Gamma-ray pulsar binary search #1 on GPUs

Bernd Machenschalk
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 4266
Credit: 244925706
RAC: 16562
Topic 203459

Thanks to the help of Christophe Choquet we now do have an efficient GPU version of the Gamma-ray pulsar search application. This is meant to replace the Radio-Pulsar search as our main GPU application when we run out of Arecibo data to process. We have been testing this application versions on Albert@Home for a couple of weeks. The Windows build still doesn't work, and we're working on that. However, we are releasing the versions for Linux and OSX now on Einstein@home for Beta test to get a wider test bed and get an impression how it compares to the CPU version.

 

BM

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

I suppose there's no way of

I suppose there's no way of running those without getting CPU tasks too? I'd like to avoid that, especially v1.05.

 

22
22
Joined: 6 Nov 11
Posts: 14
Credit: 757787830
RAC: 3878

Observations: the percentage

Observations:

the percentage clock starts ticking (after a while) at about 2%; the WU is finished at below 6%, the counter then jumps to 100%. These figures differ when you run two GPU WUs concurrently (finish at around 9%). Still different figures applied with these apps at Albert: single GPU WU finished at about 12%

Jim1348
Jim1348
Joined: 19 Jan 06
Posts: 463
Credit: 257957147
RAC: 0

floyd_7 wrote:I suppose

floyd_7 wrote:
I suppose there's no way of running those without getting CPU tasks too? I'd like to avoid that, especially v1.05.

You can just select the Binary Radio Pulsar Search (Parkes PMPS XT) application, since it doesn't have any tasks available anyway.  And then select "Run test applications" of course. 

 It works for me, other than the fact that they were all erroring out after 10 seconds when they were running on my two GTX 960s running under Ubuntu 16.10.  However, I now just have one task left running on one card, and it is doing OK after 7 minutes.  So it seems that the problem is only when running on two cards.

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

Jim1348 wrote:And then select

Jim1348 wrote:
And then select "Run test applications" of course.


If I'm not mistaken that enables all test applications no matter what. But I've disabled it again anyway since BOINC seriously overfetched. With an estimated run time of 2:08 hours and a 2 days cache I got 18 tasks every minute.

(Meant to write this as an edit to my previous post but then somehow got "no permission")

Quote:
So it seems that the problem is only when running on two cards.


I'll try that later today. Need to be off now ...

 

Richie
Richie
Joined: 7 Mar 14
Posts: 656
Credit: 1702989778
RAC: 0

I see this Einstein-version

I see this Einstein-version takes about 0,54x time compared to Albert-version, even though the estimated computation size was 105,000 GFLOP's for both kind of tasks. This was observed on a host with AMD/ATI card, Linux Mint 18, two tasks running parallel on both cases.

edit: Same observation also with a host with Nvidia card.

Sid
Sid
Joined: 17 Oct 10
Posts: 160
Credit: 920917209
RAC: 279715

I've got some a bit strange

I've got some a bit strange WUs for windows machine with name:

Gamma Ray pulsar binary search #1 onGPUs 1.05(FGRPSSE)  

However, they are purely CPU.  

Mad_Max
Mad_Max
Joined: 2 Jan 10
Posts: 153
Credit: 2134792969
RAC: 440417

Same strange behavior here. I

Same strange behavior here. I got Gamma-ray pulsar binary search #1 on GPUs on windows host (while Bernd  says it not ready for Win platform).

https://einsteinathome.org/task/593239043

 

And more interesting - this host not even have any GPU available to BOINC. But WU still running(on CPU of course) and looks fine so far:

20:46:24 (1628): [normal]: This Einstein@home App was built at: Sep 21 2016 10:12:24

20:46:24 (1628): [normal]: Start of BOINC application 'projects/einstein.phys.uwm.edu/hsgamma_FGRPB1G_1.05_windows_intelx86__FGRPSSE.exe'.
20:46:24 (1628): [debug]: 2.1e+015 fp, 3.4e+009 fp/s, 614089 s, 170h34m48s53
command line: projects/einstein.phys.uwm.edu/hsgamma_FGRPB1G_1.05_windows_intelx86__FGRPSSE.exe --inputfile ../../projects/einstein.phys.uwm.edu/LATeah2003L.dat --alpha 4.42281478648 --delta -0.0345027837249 --skyRadius 2.152570e-06 --ldiBins 15 --f0start 12 --f0Band 8 --firstSkyPoint 0 --numSkyPoints 1 --f1dot -1.1e-11 --f1dotBand 1e-12 --df1dot 3.344368011e-15 --ephemdir ..\..\projects\einstein.phys.uwm.edu\JPLEPH --Tcoh 2097152.0 --toplist 10 --cohFollow 10 --numCells 1 --useWeights 1 --Srefinement 1 --CohSkyRef 1 --cohfullskybox 1 --mmfu 0.1 --reftime 56100 --model 0 --f0orbit 0.005 --mismatch 0.1 --demodbinary 1 --BinaryPointFile ../../projects/einstein.phys.uwm.edu/templates_LATeah2003L_0020_1224.dat --debug 1 -o LATeah2003L_20.0_0_-1e-11_1224_0_0.out
output files: 'LATeah2003L_20.0_0_-1e-11_1224_0_0.out' '../../projects/einstein.phys.uwm.edu/LATeah2003L_20.0_0_-1e-11_1224_0_0' 'LATeah2003L_20.0_0_-1e-11_1224_0_0.out.cohfu' '../../projects/einstein.phys.uwm.edu/LATeah2003L_20.0_0_-1e-11_1224_0_1'
20:46:25 (1628): [debug]: Flags: i386 SSE GNUC X86 GNUX86
20:46:25 (1628): [debug]: Set up communication with graphics process.
% Opening inputfile: ../../projects/einstein.phys.uwm.edu/LATeah2003L.dat
% Total amount of photon times: 30007
% Preparing toplist of length: 10
% Read 36 binary points
read_checkpoint(): Couldn't open file 'LATeah2003L_20.0_0_-1e-11_1224_0_0.out.cpt': No such file or directory (2)
% fft_size: 16777216 (0x1000000)
% Sky point 1/1
% Binary point 1/36
% Creating FFT plan.
% Starting semicoherent search over f0 and f1.
% nf1dots: 301 df1dot: 3.344368011e-015 f1dot_start: -1.1e-011 f1dot_band: 1e-012
% Filling array of photon pairs
.............................................................................................................................................................................................................................................................................................................
% Binary point 2/36
% Starting semicoherent search over f0 and f1.
% nf1dots: 301 df1dot: 3.344368011e-015 f1dot_start: -1.1e-011 f1dot_band: 1e-012
% Filling array of photon pairs
...................................
------------- cut -------------- (at 25/36 Binary point now after ~5 hours of runtime)
Richie
Richie
Joined: 7 Mar 14
Posts: 656
Credit: 1702989778
RAC: 0

Mad_Max wrote:Same strange

Mad_Max wrote:
Same strange behavior here. I got Gamma-ray pulsar binary search #1 on GPUs on windows host (while Bernd  says it not ready for Win platform).

 

I believe a logical explanation for CPU version (FGRPSSE) also being offered under this application title could be that CPU versions help to gather further information for development. So they can compare results between CPU and GPU and fine-tune these applications etc. Those CPU versions for every platform existed first. They were the original 'form' of this thing. GPU versions saw the sunlight only later (except for the third giant... Windows platform... not yet).

Bernd Machenschalk
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 4266
Credit: 244925706
RAC: 16562

Sorry for the

Sorry for the misunderstanding - currently we have CPU versions of FGRPB1G, too, to validate the GPU results.

BM

MAGIC Quantum Mechanic
MAGIC Quantum M...
Joined: 18 Jan 05
Posts: 1695
Credit: 1042554346
RAC: 1386399

Mine are all set to not get

Mine are all set to not get any of these and yet I did  get over 30 of them on the first one I check.

Didn't help my VB tasks at all

Comment viewing options

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