Re: [HTCondor-users] GPU discovery - empty results with older cards on 8.6.13

[Michael Pelletier <Michael.V.Pelletier@xxxxxxxxxxxx>]

Thanks, TJ! Found the problem - my LD_LIBRARY_PATH in my shell was masking the fact that there was an ld.so.conf.d file covering the CUDA 9.0 bundle that was installed on the system a while back. Once I removed that and ran ldconfig, the discovery ran without the path variable set, as it would under the startd, and now it's advertising things properly. Thanks!

Michael V. Pelletier
Information Technology
Digital Transformation & Innovation
Integrated Defense Systems
Raytheon Company

-----Original Message-----
From: HTCondor-users <htcondor-users-bounces@xxxxxxxxxxx> On Behalf Of John M Knoeller
Sent: Thursday, May 16, 2019 5:24 PM
To: HTCondor-Users Mail List <htcondor-users@xxxxxxxxxxx>
Subject: [External] Re: [HTCondor-users] GPU discovery - empty results with older cards on 8.6.13

it's probably a problem with the path that is in effect when gpu discovery is run under the startd.

why don't you replace condor_gpu_discovery with a wrapper script that runs the real discovery with the -diag and -verbose flags
and dumps the output to a file.   I suspect it will say that it can't find the cuda runtime library or that there is a mismatch between
library and driver.

-tj