CUDA32 vs CUDA50


log in

Advanced search

Message boards : Number crunching : CUDA32 vs CUDA50

Author Message
Profile Michael BantaProject donor
Avatar
Send message
Joined: 16 Jul 01
Posts: 100
Credit: 5,410,124
RAC: 41
United States
Message 1458567 - Posted: 29 Dec 2013, 21:50:32 UTC
Last modified: 29 Dec 2013, 21:52:49 UTC

Are cuda50 work units bigger/more complex than cuda42 work units? I assume so but don't know.

My GTX 560 runs cuda42 WUs faster than my GTX 780 run cuda50 WUs. Both machines are 'optimized'.

Any thoughts?

Thanks
____________

Profile MikeProject donor
Volunteer tester
Avatar
Send message
Joined: 17 Feb 01
Posts: 24500
Credit: 33,831,556
RAC: 24,059
Germany
Message 1458569 - Posted: 29 Dec 2013, 22:02:43 UTC

No the work units are similar.
Just different angle ranges.
But has nothing to do with the cuda version.
____________

Richard HaselgroveProject donor
Volunteer tester
Send message
Joined: 4 Jul 99
Posts: 8631
Credit: 51,493,013
RAC: 48,432
United Kingdom
Message 1458570 - Posted: 29 Dec 2013, 22:02:59 UTC - in response to Message 1458567.
Last modified: 29 Dec 2013, 22:03:39 UTC

The work units are identical.

The cuda42 application is better matched to the GTX 560 hardware: the cuda50 application is 'better' (but not ideally) matched to the GTX 780 hardware.

Profile Michael BantaProject donor
Avatar
Send message
Joined: 16 Jul 01
Posts: 100
Credit: 5,410,124
RAC: 41
United States
Message 1458571 - Posted: 29 Dec 2013, 22:03:39 UTC - in response to Message 1458569.

The cuda42 units are smaller?
____________

Profile Wiggo
Avatar
Send message
Joined: 24 Jan 00
Posts: 7330
Credit: 96,769,941
RAC: 67,461
Australia
Message 1458581 - Posted: 29 Dec 2013, 22:21:44 UTC

No, there is no difference in size, but there are short, medium and long workunits which all depend on the angle range contained in them.

Cheers.

Profile Michael BantaProject donor
Avatar
Send message
Joined: 16 Jul 01
Posts: 100
Credit: 5,410,124
RAC: 41
United States
Message 1458595 - Posted: 29 Dec 2013, 23:44:27 UTC - in response to Message 1458581.
Last modified: 29 Dec 2013, 23:44:57 UTC

Which means they can take longer to process?
____________

Profile Wiggo
Avatar
Send message
Joined: 24 Jan 00
Posts: 7330
Credit: 96,769,941
RAC: 67,461
Australia
Message 1458597 - Posted: 29 Dec 2013, 23:52:00 UTC
Last modified: 29 Dec 2013, 23:53:54 UTC

Yes, Vlars take the longest while Vhars are the shortest while the rest fall somewhere in between the 2.

[edit] Currently Vlars are not sent to GPU's though it has been suggested many times that there should be an "opt in" for these. [/edit]

Cheers.

Profile petri33
Volunteer tester
Send message
Joined: 6 Jun 02
Posts: 382
Credit: 69,677,927
RAC: 122,115
Finland
Message 1458626 - Posted: 30 Dec 2013, 2:11:04 UTC

What is the current AR limit for for a task to be labelled as vlar?
____________

Josef W. SegurProject donor
Volunteer developer
Volunteer tester
Send message
Joined: 30 Oct 99
Posts: 4301
Credit: 1,070,204
RAC: 1,104
United States
Message 1458634 - Posted: 30 Dec 2013, 3:34:06 UTC - in response to Message 1458626.

What is the current AR limit for for a task to be labelled as vlar?

AR less than 0.120000, and that's hard coded in the splitter source so could only be changed by modifying the source and recompiling.
Joe

Message boards : Number crunching : CUDA32 vs CUDA50

Copyright © 2014 University of California