You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi.
I'm trying to run phoronix-test-suite install pytorch-1.0.1 on the Google Cloud Platform VM instance with support for GPU (Nvidia Tesla T4 15GB). However, I did not see the option to choose between CPU vs Cuda. I'm always forced to use the CPU. Phoronix did not ask me which hardware to use (GPU vs CPU).
I have made sure that PyTorch 1.0.1 supports Nvidia in its test-definition.xml. After spending some time reading the Phoronix code I think the simple validation at
// Only show NVIDIA / CUDA options when running with NVIDIA hardware
$error = 'NVIDIA support is not available.';
returnfalse;
}
doesn't work on my case.
I tried to print out the phodevi::read_property('gpu', 'model') of my VM instance, and it yields Tesla T4 15GB which does not contain substring NVIDIA in it, even though it's also an Nvidia GPU with CUDA support.
Some solutions I propose to this issue are:
Add alternative validation. If the substring "nvidia" is not found, then try to run command nvidia-smi and see if it returns "command not found error" or not.
Add an option to disable all validation entirely (which may not be an ideal solution, but easier to implement)
Rely on other properties of the GPU besides the "model" property
However, I believe there may be some more ideal and better solutions than my solutions. Should that be the case, feel free to use the better one
The text was updated successfully, but these errors were encountered:
Hzzkygcs
changed the title
Edge Case for NVIDIA Cuda Validation
NVIDIA Cuda Validation did not detect Nvidia for Tesla T4 15GB
Jan 20, 2024
Hi.
I'm trying to run
phoronix-test-suite install pytorch-1.0.1
on the Google Cloud Platform VM instance with support for GPU (Nvidia Tesla T4 15GB). However, I did not see the option to choose between CPU vs Cuda. I'm always forced to use the CPU. Phoronix did not ask me which hardware to use (GPU vs CPU).I have made sure that PyTorch 1.0.1 supports Nvidia in its
test-definition.xml
. After spending some time reading the Phoronix code I think the simple validation atphoronix-test-suite/pts-core/objects/pts_test_run_options.php
Lines 759 to 770 in f036573
I tried to print out the
phodevi::read_property('gpu', 'model')
of my VM instance, and it yieldsTesla T4 15GB
which does not contain substringNVIDIA
in it, even though it's also an Nvidia GPU with CUDA support.Some solutions I propose to this issue are:
nvidia-smi
and see if it returns "command not found error" or not.However, I believe there may be some more ideal and better solutions than my solutions. Should that be the case, feel free to use the better one
The text was updated successfully, but these errors were encountered: