Advanced search

Message boards : Graphics cards (GPUs) : Strange Calculation Behaviour

Author Message
Profile [AF>Amis des Lapins] Phil...
Send message
Joined: 16 Jul 13
Posts: 56
Credit: 1,626,354,890
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 34225 - Posted: 12 Dec 2013 | 6:25:28 UTC

Hello !
I am currently running 77x-SANTI_MARwtcap310-1-32-RND1129_0
Unité de travail (WU) 4985773.
First time I see this, but the GPU charge (graph) is sinusoidal.

Instead of being stable at 98 % or 99 % :

X seconds at 50 % or less, followed by Y seconds at 99 % etc

=> after 13 hours on a GTX660Ti : 91 % only, while this card
normally takes between 10 up to 12 hours to completed long runs.

Any idea about the origin of this "strange" behaviour ?

Thank You, Kind Regards,
Philippe

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 34257 - Posted: 12 Dec 2013 | 19:29:46 UTC - in response to Message 34225.

If by 'charge' you mean power then it should be linear or very close to it.
If it's a square-toothed graph then it's likely that something else is interfering with GPU crunching, the GPU itself, or there might be something wrong with the WU.
Sinusoidal, as in sine wave (http://people.sc.fsu.edu/~jburkardt/latex/svd_2006_vt/inout_flow_sinusoid.png) would be very odd!

What software are you using to view the graph?
What else are you crunching and what Boinc settings are you using?
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Profile [AF>Amis des Lapins] Phil...
Send message
Joined: 16 Jul 13
Posts: 56
Credit: 1,626,354,890
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 34292 - Posted: 14 Dec 2013 | 11:00:16 UTC - in response to Message 34257.
Last modified: 14 Dec 2013 | 11:01:59 UTC

Thank you for your message.
After a reboot, thinks look fine now.
However, this UT took much more time to complete ... (WU) 4985773

http://www.gpugrid.net/result.php?resultid=7540482

Yes, the GPU charge indicator (EVGA Precision X) showed
a sinusoidal use i/o linear ...

Made neither change to the PC nor to BOINC ...

Posted this message as this was the first time I saw such behaviour ...

GPU = GTX660Ti - PNY

Again, Thank You

Kind Regards

Philippe

Dagorath
Send message
Joined: 16 Mar 11
Posts: 509
Credit: 179,005,236
RAC: 0
Level
Ile
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 34294 - Posted: 14 Dec 2013 | 11:45:32 UTC - in response to Message 34292.

I think by GPU charge he means GPU utilization rather than something electrical. My French is not very good but I think charge can mean something like usage or utilization in French.

____________
BOINC <<--- credit whores, pedants, alien hunters

VT
Send message
Joined: 28 Nov 13
Posts: 17
Credit: 153,786,987
RAC: 0
Level
Ile
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 34340 - Posted: 16 Dec 2013 | 8:11:10 UTC - in response to Message 34294.

That tool can show a graph of electrical power consumption for the card. Even if clock stays the same(and I really mean the same - no boost or "idle" freq drop), but there is no compute load on the card, the power consumption will drop - seems like a good indicator of how busy the card is at a given moment.

That tool also can dynamically boost Memory and compute clocks - if memory load is small, it drops it's clock and can boot compute up to 1058 for Titan (from 836 advertized or 928 "official" boost).

Post to thread

Message boards : Graphics cards (GPUs) : Strange Calculation Behaviour

//