Forum > GPU crunching
V10/11 of modified SETI MB CUDA + opt AP package for full multi-GPU+CPU use
k6xt:
--- Quote from: Raistmer on 09 Mar 2009, 10:33:53 am ---My own Q9450 did ~4400 with MB tasks before CUDA start. Now it ~5200 with 9600GSO addition. But it still not stabilized (I expect it to be higher).
--- End quote ---
Am I correct to thinik the important GPU spec is core clock?
You hinted in earlier post that stream processors are not involved in GPU crunching. How important is memory clock?
Thanks Art
Raistmer:
--- Quote from: k6xt on 09 Mar 2009, 10:43:14 am ---
--- Quote from: Raistmer on 09 Mar 2009, 10:33:53 am ---My own Q9450 did ~4400 with MB tasks before CUDA start. Now it ~5200 with 9600GSO addition. But it still not stabilized (I expect it to be higher).
--- End quote ---
Am I correct to thinik the important GPU spec is core clock?
You hinted in earlier post that stream processors are not involved in GPU crunching. How important is memory clock?
Thanks Art
--- End quote ---
Not correct to think stream processors not included. I said "texture-related hardware", not stream processors. There is dedicated hardware that will do linear approximation between 2 dots in texture for example. This particular part can be used inside CUDA BTW, but I don't think this app use it. But there are another parts of GPU that can't be accessed via CUDA at all.
Will try to find relevant info in CUDA FAQ or forums (I seen that there).
Raistmer:
Ok, here:
"
What texture features does CUDA support?
CUDA supports 1D and 2D textures, which can be accessed with normalized (0..1) or integer coordinates. Textures can also be bound to linear memory and accessed with the "tex1Dfetch" function.
The hardware only supports 1, 2 and 4-component textures, not 3-component textures.
3D textures, cube maps, texture arrays, compressed textures and mip-maps are not currently supported.
"
and
"
Are graphics operations such as z-buffering and alpha blending supported in CUDA?
No. Access to video memory in CUDA is done via the load/store mechanism, and doesn't go through the normal graphics raster operations like blending. We don't have any plans to expose blending or any other raster ops in CUDA.
"
AND AFAIK any game will do blending a lot :)
Raistmer:
--- Quote from: k6xt on 09 Mar 2009, 10:43:14 am ---Am I correct to thinik the important GPU spec is core clock?
--- End quote ---
Think no.
look beta forum for GPU OCing questions. It was showed that shader clock most important and engine clock least important.
My own GPU can support high memory freqs (for now it works on 2x1140 and 2x1120 known to be stable already). With shader freqs it still awaiting evaluation...
k6xt:
--- Quote from: Raistmer on 09 Mar 2009, 10:52:10 am ---Ok, here:
"
What texture features does CUDA support?
CUDA supports 1D and 2D textures, which can be accessed with normalized (0..1) or integer coordinates. Textures can also be bound to linear memory and accessed with the "tex1Dfetch" function.
The hardware only supports 1, 2 and 4-component textures, not 3-component textures.
3D textures, cube maps, texture arrays, compressed textures and mip-maps are not currently supported.
"
and
"
Are graphics operations such as z-buffering and alpha blending supported in CUDA?
No. Access to video memory in CUDA is done via the load/store mechanism, and doesn't go through the normal graphics raster operations like blending. We don't have any plans to expose blending or any other raster ops in CUDA.
"
AND AFAIK any game will do blending a lot :)
--- End quote ---
From this maybe I can infer CUDA uses some but not all capabilities. Just considering SETI or Einstein processing, then, there must be a point beyond which it is wasted money to buy more performance. Would it be true that increased core clock is always valuable, memory less so? And, is it possible to decide how many stream processors is enough, or is the case "more is better"?
I am trying to decide how best to implement the new PC with regard to SETI processing. Given I don't understand much about video processing, guessing which specs will yield the most result per dollar is a task. The spec sheets that I've read for example don't talk about texture so it is hard to relate to specs.
Regards Art
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version