How many vertex pipes could a CPU core equal?

[Bill]

Reason I ask is, I think X360 should use one core for that. Then the GPU is all shader.

It could also make more sense out of this:

http://www.beyond3d.com/forum/showthread.php?t=25152

 

[nAo]

Reason I ask is, I think X360 should use one core for that. Then the GPU is all shader.

It's very difficult to answer your question cause CPUs and GPUs have very different architectures and it's like comparing apples to oranges.
If we compare a VMX unit (on the X360's CPU) and a Xenos' ALU the latter can do a lot more work per clock cycle than its CPU counterpart so even if the CPU has a 8x higher clock than the GPU I'm willing to bet that a single VMX unit is not going to be much more efficient than 2 or 3 GPU ALUs.
Futhermore it's way easier to write and efficiently run your vertex shaders on a GPU than on a CPU (and I don't want even to think about vertex texturing running on a CPU..), even if Microsoft could provide some higly optimized hlsl compiler for X360 CPU...
If you look at a traditional (non unfied) GPU like G70 only 15% of its programmable computational power is devoted to vertex shading. Assuming a G70-like architecture is less efficient than Xenos in allocating its resources we can say that in many cases the amount of vertex shading work it's not that relevant to take the burden out of the GPU, because you are probably not going have a big payback, for the amount of work you need to do..

 

[Bill]

Interesting.

Well, apparantly as late as the GF2MX cards, they only had partial vertex "assistors", and not pipes at all.

Also, didn't some Sony guy claim Devs didn't like unified shaders, and were doing vertexs on the Xcpu as a slam?

 

[nAo]

Well, apparantly as late as the GF2MX cards, they only had partial vertex "assistors", and not pipes at all.

Never heard about this..

Also, didn't some Sony guy claim Devs didn't like unified shaders, and were doing vertexs on the Xcpu as a slam?

Never heard about this too..

 

[see colon]

Interesting.

Well, apparantly as late as the GF2MX cards, they only had partial vertex "assistors", and not pipes at all.

GF2MX (and all nv1x varients) don't have vertex shaders at all, but they do have hardware transform and hardware lighting. though it lacks flexability and any real programability, the hardware T&L in NV1x can help accelerate some fuctions of the software vertex shader included in DX8.

 

[one]

Also, didn't some Sony guy claim Devs didn't like unified shaders, and were doing vertexs on the Xcpu as a slam?

Is David Kirk a Sony guy now?

 

[Guden Oden]

Is David Kirk a Sony guy now?

David Kirk is whatever you pay him to be.

 

[Gholbine]

David Kirk is whatever you pay him to be.

What's his going rate?

 

[Gholbine]

Dp.