Folding Forum

If the problem were memory related, it would seem to affect all fahcores, not just one type, and that is not the case.

CUDA fahcore works, OpenCL fahcore does not. So far, that's the simplest explanation, making it the most likely answer.

Yes, it could be any of 6 other things that is the problem, one of them being the new shared memory. Just not as likely, IMO.

We're still looking into this. It's hard to tell if it's a driver issue or an OpenMM issue.

VijayPande wrote:We're still looking into this. It's hard to tell if it's a driver issue or an OpenMM issue.

thank you for your reply

ud

Freightanimal wrote:

Zagen30 wrote:Anyone with a 750 (Ti) could temporarily install the v6 GPU client, as it cannot get core 17 projects and would therefore only get core 15.

Thanks for the info. I personally wouldn't want to do that. Unfortunately I kind of doubt most people will either. Hopefully they can fix core 17 soon.

FAH GPU Tracker v2 http://fahtracker.com/setup.html has easily let me make use of my new card thanks for the tip about v6. I used this: http://www.overclock.net/t/1271684/wind ... kepler-gpu guide which has been perfect for kepler cards pre V7. I configured as a kepler. Could happily change to V7 when problems get ironed out, but will run this til then.

Thank you all for the help.

I changed the title again. It turns out there are two entirely different GPUs known as the 750 Ti. I'm sure this complicates the issue terribly.

Please confirm which one you're reporting.

Same thing here with a new Maxwell card (750 ti) on FaH 7.3.6: Failing after a few minutes on core 17 (v. 0.0.52), running to completion with core 15 (v 2.25 dated 9 May 2012).

I see there is a beta version of core 17 dated 14 March of this year. I set client-type=beta in advanced options and nuked the Core17_fah directory that held the FahCore_17.exe. Hopefully if I get another core 17 WU it will pull down the new core and perhaps it may work better.

^^bruce, my research comes up with a 750M that is GK106, but not any other 750ti. Nividia recently released *two* versions of the 860M, one Kepler with 1152 cores and one Maxwell with 640 cores. The 850M is a lower-clocked version of the Maxwell part, the 870M and 880M are Kepler parts with more cores.

i've set up my sig.
gm 107

ud

GM107 here

Are there any GM107's still running V7 or has everybody switched to V6?

GentleGiant:
The beta core_17 isn't supposed to fix this problem but a report of your test will confirm.

this is a backup rig and is still set up for v 7.3.6.

ud

The beta, which reports version 0.0.54 does not work either. I'm deleting my GPU slot in the version 7 client and going to mess with the v6 client. I'm thinking the command line gpu client would be easier to to get going, rather than the systray version. If anybody has experience either way I'd be happy to hear it....

Edit: addition:
Quick trial didn't work for me, got a core 11 WU:
[15:46:24] Project: 10502 (Run 157, Clone 0, Gen 1740)

[15:46:30] Calling fah_main args: 14 usage=100
[15:46:30]
[15:46:33] CoreStatus = 63 (99)
[15:46:33] + Error starting Folding@Home core.

After trying a couple of times it tries to download the core again. My "installation method" was just to create a directory in AppData\Roaming and unzip Folding@home-Win32-GPU_XP-641.zip there, start a command shell in that directory and run "Folding@home-Win32-gpu -configonly" and then "Folding@home-Win32-gpu" with no arguments.

It sounds like you got project 10502 using V6 ... or did the V7 slot do that?

You probably have to use -forcegpu nvidia_fermi, though nobody who is successfully using V6 has mentioned that yet.

GM107 here & i've switched to GPU Tracker V2 which pumps out nothing but 8018's

That switch did the trick -- I killed off a few stray FahCore_11 processes, then deleted work directory and queue.dat, and restarted the V6 client with -forcegpu nvidia_fermi. It started a project 8018 core 15 WU and is happily crunching...

The fahtracker.com site tofuwombat mentions above was down, or I might have tried that route.

OK. The V6 option works for Windows. Let's call it a temporary solution until a better choice becomes available. Linux folks will have to wait.

I have no ETA for when more options will be available. The only information is what Dr. Pande posted above.

Would somebody with a GM107 try V7 again and post the applicable segment of the log. It has probably changed since the logs that were posted above.