Can't track metrics with nvprof 12.9 on 1080TI

I have been following a CUDA refresher course but have run into some trouble when trying to profile my code. If I compile the solution code and run nvprof ./matrix_sums, I get a profiling result as expected. However, if I try and add any metrics, e.g. nvprof --metrics gld_transactions ./matrix_sums, I get
==4481== NVPROF is profiling process 4481, command: ./matrix_sums
==4481== Error: Internal profiling error 4212:27.
row sums correct!
column sums correct!
======== Profiling result:
No events/metrics were profiled.
======== Error: CUDA profiling error.

Looking up the error code in CUPTI suggests that it is CUPTI_ERROR_NOT_SUPPORTED, but it’s not clear exactly what isn’t supported. I have tried running with sudo and following the steps here so it does not appear to be a permission issue. My GPU is a 1080Ti and I am using the following software versions:

$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2025 NVIDIA Corporation
Built on Tue_May_27_02:21:03_PDT_2025
Cuda compilation tools, release 12.9, V12.9.86
Build cuda_12.9.r12.9/compiler.36037853_0
$ nvprof --version
nvprof: NVIDIA (R) Cuda command line profiler
Copyright (c) 2012 - 2025 NVIDIA Corporation
Release version 12.9.79 (21)

I don’t think newer versions (or Nsight Compute) support my GPU. My OS is Arch Linux. Any help would be very welcome!

Hi, @halawiye

Sorry for the issue you met.

Can you tell your driver version ?
Is your sample built with 12.9 nvcc compiler ?

Also please try
$nvprof -e inst_executed
$nvprof -m gld_efficiency,shared_efficiency

Thanks for your reply.

My driver version is 580.82.09. Is it possible that this is too new? I haven’t frozen it like I have with the CUDA compiler and tools.

The sample is built with the 12.9 nvcc compiler

Here is the output for the commands you requested:
$ nvprof -e inst_executed ./matrix_sums
======== Warning: Event "inst_executed" cannot be found on device 0.
==3605== NVPROF is profiling process 3605, command: ./matrix_sums
row sums correct!
column sums correct!
==3605== Profiling application: ./matrix_sums
==3605== Profiling result:
No events/metrics were profiled.
$ nvprof -m gld_efficiency,shared_efficiency ./matrix_sums
==3684== NVPROF is profiling process 3684, command: ./matrix_sums
==3684== Error: Internal profiling error 4212:27.
row sums correct!
column sums correct!
======== Profiling result:
No events/metrics were profiled.
======== Error: CUDA profiling error.

Hi, @halawiye

580 driver is released with CUDA 13.0, and nvprof has been dropped in CUDA 13.0.
So yes, this might be the problem.

Please try with R575 drivers.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.