Increased Credit Values for GPU Jobs

Message boards : News : Increased Credit Values for GPU Jobs

To post messages, you must log in.

AuthorMessage
Thomas Koch
Project administrator
Project developer
Project scientist

Send message
Joined: 17 Feb 12
Posts: 436
Credit: 37,847
RAC: 0
Message 10433 - Posted: 23 Apr 2015, 10:00:27 UTC

We have noticed increasing runtimes for 1vii and 2f21 work units due to conformational changes to the system, so the credit value for these jobs has been adjusted. Further adjustments may follow.

Wir haben festgestellt, dass sich die Laufzeiten der 1VII und 2F21 Workunits wegen Änderungen an der Konformation des Systems verlängert haben, deshalb wurde der Credit-Wert dieser Jobs angepasst. Eventuell folgen weitere Anpassungen.
ID: 10433 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
tbritt

Send message
Joined: 11 Oct 12
Posts: 1
Credit: 16,442,882
RAC: 0
Message 10437 - Posted: 23 Apr 2015, 20:26:07 UTC

Jobs running for 29 hours are ridiculous. Anything over 8 hours is unworkable. Break them down into smaller units or they will be aborted by user.
ID: 10437 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
floyd

Send message
Joined: 17 Sep 11
Posts: 15
Credit: 14,326,640
RAC: 0
Message 10438 - Posted: 24 Apr 2015, 9:27:35 UTC - in response to Message 10437.  

Jobs running for 29 hours are ridiculous. Anything over 8 hours is unworkable. Break them down into smaller units or they will be aborted by user.

If there are frequent checkpoints and I'm given enough time, why wouldn't I run long tasks? Besides, it seems it's just your GT620 being slow.
ID: 10438 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Thomas Koch
Project administrator
Project developer
Project scientist

Send message
Joined: 17 Feb 12
Posts: 436
Credit: 37,847
RAC: 0
Message 10439 - Posted: 24 Apr 2015, 9:32:11 UTC

As you can see by comparing the runtimes on your own hosts, the times for GPU calculations are highly device dependant. As we are shipping work units of a fixed simulation time, we won't be able to perfectly fit runtimes on all available devices.

However there are ways to improve the runtime. Running OpenCL jobs on Nvidia GPUs, the driver needs a full CPU core to keep the GPU fed. If you are running CPU jobs in parallel, you can improve GPU job runtimes by leaving one spare CPU core.
Also keep in mind that BOINC using its default configuration will suspend GPU work as soon as the computer is in use. This will lead to much checkpointing, and also increase the overall runtime.
ID: 10439 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Vladimir Tankovich
Volunteer developer
Volunteer tester

Send message
Joined: 12 Nov 10
Posts: 182
Credit: 429,219,133
RAC: 0
Message 10443 - Posted: 24 Apr 2015, 16:50:31 UTC - in response to Message 10439.  

I see one of your hosts having 620 GT. that's a very old and a very small card and it's expected to have 60X longer runtime than modern 200$ cards. considering compute it delivers and power it consumes, I'd say it's not cost-efficient to run it (savings from electric bill over a year could be spent on upgrades).

PS. this thread was hijacked :)
ID: 10443 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
hunter1978

Send message
Joined: 6 Oct 12
Posts: 5
Credit: 385,030,676
RAC: 0
Message 10447 - Posted: 24 Apr 2015, 19:10:02 UTC
Last modified: 24 Apr 2015, 19:18:57 UTC

WOW 29 hours. Running a Sapphire TRIXX r9 290. Finishing either side of 30 minutes per task. Also running 8 tasks on WCG at the same time. CPU i7-3770K.
ID: 10447 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
User

Send message
Joined: 11 Mar 15
Posts: 2
Credit: 2,749,749
RAC: 0
Message 10457 - Posted: 28 Apr 2015, 9:50:53 UTC
Last modified: 28 Apr 2015, 9:55:38 UTC

FAO Thomas Koch.

How strict are you on deadlines? I have a work unit running on opencl_ati_101, processing time thus far 80 hours, time to completion 262 hours which is 10.75 hours after deadline.

Is this going to be of any use to the project, or should I abort now?
ID: 10457 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Thomas Koch
Project administrator
Project developer
Project scientist

Send message
Joined: 17 Feb 12
Posts: 436
Credit: 37,847
RAC: 0
Message 10461 - Posted: 29 Apr 2015, 9:50:36 UTC

Hello User,

as our project depends on generating new work units from returned ones, we have strict deadlines.
The time remaining is not an exact value, but with such long runtimes, I'd recommend to abort the job, because chances are high it will be rejected by the server.

As far as I can see, you are running our jobs on your E-300 APU, which is definitely not made for such work.

Best,
Thomas
ID: 10461 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Jim1348

Send message
Joined: 4 Jul 12
Posts: 104
Credit: 195,960,357
RAC: 0
Message 10462 - Posted: 29 Apr 2015, 14:55:24 UTC

The longest that I see on my GTX 750 Ti is 2 hours 7 minutes, and the shortest 1 hour 40 minutes. So it all works fine for me; my HD 7790 was somewhat faster, but at correspondingly more power. It all scaled very well. So the real key is to have a newer card (Nvidia Maxwell or AMD HD 7000 or later), as the developer posted at the beginning.
ID: 10462 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
User

Send message
Joined: 11 Mar 15
Posts: 2
Credit: 2,749,749
RAC: 0
Message 10463 - Posted: 29 Apr 2015, 17:07:02 UTC - in response to Message 10461.  

Thanks, will so do.
ID: 10463 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Ramy O

Send message
Joined: 31 Dec 15
Posts: 1
Credit: 0
RAC: 0
Message 10831 - Posted: 31 Dec 2015, 9:05:21 UTC

Hello,

I just wanted to ask if Intel 4000 is compatible?
You indicated that you are implementing support for Intel in a thread in News section;
I am not sure if it is compatible yet or still in testing?

Thanks.
Please donate to my Mathematics Papers.
ID: 10831 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Thomas Koch
Project administrator
Project developer
Project scientist

Send message
Joined: 17 Feb 12
Posts: 436
Credit: 37,847
RAC: 0
Message 10836 - Posted: 13 Jan 2016, 14:43:35 UTC

I'm afraid the Intel release is currently put on hold.
We had a short public test phase, which pointed out errors we couldn't fix yet.
The investigation of this problem isn't set on high priority, as Intel OnChip-GPUs increase in capability, but still don't compete with dedicated GPUs.
ID: 10836 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote

Message boards : News : Increased Credit Values for GPU Jobs


Copyright © 2017 KIT-INT