Advanced search

Message boards : News : New CPU WUs - DUDHIVPR

Author Message
Profile MJH
Project administrator
Project developer
Project scientist
Send message
Joined: 12 Nov 07
Posts: 696
Credit: 27,266,655
RAC: 0
Level
Val
Scientific publications
watwat
Message 35848 - Posted: 24 Mar 2014 | 10:38:10 UTC
Last modified: 24 Mar 2014 | 10:38:25 UTC

Hi All,

I've just submitted our first large batch of work units for the CPU application, named "DUDHIVPR*". The purpose of these WUs is to test the correctness of operation of our application by attempting to reproduce published benchmark data.

Matt

Jacob Klein
Send message
Joined: 11 Oct 08
Posts: 1068
Credit: 1,149,603,614
RAC: 1,026,674
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 35856 - Posted: 24 Mar 2014 | 12:23:27 UTC - in response to Message 35848.
Last modified: 24 Mar 2014 | 12:27:11 UTC

1) When I suspend the GPUGrid "CPU Only" tasks in BOINC, they are actually still running, according to Task Manager and Process Explorer. Surely this is a huge bug?

2) How often do these tasks checkpoint? I had several running for 20 minutes, I think, then restarted BOINC, and lost all of that work. :(

Windows 8.1 Update 1 x64, BOINC 7.3.11 x64

Profile MJH
Project administrator
Project developer
Project scientist
Send message
Joined: 12 Nov 07
Posts: 696
Credit: 27,266,655
RAC: 0
Level
Val
Scientific publications
watwat
Message 35859 - Posted: 24 Mar 2014 | 13:52:52 UTC - in response to Message 35856.

Hi Jacob,


These particular WUs don't checkpoint - the code doesn't currently support it, and the WUs shouldn't last for more than 30mins or so.

However, suspending ought to work. I'll check it out.

Matt

Jacob Klein
Send message
Joined: 11 Oct 08
Posts: 1068
Credit: 1,149,603,614
RAC: 1,026,674
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 35860 - Posted: 24 Mar 2014 | 14:23:43 UTC - in response to Message 35859.

Regarding run time, I'm running several on my high-powered i7-965-XE (overclocked to 3.7 GHz).... and after a full hour, they're only 61% done. That's quite a long time to go without checkpointing.

Profile MJH
Project administrator
Project developer
Project scientist
Send message
Joined: 12 Nov 07
Posts: 696
Credit: 27,266,655
RAC: 0
Level
Val
Scientific publications
watwat
Message 35862 - Posted: 24 Mar 2014 | 14:43:37 UTC - in response to Message 35860.


Regarding run time, I'm running several on my high-powered i7-965-XE (overclocked to 3.7 GHz).... and after a full hour, they're only 61% done. That's quite a long time to go without checkpointing.


There's variation as the runtime depends on the complexity of the input molecualr (all WUs being different).

Future WUs will be shorter, I hope. Part of the test here is to determine how long WUs have to run for before converging on reproducible results.

We can't practically add checkpointing to this code, its design makes it difficult to retrofit.

MJH

Profile Neil Polson
Avatar
Send message
Joined: 23 Jun 13
Posts: 1
Credit: 2,017
RAC: 0
Level

Scientific publications
wat
Message 35863 - Posted: 24 Mar 2014 | 14:50:19 UTC

Any chance we can get more than one per core?

Profile MJH
Project administrator
Project developer
Project scientist
Send message
Joined: 12 Nov 07
Posts: 696
Credit: 27,266,655
RAC: 0
Level
Val
Scientific publications
watwat
Message 35865 - Posted: 24 Mar 2014 | 14:59:17 UTC - in response to Message 35863.


Any chance we can get more than one per core?



No sure what you mean. The tasks are serial (they'll use a single core) but you can have one on every CPU core you wish to allocate to the project.

Matt

Post to thread

Message boards : News : New CPU WUs - DUDHIVPR