oohhh shiney. My box is ranked 34 on that list. I wish the metrics posted included actual clock speeds and duty cycles so to facilitate bank for the buck comparisons as well.
The SETI guys swear by Intel -- the only computers in the top 20 or so with AMD chips also have four or five GTX 295s. :)
You'll find the same Intel dominance here at E@H, it's a bit sad :-( .
CU
HB
which compiler do you use? I hope you know that ICC has been identified recently as crippling performances of amd chips...
Recently? Source and link, please.
Most people here are familiar with http://www.swallowtail.org/naughty-intel.shtml, which dates back several years. We periodically test to see whether the tell-tale signs are present in the Einstein executables, both Linux and Windows, and have always come up negative so far.
And so far as I know there isn't a version of ICC targeted on a GPU platform on general release yet - maybe because Intel hasn't released any graphics chips comparable to NVidia's CUDA or ATI's Stream technology yet. Maybe it will come: we'll be watching.
Yes. The current E@H cuda app is more of a proof of concept than anything else, it only offloads a single part of the calculation to the GPU while still doing most of the work on the CPU. A future version of the app with more gpu computations is promised but no ETA has been given.
1 CPU + 1 GPU = very low acceleration (less than 20 %).
Is this situation normal?
Yes. Unfortunately, yes.
E@H (currently) only uses the GPU to do FFT. Most of the work is still done on the CPU. At best, since it's only doing a small portion of the work on the GPU, you'll see a small improvement in speed. If you have a fast CPU and slow GPU, you could even see it take longer running on the GPU.
If you run any other projects on your GPU, you probably will want to turn off Einstein's ability to use your GPU unti they release an app that fully runs on the GPU. Right now, it's really just wasting the GPU's resources.
I think they're working on a version that more fully utilizes the GPU, but there hasn't been a lot of information about that. Stay tuned.
Want to find one of the largest known primes? Try PrimeGrid. Or help cure disease at WCG.
oohhh shiney. My box is
)
oohhh shiney. My box is ranked 34 on that list. I wish the metrics posted included actual clock speeds and duty cycles so to facilitate bank for the buck comparisons as well.
RE: RE: The SETI guys
)
which compiler do you use? I hope you know that ICC has been identified recently as crippling performances of amd chips...
RE: RE: RE: The SETI
)
Recently? Source and link, please.
Most people here are familiar with http://www.swallowtail.org/naughty-intel.shtml, which dates back several years. We periodically test to see whether the tell-tale signs are present in the Einstein executables, both Linux and Windows, and have always come up negative so far.
And so far as I know there isn't a version of ICC targeted on a GPU platform on general release yet - maybe because Intel hasn't released any graphics chips comparable to NVidia's CUDA or ATI's Stream technology yet. Maybe it will come: we'll be watching.
E@H is using GCC now so
)
E@H is using GCC now so intel's compiler shenanigans can't have any effect on it.
RE: E@H is using GCC now so
)
Indeed. Even the Windows app is cross compiled on a Linux box using MinGW (gcc based).
CU
HB
1 CPU + 1 GPU = very low
)
1 CPU + 1 GPU = very low acceleration (less than 20 %).
Is this situation normal?
Yes. The current E@H cuda
)
Yes. The current E@H cuda app is more of a proof of concept than anything else, it only offloads a single part of the calculation to the GPU while still doing most of the work on the CPU. A future version of the app with more gpu computations is promised but no ETA has been given.
RE: 1 CPU + 1 GPU = very
)
Yes. Unfortunately, yes.
E@H (currently) only uses the GPU to do FFT. Most of the work is still done on the CPU. At best, since it's only doing a small portion of the work on the GPU, you'll see a small improvement in speed. If you have a fast CPU and slow GPU, you could even see it take longer running on the GPU.
If you run any other projects on your GPU, you probably will want to turn off Einstein's ability to use your GPU unti they release an app that fully runs on the GPU. Right now, it's really just wasting the GPU's resources.
I think they're working on a version that more fully utilizes the GPU, but there hasn't been a lot of information about that. Stay tuned.
Want to find one of the largest known primes? Try PrimeGrid. Or help cure disease at WCG.
Thank You, friends, for Your
)
Thank You, friends, for Your comments.
Two projects that seem to be
)
Two projects that seem to be almost totally over to GPU crunching are Milkyway and Collatz.
These are dominated by the AMD ATI graphics cards as the nVidia ones are so, relatively, slow.
Looking in on these 2 projects can give you a good insight to fast GPU crunching, with minimum use of the CPU in support (@20% of 1 core).
There are other GPU supporting projects, which I have not contributed to which will help more (GPUGrid).
Shih-Tzu are clever, cuddly, playful and rule!! Jack Russell are feisty!