nvprof does not complete without sudo
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
nvidia-cuda-toolkit (Ubuntu) |
Expired
|
Undecided
|
Unassigned |
Bug Description
Description: Ubuntu 18.04 LTS
Release: 18.04
Expected behavior: profile output
Actual behavior: error messages
Reproduce as follows:
cd NVIDIA_
nvcc -I ../../common/inc matrixMul.cu -o matrixMul
# check the exe works
./matrixMul
[Matrix Multiply Using CUDA] - Starting...
GPU Device 0: "GeForce GTX 1080" with compute capability 6.1
MatrixA(320,320), MatrixB(640,320)
Computing result using CUDA Kernel...
done
Performance= 1137.23 GFlop/s, Time= 0.115 msec, Size= 131072000 Ops, WorkgroupSize= 1024 threads/block
Checking computed result for correctness: Result = PASS
NOTE: The CUDA Samples are not meant for performance measurements. Results may vary when GPU Boost is enabled.
# now try nvprof
nvprof ./matrixMul
[Matrix Multiply Using CUDA] - Starting...
==4775== NVPROF is profiling process 4775, command: ./matrixMul
GPU Device 0: "GeForce GTX 1080" with compute capability 6.1
MatrixA(320,320), MatrixB(640,320)
Computing result using CUDA Kernel...
done
==4775== Error: Internal profiling error 4168:999.
Performance= 1130.40 GFlop/s, Time= 0.116 msec, Size= 131072000 Ops, WorkgroupSize= 1024 threads/block
Checking computed result for correctness: Result = PASS
NOTE: The CUDA Samples are not meant for performance measurements. Results may vary when GPU Boost is enabled.
======== Error: CUDA profiling error.
# run with sudo
sudo nvprof ./matrixMul
[Matrix Multiply Using CUDA] - Starting...
==4797== NVPROF is profiling process 4797, command: ./matrixMul
GPU Device 0: "GeForce GTX 1080" with compute capability 6.1
MatrixA(320,320), MatrixB(640,320)
Computing result using CUDA Kernel...
done
Performance= 1132.95 GFlop/s, Time= 0.116 msec, Size= 131072000 Ops, WorkgroupSize= 1024 threads/block
Checking computed result for correctness: Result = PASS
NOTE: The CUDA Samples are not meant for performance measurements. Results may vary when GPU Boost is enabled.
==4797== Profiling application: ./matrixMul
==4797== Profiling result:
Type Time(%) Time Calls Avg Min Max Name
GPU activities: 99.54% 34.644ms 301 115.10us 114.15us 116.07us void matrixMulCUDA<
API calls: 74.77% 110.27ms 3 36.757ms 3.4300us 110.26ms cudaMalloc
ProblemType: Bug
DistroRelease: Ubuntu 18.04
Package: nvidia-profiler 9.1.85-3
ProcVersionSign
Uname: Linux 4.15.0-20-generic x86_64
NonfreeKernelMo
ApportVersion: 2.20.9-0ubuntu7
Architecture: amd64
Date: Thu Apr 26 17:28:48 2018
Dependencies:
gcc-8-base 8-20180414-1ubuntu2
libc6 2.27-3ubuntu1
libcuinj64-9.1 9.1.85-3
libgcc1 1:8-20180414-
InstallationDate: Installed on 2018-04-21 (5 days ago)
InstallationMedia: Ubuntu 18.04 LTS "Bionic Beaver" - Alpha amd64 (20180421)
ProcEnviron:
TERM=xterm-
PATH=(custom, no user)
XDG_RUNTIME_
LANG=en_US.UTF-8
SHELL=/bin/bash
SourcePackage: nvidia-cuda-toolkit
UpgradeStatus: No upgrade log present (probably fresh install)
Changed in nvidia-cuda-toolkit (Ubuntu): | |
status: | New → Incomplete |
Martin, thanks for the detailed bug report.
I have tried on two different machines and on both, nvprof works without requiring sudo access.
However, neither of these are fresh installations of 18.04.
After running nvprof with sudo, please try running it again without.
Also, would you be able to test on a different machine?