The GPU only accelerates a limited part of the WU; once you get to a haflway decent GPU (say a 250 or better) the task is effectively CPUbound again, so a fast CPU can easily beat a slow CPU with a fast GPU.
Boinc doesn't show overclock speeds; but the GPUless computer is an i7 chip and is scoring at about the same rate as one of mine running at 4.1ghz (HT enabled). The second is an old core2 chip. The i7 has a really large advantage on ABP2 WUs vs GCE ones, doing them in about 2/3rds the time vs core2 (and IIRC AMD) systems that do both types in about the same amount of time.
The GPU only accelerates a limited part of the WU; once you get to a haflway decent GPU (say a 250 or better) the task is effectively CPUbound again, so a fast CPU can easily beat a slow CPU with a fast GPU.
But they ARE working on the new Cuda client that supposedly will push more of the unit onto the gpu, something similar to the other gpu projects.
Are they actually working on it now? It's been 2 or 3 weeks from being done for months now, but other stuff has had the priority....
There have been some problems or whatever that have come up, but they are still working on it. Priorities are always a factor when the workforce is not unlimited! I do not know the absolute most current status though.
And a example how CPU only is
)
And a example
how CPU only is faster then CUDA with an GTX480!
[url=http://en.wikipedia.org/
)
[url=http://en.wikipedia.org/wiki/Amdahl's_law]Amdahl's law[/url].
The GPU only accelerates a limited part of the WU; once you get to a haflway decent GPU (say a 250 or better) the task is effectively CPUbound again, so a fast CPU can easily beat a slow CPU with a fast GPU.
Boinc doesn't show overclock
)
Boinc doesn't show overclock speeds; but the GPUless computer is an i7 chip and is scoring at about the same rate as one of mine running at 4.1ghz (HT enabled). The second is an old core2 chip. The i7 has a really large advantage on ABP2 WUs vs GCE ones, doing them in about 2/3rds the time vs core2 (and IIRC AMD) systems that do both types in about the same amount of time.
RE: [url=http://en.wikipedi
)
But they ARE working on the new Cuda client that supposedly will push more of the unit onto the gpu, something similar to the other gpu projects.
Are they actually working on
)
Are they actually working on it now? It's been 2 or 3 weeks from being done for months now, but other stuff has had the priority....
RE: Are they actually
)
There have been some problems or whatever that have come up, but they are still working on it. Priorities are always a factor when the workforce is not unlimited! I do not know the absolute most current status though.