If you did the same comparison at GPUGrid, you'd find the GTX 670 to be almost twice as fast as the 560Ti, but that's CUDA42.
exactly - its an optimization issue. and your point lends further credence to what i said earlier - Einstein@Home just isn't optimized as well for Kepler as it is for Fermi. if it were (i.e. if Einstein@Home were taking advantage of CUDA42, and not just CUDA32), then we would actually see substantial differences in performance between Kepler GPUs and Fermi GPUs on the Einstein@Home project. but until then...
I asked about this a while ago. The reason for Einstein using CUDA32 (from Bernd) is:
"Our BRP application triggers a bug in CUDA versions 4 & 5 that has been reported to NVida. The bug was confirmed but not fixed yet. Until it has, we're stuck with CUDA 3."
Still not fixed. Glad he didn't hold his breath...
This seems to be finally fixed in the current release candidate of CUDA 5.5. We'll see how it comes out when they actually release it.
BM
This is *potentially* absolutely fantastic news. The GK110s are horribly optimized for this research (as well as every other BOINC program), and could really use a boost right now :P
We'll certainly keep the CUDA 3.2 version, and maintain it as long as possible with reasonable effort. Besides a possible CUDA 5.5 version has yet to show a speedup over the 3.2 version that makes up for the additional maintenance work of - yes - yet another app version (well, actually three).
Question:
Why you not use thread priority trick with BRP5 as was in previos GPU searches?
Process (einsteinbinary_BRP5_1.34_windows_intelx86__opencl-ati.exe) priority is above idle(below normal) - OK. But thread priority too low (=1):
If i not reserve 1 CPU core for E@H in Boinc prefs GPU almost idle waiting while CPU part of app gets some of CPU time.
Since yesterday I'm not getting any BRP5 work for my ATI/AMD GPUs. The message on all machines is:
Einstein@Home 08-01-13 11:35 Sending scheduler request: To fetch work.
Einstein@Home 08-01-13 11:35 Requesting new tasks for ATI
Einstein@Home 08-01-13 11:35 Scheduler request completed: got 0 new tasks
Einstein@Home 08-01-13 11:35 No work sent
Einstein@Home 08-01-13 11:35 No work is available for Binary Radio Pulsar Search (Perseus Arm Survey)
Einstein@Home 08-01-13 11:35 No work available for the applications you have selected. Please check your preferences on the web site.
My preferences have not changed and are set "use ATI GPU" and run 2 WUs at a time (0.5 GPU utilization factor of BRP apps).
The server status shows that work is available and i can find no news about an outage. Any ideas?
Edit: Einstein is the only ATI project on these GPUs...
RE: RE: If you did the
)
I asked about this a while ago. The reason for Einstein using CUDA32 (from Bernd) is:
"Our BRP application triggers a bug in CUDA versions 4 & 5 that has been reported to NVida. The bug was confirmed but not fixed yet. Until it has, we're stuck with CUDA 3."
http://einsteinathome.org/node/196873&nowrap=true#123838
He goes on to say they reported the bug to NVidia 2.5 years ago. Still not fixed. Glad he didn't hold his breath...
RE: Still not fixed. Glad
)
This seems to be finally fixed in the current release candidate of CUDA 5.5. We'll see how it comes out when they actually release it.
BM
BM
RE: This seems to be
)
I'll keep my fingers crossed for this :)
That's great news :) Hope to
)
That's great news :)
Hope to see a BRP4cuda55 app soon ;-)
-----
RE: RE: Still not fixed.
)
This is *potentially* absolutely fantastic news. The GK110s are horribly optimized for this research (as well as every other BOINC program), and could really use a boost right now :P
Best of luck.
Will building the workunits
)
Will building the workunits against CUDA 5 mean that older cards will stop working with the GPU tasks? Or will you be maintaining yet more versions?
Thanks!
We'll certainly keep the CUDA
)
We'll certainly keep the CUDA 3.2 version, and maintain it as long as possible with reasonable effort. Besides a possible CUDA 5.5 version has yet to show a speedup over the 3.2 version that makes up for the additional maintenance work of - yes - yet another app version (well, actually three).
BM
BM
Question: Why you not use
)
Question:
Why you not use thread priority trick with BRP5 as was in previos GPU searches?
Process (einsteinbinary_BRP5_1.34_windows_intelx86__opencl-ati.exe) priority is above idle(below normal) - OK. But thread priority too low (=1):
If i not reserve 1 CPU core for E@H in Boinc prefs GPU almost idle waiting while CPU part of app gets some of CPU time.
Since yesterday I'm not
)
Since yesterday I'm not getting any BRP5 work for my ATI/AMD GPUs. The message on all machines is:
My preferences have not changed and are set "use ATI GPU" and run 2 WUs at a time (0.5 GPU utilization factor of BRP apps).
The server status shows that work is available and i can find no news about an outage. Any ideas?
Edit: Einstein is the only ATI project on these GPUs...
Same situation here. No
)
Same situation here. No work
regards