Open fertiliz opened 2 weeks ago
I got the same error within apptainer.
May I ask where warp find the libs ? (the path)
my cuda is cuda_11.3.r11.3/compiler.29920130_0 my warp is 1.0.0
Looking forward to the reply,
Best,
Thank you for reporting this. We just load libcuda.so
dynamically using dlopen()
, without specifying a directory.
The error you're getting is defined as:
* This indicates that the CUDA driver that the application has loaded is a
* stub library. Applications that run with the stub rather than a real
* driver loaded will result in CUDA API returning this error.
*/
CUDA_ERROR_STUB_LIBRARY = 34,
The stub library is essentially an empty version of libcuda.so
that comes with the CUDA Toolkit and that people can link their applications against so they don't actually need to have the driver installed on the build system; only the system on which the software will get deployed needs to have the real library under a path searched by dlopen()
.
So it appears that your system has the directory of the stub (typically /usr/local/cuda/lib64/stubs
) configured to be searched before the user-mode driver.
thanks for your reply. Do you have any suggestions to solve it? reinstall the cuda?or change the ld path?
best
zhi hou
From: Nicolas Capens @.> Sent: Friday, May 17, 2024 1:36:33 AM To: NVIDIA/warp @.> Cc: Zhi Hou @.>; Comment @.> Subject: Re: [NVIDIA/warp] Unable to determine CUDA driver version (Issue #219)
Thank you for reporting this. We just load libcuda.so dynamically using dlopen(), without specifying a directory.
The error you're getting is defined as:
* This indicates that the CUDA driver that the application has loaded is a
* stub library. Applications that run with the stub rather than a real
* driver loaded will result in CUDA API returning this error.
*/
CUDA_ERROR_STUB_LIBRARY = 34,
The stub library is essentially an empty version of libcuda.so that comes with the CUDA Toolkit and that people can link their applications against so they don't actually need to have the driver installed on the build system; only the system on which the software will get deployed needs to have the real library under a path searched by dlopen().
So it appears that your system has the directory of the stub (typically /usr/local/cuda/lib64/stubs) configured to be searched before the user-mode driver.
— Reply to this email directly, view it on GitHubhttps://github.com/NVIDIA/warp/issues/219#issuecomment-2115836403, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AQGPLYDAYNOHDITQ2O7QWZ3ZCTVCDAVCNFSM6AAAAABHZWQXC2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCMJVHAZTMNBQGM. You are receiving this because you commented.Message ID: @.***>
how to fix it? same question
(foundationpose) robot@robot-System-Product-Name:~/CODE/FoundationPose$ python Python 3.9.19 (main, May 6 2024, 19:43:03) [GCC 11.2.0] :: Anaconda, Inc. on linux Type "help", "copyright", "credits" or "license" for more information.
My nvcc -V is: nvcc: NVIDIA (R) Cuda compiler driver Copyright (c) 2005-2022 NVIDIA Corporation Built on Wed_Sep_21_10:33:58_PDT_2022 Cuda compilation tools, release 11.8, V11.8.89 Build cuda_11.8.r11.8/compiler.31833905_0