Amd dual rasterisers

[Davros]

been reading up on amd cards and found this quote
"as with barts pro and xt the gpu is now split into two rasterisers"

Whats the benefit of that ?

 

[rpg.314]

It can now assemble and rasterize 2 triangles per clock.

 

[pjbliverpool]

Whats the situation with NV 4xx and 5xx. Are they still limited to a single triangle per clock?

 

[rpg.314]

Whats the situation with NV 4xx and 5xx. Are they still limited to a single triangle per clock?

The geforce versions. The quadro versions are unlocked, so to speak.

 

[Davros]

so the standard geforces have 2 rasterisers but ones disabled ?

 

[DarthShader]

They have 4 rasterizers, only one is working with tesselation off for standard geometry, all four when tesselation is being done.

 

[MDolenc]

GF100/GF110 has 4 rasterizers, but is limited to 1 triangle per clock output from vertex assembly. So unless triangles come from tesselation, you won't see full triangle rates on GeForce. It'll also cull at full speed, if I'm not mistaken.
Quadro will also run close to peak without tesselation.

 

[Davros]

so does that partly explain nv's superior tessellation performance
and the increase in tessellation performance between 5000 + 6000 cards

 

[fellix]

Fermi uses its coherent L2 cache to avoid the slow access to the global memory for sync'ing the primitives. Also the tessellation workload is much more distributed than in Cayman, removing a potential bottleneck.

 

[3dcgi]

Only GTX480/580 have 4 rasterizers. The GTX460 chip has 2 so it fights on equal theoretical ground with Cayman.

 

[AlexV]

It'll also cull at full speed, if I'm not mistaken.

Yep, it does.

 

[Davros]

so barts has 2 rasterisers, multiple rops/shaders/tmu's/memory controllers
how far away is it from being sli on a chip ?

 

[CNCAddict]

What are the advantages to having high polygon count without tessellation? I notice that textures get stretched on most tessellated surfaces I've seen. Does physics collision detection work on tessellated geometry? It seems like tessellation is almost a gimmick compared to highly detailed primary geometry.

 

[silent_guy]

so barts has 2 rasterisers, multiple rops/shaders/tmu's/memory controllers
how far away is it from being sli on a chip ?

The units have the same high bandwidth to the same memory pool. That makes all the difference in the world.

 

[3dcgi]

What are the advantages to having high polygon count without tessellation? I notice that textures get stretched on most tessellated surfaces I've seen. Does physics collision detection work on tessellated geometry? It seems like tessellation is almost a gimmick compared to highly detailed primary geometry.

Tessellation isn't a gimmick. It's essentially a form of compression. Though hardware tessellation also gives the ability to perform some operations prior tessellation such as vertex animation and patch culling.