Binary Radio Pulsar Search (Parkes PMPS XT) "BRP6"

floyd
floyd
Joined: 12 Sep 11
Posts: 133
Credit: 186610495
RAC: 0

RE: So, is there any

Quote:
So, is there any noticeable benefit of 1.52 over 1.39 on BRP4G?


HD 7750: No improvement
GTX 750: Run times seem good but I don't have any old data to compare. A quick test indicates 1.39 would take about 15% more time

ravenigma
ravenigma
Joined: 20 Aug 10
Posts: 69
Credit: 80558758
RAC: 278

My GTX 750Ti cards have sped

My GTX 750Ti cards have sped up by a few minutes with the new app. Comes to around 14% improvement.

|MatMan|
|MatMan|
Joined: 22 Jan 05
Posts: 24
Credit: 249005261
RAC: 0

RE: BRP4 and BRP4G is

Quote:
BRP4 and BRP4G is actually the same data (Arecibo), the post-processing doesn't distinguish whether a task ran on GPUs or CPUs, and so doesn't the progress counter.


Ah ok, thanks for the info.

Quote:
So, is there any noticeable benefit of 1.52 over 1.39 on BRP4G?


My Titan Black (on PCIe 3.0 x4 only) went from ~2.200s per task (1.39) to ~1.100s per task (1.52).
Three GTX 750 Ti (PCIe 3.0 x8, PCIe 3.0 x4, PCIe 2.0 x4) went from ~3.200s - 4.200s (1.39) to ~2.400s - ~2.800s.

(one tasks per GPU, one GTX 750 Ti clocks ~100 MHz slower than the others; all GPUs are in a single computer and the integrated Intel HD4600 is also processing Einstein tasks!)

So yes, the benefit is huge in my case!! Thank you! :D

Mumak
Joined: 26 Feb 13
Posts: 325
Credit: 3520908280
RAC: 1614484

BRP4G change v1.39 -> v1.52

BRP4G change v1.39 -> v1.52 Beta; Total Time (CPU time)

NVIDIA Tesla K20c: 1900 (500) -> 1400 (300)
AMD Radeon R9 280X: 1650 (900) -> 1150 (600)

So a significant improvement! Let's roll it out into production ;-)

-----

Bernd Machenschalk
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 4312
Credit: 250472777
RAC: 35034

RE: So a significant

Quote:
So a significant improvement! Let's roll it out into production ;-)

Validation statistics look surprisingly bad for a few app versions (e.g. Linux 64 ATI). I'll leave it at Beta until we have collected a bit more data.

BM

BM

Bernd Machenschalk
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 4312
Credit: 250472777
RAC: 35034

There is a Beta version of a

There is a Beta version of a CUDA 5.5 BRP6 app for 64Bit Linux. This is mainly meant to test the build and deployment process for CUDA 5.5.

BM

BM

AgentB
AgentB
Joined: 17 Mar 12
Posts: 915
Credit: 513211304
RAC: 0

RE: There is a Beta version

Quote:
There is a Beta version of a CUDA 5.5 BRP6 app for 64Bit Linux.

I can not seem to start these, they error out for me immediately. vanilla BRP6 crunches fine.

Running 2 gtx-460 gpus, each running x2.

http://einsteinathome.org/task/506898750
http://einsteinathome.org/task/506890013
http://einsteinathome.org/task/506862660

I have enough vanilla BRP6 to munch on so i will leave beta on

AgentB
AgentB
Joined: 17 Mar 12
Posts: 915
Credit: 513211304
RAC: 0

RE: I have enough vanilla

Quote:
I have enough vanilla BRP6 to munch on so i will leave beta on

Beta now off for this host, i will have another look after the weekend.

Bernd Machenschalk
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 4312
Credit: 250472777
RAC: 35034

Thanks for testing. That was

Thanks for testing. That was a problem in deployment and should be fixed in the new version 1.54.

BM

BM

AgentB
AgentB
Joined: 17 Mar 12
Posts: 915
Credit: 513211304
RAC: 0

Thanks Bernt, that has solved

Thanks Bernt, that has solved the problem, the first two 1.54 completed uneventfully (yet to verify).

A little early to make any comments on performance.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.