GPU Workunits as big as CPU Workunits?

Christian
Christian
Joined: 8 Feb 06
Posts: 3
Credit: 2,043,673
RAC: 0
Topic 218717

Hi! I think the title says what I mean, but to clarify:

Is a WU for a GPU as big as a CPU workunit in the sense that it has the same amount of data to crunch or is it smaller/bigger than what a CPU WU asks from the CPU?

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 747
Credit: 638,550,614
RAC: 1,234,271

I think for most projects, a

I think for most projects, a task is just a task, no differences, only what app processes it.

 

Jim1348
Jim1348
Joined: 19 Jan 06
Posts: 371
Credit: 201,370,089
RAC: 539

I concluded early on that

I concluded early on that they are apparently identical, since they validate each other insofar as I can see.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,179
Credit: 39,843,934,346
RAC: 45,758,354

Christian wrote:Is a WU for a

Christian wrote:
Is a WU for a GPU as big as a CPU workunit ...

It depends on the particular search you are referring to.

Recently, the gravity wave (GW) search using LIGO data had test CPU and GPU applications (O1OD1E) which used the same workunits.  One of the aims was to make sure that both the CPU version and the GPU version gave answers that were sufficiently 'close' for the results to validate each other.  That seemed to work OK - in fact surprisingly so.  That search has been put on hold and the non-GPU O2AS search will be resumed in its place.

At some future time a different 'version' of the O1OD1 search (called "injection") will resume.  When that happens there will probably be once again both a CPU and a GPU version of the app.  If the GPU app improves in efficiency, it could happen that larger tasks might be used to prevent overly short GPU crunch times.  At the moment, present indications are that the same task size is likely to be used for both.

With the gamma-ray pulsar searches that use data from the Fermi satellite (FGRP5 and FGRPB1G), different data is being used and there is no cross comparison.  The GPU app processes data that has 5 times the 'work content' of the much smaller CPU tasks.  A GPU task is just a 'bundle' of 5 CPU sized tasks.

 

Cheers,
Gary.

Mad_Max
Mad_Max
Joined: 2 Jan 10
Posts: 147
Credit: 1,673,220,152
RAC: 355,133

O1OD1E is put "on hold" only

O1OD1E is put "on hold" only in aspect of generating new WUs.
But before shutdown WU generator created so many WUs so it will be enough for atleast for two or few more weeks of work based at current precessing speed.

Its about 2-6 thousands tasks per day and 82 thousands more is ready to send  + there will be some resend due to errors and timeouts

Christian
Christian
Joined: 8 Feb 06
Posts: 3
Credit: 2,043,673
RAC: 0

Thanks a lot to everyone who

Thanks a lot to everyone who replied! This gives me some new insights into the project and how fast GPUs acutally are.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,179
Credit: 39,843,934,346
RAC: 45,758,354

Mad_Max wrote:O1OD1E is put

Mad_Max wrote:
O1OD1E is put "on hold" only in aspect of generating new WUs.

Whilst the announcement about this was very brief, the implication was that they really wanted to start work again on the O2AS search.  It was originally suspended last year because there was some doubt about the data.  That must have been sorted out in such a way that they are now ready to resume that search.  Since O2 (Observation run 2) is 'better' data (collected with a more sensitive instrument) it seems likely that they may want to get back to it as quickly as possible.

There doesn't seem to be any reason why they couldn't completely suspend the remaining unsent tasks of the O1OD1E search if there was a strong motivation to get back to O2AS.  So best not to count on a long lifetime for what remains.  I would guess that as soon as O2AS is ready to go, we might see the changeover.  After all, there is still a lot more to generate for the full O1OD1E search so what is generated but not distributed could probably just be added to what is still remaining to be generated.  I guess we'll find out soon enough :-).

 

Cheers,
Gary.

Jim1348
Jim1348
Joined: 19 Jan 06
Posts: 371
Credit: 201,370,089
RAC: 539

Gary Roberts wrote:Whilst the

Gary Roberts wrote:
Whilst the announcement about this was very brief, the implication was that they really wanted to start work again on the O2AS search.

I just attached a new Linux machine, selecting only "Continuous Gravitational Wave search O2 All-Sky", and immediately got three of them (_0, not resends).  So we are back in business.  Thanks.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,179
Credit: 39,843,934,346
RAC: 45,758,354

Interesting!  I just had a

Interesting!  I just had a look at the server status page and it shows the WU generator 'running' - which is unusual as usually it shows as 'not running' whenever the page is refreshed.  There were just 4 tasks listed as ready to send but 10,000+ tasks in progress.

I hadn't looked at that page for a while so I don't know how quickly (or when) the 10,000+ tasks got out there.  Looks like the WU generator isn't keeping up with the demand at the moment.

EDIT: 10 mins later and the ready to send shows as 8.  The WU generator still shows 'running'.

Cheers,
Gary.

Comment viewing options

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