Originally posted by Jon`C:
This might sound condescending but I don't mean it to be.
no you can't, because:
- The pixel shading performance of a GPU absolutely demolishes the performance of the Cell and any other general-purpose processor in the world.
Cell: 100 GFLOPS measured real-world single-precision floating point performance. It could do a fixed-function transform on about 590 million vertices a second or 2 billion trilinearly filtered texture samples a second. That's super-keen.
RSX: Theoretical floating point output, if you combine the number of floating point operations performed by all of the shader units including the texture samplers? 1.8 TFLOPS. About a billion shaderized, transformed and lit vertices a second. 12 billion anisotropically filtered texture samples a second. This isn't even in the same ballpark. Throwing Cell at the RSX would be like trying to speed up a freight train by ramming it with a Civic. I'm sorry, but a 4 cyl 1.8L VTEC just isn't going to make a difference. Actually it's going to hurt performance, because....
- RSX and Cell do not share memory.
And even if they did, trying to feed data from the CPU straight into the GPU would absolutely strangle the GPU.
Back before the original GeForce and Radeon, computers used the CPU for transformation and lighting. This was "okay" because the polygon counts were relatively low - the CPU could handle the math, and the PCI bus could handle the relatively minimal amount of data that was being sent across.
Even then, the more bandwidth intensive data (textures) were uploaded to the card permanently.
At a certain point, though, scenes simply became too complex and the bus couldn't handle updating the whole thing every frame. So, instead, you upload the scene data to the graphics hardware and issue draw instructions along with a few variables like the camera position.
This is nice for another reason: the GPU can draw the scene asynchronously while the CPU moves ahead and processes gameplay updates for the next frame.
The fastest way you could do it is to lock a vertex buffer, fill it, unlock it and then pass the rendering call, and all of those are blocking operations. And that's still only if you're using the vertex shader for transformation and lighting (just using the Cell for stuff like animation). It goes down from there. Quite sharply.
- The only part of the Cell that's any good for general-purpose processing is the PPE.
The SPEs are only really good at floating point math. They also have no way of directly accessing memory and very small caches. The PPE has to keep them on a short leash - they're more like vertex shaders than full-fledged processors.
The PPE is also the only part of the Cell that can issue instructions to the RSX. Including things like filling vertex buffers and other wonderfully synchronous operations. Thread starvation is :awesome:.
umm... I sort of lost track of where I was going with this post, but in conclusion: the Cell can't be used for any significant rendering operations, and because of the way the PS3 is built your performance would actually suffer if you tried.
no you can't, because:
- The pixel shading performance of a GPU absolutely demolishes the performance of the Cell and any other general-purpose processor in the world.
Cell: 100 GFLOPS measured real-world single-precision floating point performance. It could do a fixed-function transform on about 590 million vertices a second or 2 billion trilinearly filtered texture samples a second. That's super-keen.
RSX: Theoretical floating point output, if you combine the number of floating point operations performed by all of the shader units including the texture samplers? 1.8 TFLOPS. About a billion shaderized, transformed and lit vertices a second. 12 billion anisotropically filtered texture samples a second. This isn't even in the same ballpark. Throwing Cell at the RSX would be like trying to speed up a freight train by ramming it with a Civic. I'm sorry, but a 4 cyl 1.8L VTEC just isn't going to make a difference. Actually it's going to hurt performance, because....
- RSX and Cell do not share memory.
And even if they did, trying to feed data from the CPU straight into the GPU would absolutely strangle the GPU.
Back before the original GeForce and Radeon, computers used the CPU for transformation and lighting. This was "okay" because the polygon counts were relatively low - the CPU could handle the math, and the PCI bus could handle the relatively minimal amount of data that was being sent across.
Even then, the more bandwidth intensive data (textures) were uploaded to the card permanently.
At a certain point, though, scenes simply became too complex and the bus couldn't handle updating the whole thing every frame. So, instead, you upload the scene data to the graphics hardware and issue draw instructions along with a few variables like the camera position.
This is nice for another reason: the GPU can draw the scene asynchronously while the CPU moves ahead and processes gameplay updates for the next frame.
The fastest way you could do it is to lock a vertex buffer, fill it, unlock it and then pass the rendering call, and all of those are blocking operations. And that's still only if you're using the vertex shader for transformation and lighting (just using the Cell for stuff like animation). It goes down from there. Quite sharply.
- The only part of the Cell that's any good for general-purpose processing is the PPE.
The SPEs are only really good at floating point math. They also have no way of directly accessing memory and very small caches. The PPE has to keep them on a short leash - they're more like vertex shaders than full-fledged processors.
The PPE is also the only part of the Cell that can issue instructions to the RSX. Including things like filling vertex buffers and other wonderfully synchronous operations. Thread starvation is :awesome:.
umm... I sort of lost track of where I was going with this post, but in conclusion: the Cell can't be used for any significant rendering operations, and because of the way the PS3 is built your performance would actually suffer if you tried.
Oh, I was under the impression the Cell was much better at video rendering than that. :saddowns:
Originally posted by Isuwen:
Might I add that devving for the 360 is just plain easier, mainly because it's essentially a windows box. There are differences, but they are not so major that you can't use a windows machine for testing.
You're right, but for the most retardedly wrong reason I've ever seen. The 360 is not a windows box, it is not compatible with windows, and you cannot test with windows. It uses an entirely different processor type than windows (x86 vs PPC), uses an entirely different video card than is currently out (although I guess you COULD emulate it with the 8800), and well, just is nothing like a windows machine. You could more easily equate it to a tri-core G5 than a windows machine.
D E A T H
Originally Posted by FastGamerr
