Einstein/GPU "GPU missing" after hardware/software upgrades
14 Dec 2014 16:39:47 UTC
Topic 197876
(moderation:
I've been running Einstein@Home, GPU-only, for several months. I recently installed a new SSD and transferred my OS (Kubuntu 14.04 64-bit) from the old platter HD. I had an issue a couple days later with an Ubuntu-pushed nVidia driver update that failed to correctly build into the (also freshly updated) kernel, so had "missing GPU" reports from Einstein along with all my desktop effects (dependent on Open GL) failing. I got the video driver problem resolved, and restored my Open GL functions, by reinstalling the updated video driver, but I'm still/again seeing "GPU missing" from Einstein.
I've checked that all my CUDA-related items are installed; as far as I can tell, they are. Suggestions on what else I can check to restore GPU operation for Einstein on nVidia 331.113 Linux-64 driver? To verify, I have packages nvidia-331, nvidia-331-uvm, libcuda1-331, nvidia-opencl-icd-331, nvidia-prime, nvidia-settings, bbswitch-dkms, along with kubuntu-driver-manager, ubuntu-drivers-common, and libvdpau1, all current versions from Kubuntu/Ubuntu repositories.
Einstein/GPU "GPU missing" after hardware/software upgrades
)
With sudo or su, run the below command:
lsmod | grep nvidia
Check to see if the kernel module nvidia_uvm is loaded. If the module is not loaded, run the below command to load the kernel module.
modprobe nvidia_uvm
Restart BOINC after running modprobe.
Jeroen
Okay, nvidia_uvm wasn't
)
Okay, nvidia_uvm wasn't running, but the only way I could find to restart BOINC after modprobe nvidia_uvm was to restart the computer. After doing that, not only didn't nvidia_uvm load during the startup (as I understood it should, after being started with modprobe previously), but the BOINC manager is reporting that "The BOINC client closed unexpectedly three times in the past two minutes" -- and BOINC manager then won't close, nor reconnect to localhost.
Edit: Next restart, the BOINC client came up normally, but still "GPU Missing" from the Einstein task, and still nvidia_uvm didn't reload during the restart, despite having (again) loaded it with modprobe nvidia_uvm before restarting. How would I restart BOINC without restarting Linux, and would that make a difference?
Further edit: Okay, I found the service name and (as su) issued service boinc-client restart, getting what appeared to be a successful client restart (no errors, good status reported); after that, BOINC manager came up automatically reconnected to the running client, but STILL (with nvidia_uvm showing loaded before and after restarting boinc-client) reports "GPU missing" for Einstein project.
NVidia screwed up the driver
)
NVidia screwed up the driver ... sounds like you need to revert to older version to get back in business.
http://einsteinathome.org/node/197867
--------------------------
- Crunch, Crunch, Crunch -
--------------------------
Cool, I'll revert to 311.38
)
Cool, I'll revert to 311.38 and wait for an update. :)
Well, there wasn't an option
)
Well, there wasn't an option to revert to 331.38 or other 331 version, so I used Driver Manager to revert to 304-updates. That means I'll have to manually watch for the 331 family to get updates, but at least it works.
RE: Well, there wasn't an
)
Unless you also game upgrading isn't always helpful, the primary market for gpu's is the high end super computers and gamers, with the software mostly being made for gamers. We crunchers are WAAAAY down the totem pole as far as things being done to make our crunching faster. It's often best to let someone else be the first to test new software when it comes out, and let them be the one to have to downgrade when it makes things worse. IF a version is really better you should start seeing posts about it on the various projects.