I am running this on linux centos with cuda-12.6 toolkit I am trying cupti sample codes pm_sampling but it is giving me error when i am executing the executable “Function cuptiPmSamplingGetCounterAvailability(&getCounterAvailabilityParams) failed with error(1): CUPTI_ERROR_INVALID_PARAMETER.”
what could be the problem .
Can you please check below info ?
- Driver version
- GPU
- What’s the result of other CUPTI sample ?
Driver Version: 550.90.07
gpu: Tesla V100-SXM2-16GB
i am running this “./pm_sampling -d 0 -c GV100 -q“
error: “Error:pm_sampling.h:354: Function cuptiPmSamplingGetCounterAvailability(&getCounterAvailabilityParams) failed with error(1): CUPTI_ERROR_INVALID_PARAMETER.“
Can you just run “pm_sampling” without any parameter ?
Please also run activity_trace_async and concurrent_profiling to check if there is similar issue.
when running only pm_sampling i am facing same error.
As for activity_trace_asycn and concurrent_profiling there are running fine.
Thanks for the info provided.
You mentioned you are using CUDA 12.6, but with driver 550.90.07.
As CUDA 12.6 is released with R560 instead of R550.
So I need to know how do you config your env ? Is it a forward compatibility config (UMD R560+KMD R550) or a minor version compatibility config (UMD R550+KMD R550) ? CUDA Compatibility — CUDA Compatibility
no it is not forward compatibility config can that cause an issue ? but other samples are working fine.
PM Sampling is supported from Turing onwards. Not supported in Volta.
From public documentation page:
These APIs are supported on Turing and later GPU architectures, i.e. devices with compute capability 7.5 and higher.
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.