GTX580 runs 2 work units
2 Jun 2012 14:32:45 UTC
Topic 196371
(moderation:
On Thursday 5/31 I rec'd a new MSI N580GTX video card to replace my MSI N460GTX Hawk. The 460 worked flawlessly for more than 1 1/2 years. Just wanted more CUDA speed. I installed the new 580GTX and it worked 1st try. Draws more than 100W extra power. But I did not buy it to save on electricity. It worked fine on every @Home project I tried (Seti, PrimeGrid, Milkway. Have yet to try GPUGRID) but when I started working with Einstein@Home data it immediately kicked off 2 work units. Several weeks back, when I installed BOINC 7.0.25 I found a new preference in Einstein Computing Preferences entitled GPU Utilization Factor. So I had changed it from the default of "1" to "0.5" in hopes of running 2 work units at the same time on my 460. It did not work but I forgot to change it back. So it was still set to "0.50" when I started processing Einstein work units on Friday morning. Sometime around noon on Friday I remembered the setting and checked it out. Changed it back to "1" and restarted BIONC. No difference, still processed 2 WUs. They showed up as "0.534? CPU + 0.5 NIVIDA GPU". A single WU runs in about 27-28 m on the 580GTX while 2 at a time take about 40-41 minutes each to complete. So I thought that might be pretty good deal. A single WU on the 460GTX took about 33 minutes. BUT, several times the 580GTX stopped working and I had to reboot. Once it took less than 1 minute to stop working. So I turned the machine off for the night and slept on it. I reloaded my 301.42 drivers just in case they put some hooks into the new hardware but that did not seem to help any. So last night, 6/1, I suspended all but 1 E@H WU forcing it to run only 1 at a time. Not more stoppages! So I feel that worked. Somehow, either my new card or the E@H app probably is the source of my problem. In either case, I will be running a single WU from now on. But I don't want to handle that manually.
SO, here's my question: Since I have the GPU Util Factor set to "1", is there something I've missed elsewhere that still allows E@H to process 2 WUs on my 580gtx, or, does the E@H app just run 2 WUs automatically?
My machine is:
AMD FX-6100 X6 3.3 MHz running at 4.2 MHz (Very stable!)
MSI 990FXA-DG65 m/b w 16GB Corsair Vengeange 1866 MHz memory running at 1600 MHz.
MSI N580GTX Video card w 1.5 GB DDR5 mem (Running stock @ 800 MHz)
Here is a copy of my Account.Einstein.Phys.Edu file:
- http://einstein.phys.uwm.edu/ 945f4452503b136f5d72e2db92781ac8 Einstein@Home - 100 1 1 0 1 1 - 1 1 0 - 100 1 1 0 1 1 - 1 1 0
TIA for any further clues to the solution of my problem. Dave
GTX580 runs 2 work units
)
Did you click update at the Einstein project? How is your host to know that you're changed the preference?
What Power Supply, you're just increased the load by changing to a GTX580, then you're increased the load further by running 2 BRP3 Wu's?
DON'T POST ACCOUNT FILES, it contains your authenticator!!!!
Claggy
Clicking 'update' alone isn't
)
Clicking 'update' alone isn't enough to activate the setting. It takes effect the next time work is allocated for GPU computation.
RE: Clicking 'update' alone
)
You're right, he needs fresh GPU work before it'll work eithier way.
Claggy
THANKS for the help! My
)
THANKS for the help! My power supply is an OCZ Fatl1ty 750W which was new when I put the GTX460 in the machine in 12/2010. I had thought about the # of WUs to be processed being set when the units were downloaded. So I have yet to get any since that last batch. Presently I'm running some GPUGRID work and so far so good. I'll get back to E@H next.
Here's another piece of the puzzle, maybe. When the box was delivered I noticed several labels pasted over one another. So yesterday late in the afternoon I began to remove the labels using OOPS! to soften the adhesive. I found 3 in all. The top label contained my boards serial number. The other two each contained different serial numbers. So it would appear that box has held at least two other GTX580 boards in the past. That sounds like at least two boards have been returned with problems. And some place I believe I saw "Rev. 1.0". That means my "new" board is an early board and not the latest off the production line as it should be. AARGH!! E-retailers!!!!!!!
Thanks again for the quick response & clues. I think I'll interrupt the GPUGRID work & try some more E@H now. Dave
SO much to think about at one
)
SO much to think about at one time!
Still waiting for work - servers are down for maintenance.
The max temp I've seen via Afterburner is 71C with ~ 87% GPU Useage during the processing of the two E@H WUs. A single E@H runs at 61C with ~ 65% GPU Useage. Running PrimeGrid WUs I've seen 71C with GPU Useage at 99%. The board is rated at 97C so I believe it is OK in the heat department even though I'd like to see it much lower.
My UPS watt meter shows that under stress only about 497 watts are being used. I've never seen it above 500 watts. And that number includes my monitor which is about 135 watts and plugged into a surge protector so it is not part of the P/S output. So the actual P/S useage should be under 400 watts.
My CPU is cooled by water and normally sits between 95F - 104F. Under stress of a full 6 core CPU workload it might reach ~120F. Right now it shows 100F. So that should not be a problem area either.
It will probably be this evening or tomorrow before I get some more work. Thanks again all!!! Dave
FINALLY I got some more work
)
FINALLY I got some more work from E@H. They are running 1 at a time now just as you said. So now all is well again in computerland. Thanks to all who pitched in to help. It's nice to have your help available when needed. At the very least I now know that I can run 2 WUs at the same time on my GTX580.
Dave