GPUGrid, Linux

[MindCrime]

Currently I have a dual core opteron with a gtx 560 sitting here that I'd like to utilize. It has been running SETI and WCG for the past few days without issues. On seti I get Cuda and a couple different openCL apps that I can crunch without issue. Today I decided to get GPUgrid setup, first issue was gpugrid download problems. Now I get pure compute errors. I know in the past I've been able to run gpugrid on linux with this card but this is a new install. I feel like theres some library im missing but anyway here's the message im getting from the errored WUs:

Stderr output
<core_client_version>7.2.42</core_client_version>
<![CDATA[
<message>
process exited with code 197 (0xc5, -59)
</message>
<stderr_txt>
# SWAN Device 0 :
# Name : GeForce GTX 560
# ECC : Disabled
# Global mem : 958MB
# Capability : 2.1
# PCI ID : 0000:05:00.0
# Device clock : 1701MHz
# Memory clock : 2052MHz
# Memory width : 256bit
#SWAN: FATAL: cannot find image for module [.nonbonded.cu.] for device version 210

</stderr_txt>
]]>

[nanoprobe]

Wish I could offer some help. The only time I had success with Nvidia and Linux was after I installed the synaptic download package and let it get me the right drivers, libs, whatever. The download issue is GPUGrid's problem not yours. It's been going on for a long time and they seem to just ignore it. I don't run their tasks any more because I'm not going to wait hours for a download to finish and then get 0 points because a task errors out after 8-10 hours. They finally have a pascal app that seems to be working well but until they do something about the download issue I won't go back.

[MindCrime]

Wish I could offer some help. The only time I had success with Nvidia and Linux was after I installed the synaptic download package and let it get me the right drivers, libs, whatever. The download issue is GPUGrid's problem not yours. It's been going on for a long time and they seem to just ignore it. I don't run their tasks any more because I'm not going to wait hours for a download to finish and then get 0 points because a task errors out after 8-10 hours. They finally have a pascal app that seems to be working well but until they do something about the download issue I won't go back.

Well after sifting through the forums and google it's possible that the Fermi's compute capability has fallen behind. Pretty sure the "210" in "#SWAN: FATAL: cannot find image for module [.nonbonded.cu.] for device version 210" is C.C. 2.1 Which i read is too old but I see some other Fermi cards still producing. Not sure whats sup.

Here's a good thread to keep around:

http://www.gpugrid.net/forum_thread.php?id=2507&nowrap=true#42173

[Fire$torm]

Well after sifting through the forums and google it's possible that the Fermi's compute capability has fallen behind. Pretty sure the "210" in "#SWAN: FATAL: cannot find image for module [.nonbonded.cu.] for device version 210" is C.C. 2.1 Which i read is too old but I see some other Fermi cards still producing. Not sure whats sup.

Here's a good thread to keep around:

http://www.gpugrid.net/forum_thread.php?id=2507&nowrap=true#42173



I seem to recall something about GTX cards needing to run at stock clocks on certain projects. Not sure if this applies to you.

[MindCrime]

I seem to recall something about GTX cards needing to run at stock clocks on certain projects. Not sure if this applies to you.

These instantly error out like it's about to start the WU but finds it's incapable of running. Whats funny is the WU is sent to you even though you can't process it. The compute.capability. required to run a particular WU is determined by that WU and it can vary from WU to WU. I guess there are still some WUs out there that can be done with cc 2.1 but they're few and far between now. I'd like to know why their research requires such capable hardware compared to all the other projects, especially since they still don't have an ati/amd app.