the breakdown of work between vertex shaders and pixel shaders in a PC GPU? If it is, why doesn't someone just run CoD4 on a G71 at 600p/2xAA and see if its vertex or pixel bound.

That should give a pretty good indication.

Testing other games aswell would obviously give an even more rounded picture.

Thinking about it, there are a couple of PS3 ports which simply choose to run at a lower resolution on PS3. At least those should not be vertex bound (on PS3).