Hi there,
recently i've upgraded my graphics card to gtx 680 and i can't force it to start crunching, it downloads all the files snd starting to count and after 5 seconds i get an error. Is it fault of GPUgrid software or something wrong with the gtx? Is it not compatible with older cuda driver or what? I'm using latest driver 301.10.

Best regards

See this thread:

http://www.gpugrid.net/forum_thread.php?id=2916

Regards
Zy

If you wish to crunch with one, the ONLY project that accepts them currently are primegrids sieve tasks. That's where I have mine until GPUgrid or WCG release their new app.

Also works fine on Seti - mine has been happily crunching now for several weeks.

Einstein just allowed several out for those using 300x, managed to grab some and report back that 300 series monitor sleep/resume does not seem to be an issue (VERY happy im off PrimeGrid) : )

Only reason I didn't like SETI was they kept sending me CPU WU and their server was slow for me in US. Didn't like seti taking priority of WCG personally

If you wish to crunch with one, the ONLY project that accepts them currently are primegrids sieve tasks. That's where I have mine until GPUgrid or WCG release their new app.

Einstein just allowed several out for those using 300x, managed to grab some and report back that 300 series monitor sleep/resume does not seem to be an issue

Also works fine on Seti - mine has been happily crunching now for several weeks.

On Einstein, it appears as though the program is designed for CC1.x so their site does not use its full potential and they can run multiple WU on one GPU w/ no way of knowing this on status page, but here are some results:

First sets are not mine, and were run in Linux, 580 vs 680 No OC

On a 3930K @4.4= (3 tasks per GPU)

2604 sec
2610 sec
2559 sec

On his 580 in 920 @4.2: 3836 Sec Avg (850 MHz)with 3 running, and he stated his lowest he's ever seen was 3000 for one task.

I did not know you could run multiples, but running one for me at stock with i2500k @4.2 =1350 with no task running on other cores and 1550 with all 4 cores loaded with WCG. Since they load multiples on one GPU at Einstein, it's hard for me to find a direct comparison, since I don't know if they're loaded with multiple WU or not in status page.

Kind of hard to say from Einstein info since they're compiled for 1.x, and since many load multiples. Their WU seem to max out when running one at a time as seen here http://einstein.phys.uwm.edu/forum_thread.php?id=9097 But when they start adding more work per GPU it appears it works well (for running on an "old" app) Many seem to prefer lower end cards

At PrimeGrid (not me again)

CW Sieve: gtx580 overclocked to 860MHz = 304s
gtx680 = 265s.

From Robert7NB1 from GPU force with 1,000,000,000 points =

We tested the new NV. In the application tpsieve, NV680@1097 (compute 3.0) is faster than NV570@840 (compute 2.0), by 44% (12.94/9.0M p/s).

I've myself have quite a few in PV at Einstein (some are against a 590), however I will not really believe anything from what I see on these sites, since they have not been run on PCI 3.0, CC 2.0+ etc., these are small tasks that are designed for older cards. Whenever more get validated from Einstein, I'll report back. Further my gpu utilization is only 70% I believe GPUgrid will be first project with good benchmarks,.but I won't have ivy w/ 3.0 until end of April when its released

As a side note. From what I've seen and read on non optimized projects at worst its a very efficient 580. Again once GPUgrid app is out I plan on running a much more extensive review with current setup, and another one at end of month. By then hopefully GPUgrid will have an OPTIMIZED app for 680 and not one it just "runs on".

EDIT: From MartinBen (ahem from above lol) on SETI:

Yes I have re. Seti !

GPU work units are quicker - down from 3m 40s on the 580 to just 3m 10s on the 680 before any overclock is applied. With a mild overclock I have got this down to about 2m 50s

So with mild OC= 25%, but performance/watt would be sig. higher than 580, and again a non-optimized app.

I've decided to not post differences in runtimes from Einstein, b/c it will be an irrelevant comparison. Their CC 1.x ALWAYS seem to finish around the same time for anything above a 560 (448). Whenever Windows is released, or if I can ever figure out this GRUB issue (haven't tried 12.04 yet but have it on USB), I will post 680 review using an i5@4.2 (of course w/ PCI2.0), and then I will do a review using Ivy i7 w/ PCI 3.0.

Hopefully Ivy is released on April 29th, and my local Microcenter has them in stock.

As a strange sidenote, it would appear as though my GPU boost has either stopped working, or maybe NVIDIA did put an INCREDIBLY long "boost timer" on the card, because after roughly a week of crunching, it is no longer boosting. Temps NEVER got higher than 55, but I am somewhat hesitant on rebooting, since I did not like my voltage 1.175 24/7. Now it's at stock speeds using 1.087V. Oh and my power % decreased from 70% to 50%, with GPU utiliaztion the same as before, 70%, it's what their apps work on, and is why some people load multiples.

Until then, its looking for pulsars.

Yes, but i don't know of its the way its coded, or if gpus there have hit their "wall" but as I've said drawing comparisons between different high end gpus only matters when u load multiples and how many you.can load. I have heard that a 680 was up to 4 at a time, and I'm depleting stock to switch to 7.0.25 boinc and try this multi Wu thing out. I truly think GPUgrid will be best place to bench 680 crunching when it arrives.

Have u also allready read about the "Red Screen of Death" ?