[New release] BRP app v1.28 feedback thread
8 Aug 2012 17:37:23 UTC
Topic 84859
(moderation:
We just released the first app versions of a new BRP4 v1.28 app release. Over the next days, we will release this app for all supported platforms, today we released
- OpenCL (ATI/AMD), 32 bit, for Linux,
- OpenCL (ATI/AMD), 32 bit, Windows and
- OpenCL (ATI/AMD) OSX (Lion and later).
All versions require BOINC 7.0.25 or later, but we encourage volunteers to actually use the latest recommended BOINC versions, currently 7.0.28 or 7.0.31.
Versions for CPU and NVIDIA CUDA GPU will follow soon, plus some app versions that are native 64 bit apps.
New in this app version:
- Various performance improvements
- cross-validation improvements for HD 69xx GPUs
Cheers
HBE
[New release] BRP app v1.28 feedback thread
)
Hmm. My ATI OSX/ML machine can't get work. 7.0.31.
Dublin, California
Team: SETI.USA
RE: Hmm. My ATI OSX/ML
)
We had a temporary problem on Albert, BRP4 work should now be available again.
Cheers
HBE
Got some. Thanks!
)
Got some. Thanks!
Dublin, California
Team: SETI.USA
I did 1
)
I did 1 task.
http://albertathome.org/host/3755/tasks
Was faster then the older one.
Good work.
http://albert.phys.uwm.edu/wo
)
http://albertathome.org/workunit/98764
Looking good so far. Great work.
Stephan
Crunched roughly 50 so far
)
Crunched roughly 50 so far with no problems; using an HD 6850 with Boinc v.7.28 and OC'd i5 2500K (4.5GHz).
Time reduced a lot from average 3,800secs to 2,100secs doing 1xWU at a time...now if you could reduce the CPU Test WUs from 16,000+ secs I wouldn't mind. :)
RE: - OpenCL (ATI/AMD) OSX
)
I have two pretty much identical hosts - 1859 and 1868 that now have this app. They are both running the latest BOINC (7.0.31) and both have AMD 5750 1GB GPUs. The only hardware difference is that the first one has 4GB RAM whilst the second has 8GB.
The one with smaller RAM has run times around 7Ksecs whilst the other is taking 18Ksecs????
Both machines run EAH tasks on all 4 CPUs and both are doing S6LV1 tasks in aroung 17-18Ksecs. All figures quoted were for tasks crunched over the weekend when there was no user sitting at the keyboard. These machines are used during office hours on weekdays. When idle, both machines power down the screen quite quickly (maybe 3 mins) so the only thing I can now think of is whether or not the screensaver process is still running when the screen turns off. I'll have to check that next time I have access to those machines.
Anyone got any other suggestions? At some point I will release a CPU core from crunching to see what difference that makes and later I may attempt to run two tasks simultaneously - maybe not when you think of the extra heat that might be produced. The aluminium cases are running pretty hot already :-).
Cheers, Gary.
Hi Gary! Freeing a CPU core
)
Hi Gary!
Freeing a CPU core is probably a good thing to try.
Another idea: What I'm always a bit suspicious about (BOINC-performance-wise) is OSX's power saving feature to automatically switch between a CPU build-in GPU (e.g. the Intel stuff in Sandy/Ivy Bridge CPUs) and the dedicated graphics card. Could this be the case that this feature is set differently on those two hosts perhaps?
Cheers
HB
Congratulations! My pc did a
)
Congratulations!
My pc did a BRP 1.28 in ~1980 sec. HD6900 Series GPU, i7 with free cpu, 7.0.31, CCC 12.6
On HD5850 a wu takes ~3450sec.
AMD-apps are grown up now.
Phantastic! The amd is a
)
Phantastic!
The amd is a HD6950 running a single wu @ 82% gpu load, the nVidia is a GTX680, unknown settings.
edit: uups, picture is missing...
wu 997334
task 279100 runtime 1,981.83 Binary Radio Pulsar Search v1.28 (opencl-ati)
task 279099 runtime 11,162.70 Binary Radio Pulsar Search v1.25 (BRP3cuda32)