New App v4.46 Released for GTX 670 and GTX 680

[Snow Crash]

http://www.gpugrid.net/forum_thread.php?id=3025#25656

I have uploaded the new cuda4.2 app for the acemd2 queue. Later in the afternoon the Linux app.
The long queue will join in a week if these tests go well.

Please post here any problem.

gdf

The new version of the application will automatically be downloaded and is the only version for the GTX6XX series of cards.
This version of the app runs faster than the old version so it runs bit hotter.

Per GDF's post, you will need to check the
'ACEMD standard:' box to get WUs as they have not released this for the longs yet.

Fire those awesome new cards up and get crunching!!!

[edit]Thanks to XS team member 5pot - uncheck beta and uncheck run test application.[/edit]
[edit]"but also uses a full CPU thread for processing" ... NOT true![/edit]

[Snow Crash]

http://www.gpugrid.net/forum_thread.php?id=3037

We have finally uploaded the new applications for Kepler supporting cuda4.2.

You should be receiving them if you have a recent driver. The new applications are substantially faster.

gdf

I'm now running them on my 295, 480 and 670 ... it's all good ... we should see the team RAC jump as soon as everyone moves to only doing the long WUs again!!!

[FaYt]

how can i set the boinc manager to do long runs only??

[Snow Crash]

Hi FaYt you set it on your GPUGrid preferences page
http://www.gpugrid.net/prefs.php?subset=project

Run only the selected applications
ACEMD standard: no
ACEMD beta: no
ACEMD for long runs (8-12 hours on fastest GPU): yes

[FaYt]

Done

[werdwerdus]

GT240 supported? Long runs would take ages if not..

[Snow Crash]

The 2XX series does not benefit like fermi and kepler. In fact some WUs are a little slower and some are a little faster (IIRC the NATHAN WUs are a little slower). If I were you I would probably do the short runs on a GT240 as the potential for losing days on a single crash would gut me.

In a brief t-storm yesterday I lost about 24 hours all together ... 17+ hrs on a NATHAN that was running on my GTX 295 which would have finished in another 1.5 hrs.

[werdwerdus]

not sure how to set preferences per machine on GPUGrid

edit: nevermind I found the home/work/school profiles

[ChaosAD]

Just finished my first ever gpugrid long run task with my 670 at stock clocks in 4h 20m. Is this ok? I fold in a 30C room and my gfx temps is 63C with the fan at 70%. Is this temp safe for 24/7 folding?

[Snow Crash]

You got it just right for a first run
Let us know if / when you want to start OCing that bad boy and you can get that time down to 4hrs. flat

[ChaosAD]

Nice to find you all. Gfx will crunch overclocked in winter sadly. To help temps some, do gpugrid need both gpu and mem, or can i lower any of the two?

edit: i noticed in afterburner that gpu load is 93-95% and never reached 100%, is this normal? i also run wcg 4c/8t at the same time.

[werdwerdus]

that's normal. AFAIK you can lower memory speed without much effect on run times

[Snow Crash]

i also run wcg 4c/8t at the same time.

Because you are running Win7x64, GPUGrid will completely use 1 of your threads for processing so you might as well make your general BOINC settings use 99% of your CPUs. Otherwise the runtimes on your WCG WUs are going to be slow as you're not really processing 8 WUs at the exact same time and due to continuous context shifting between too many processes running your likely a bit less efficient overall.

[ChaosAD]

Is it normal that some projects have more gpu usage than others? Also, when i run WCG with all 4c/8t i see 73-74% gpu usage, but when i stop WCG completely i see 81-82% gpu usage. I tried what Snow suggested but i saw no difference when i set 99-95 or 90% cpu usage. Can i do something to rum both wcg and at the same time utilize my gpu at 100%? Maybe dedicate 1 cpu thread for gpugrid? But then wouldnt that result lower total output?

[werdwerdus]

different projects and different cards will have different gpu utilization. most of them are around 94% on my GTX 470 cards. You can set BOINC to use 99% of processors and this will drop one core from being used for WCG (unless you have more than 100 processors )