Windows 10 error with Nsight: ==WARNING== No kernels were profiled

Developer Tools Nsight Compute
1

Hello,

I cant use Nsight profiler in windows 10.
I am trying to profile a binary executable but when I do so I got these errors:
C:\Windows\system32>ncu -o profilerOuput --target-processes all D:\CUDA\a.exe
==PROF== Connected to process 2344 (D:\CUDA\a.exe)
Execution Configuration:
N = 4, Number of Blocks = 1, Number of Threads Per Block = 1

Matrix
0.0225126 5.64585 1.94304 8.09741
5.86009 4.80873 3.51291 8.96962
8.2384 7.47605 1.75108 8.59943
7.11501 5.14535 3.04995 0.159846

Vector :
0.924029 3.65452 1.48313 1.66899

Kernel Result :
0 0 0 0

cuBLAS Result :
5.0888e+276 5.83022e+252 1.65376e+243 1.29062e-306

“ERRORS I AM GETTING”
==PROF== Disconnected from process 2344
==WARNING== No kernels were profiled.

With NVPROF apparently, the problem does not happen:
C:\Windows\system32>nvprof D:\CUDA\a.exe
==19608== NVPROF is profiling process 19608, command: D:\CUDA\a.exe
Execution Configuration:
N = 4, Number of Blocks = 1, Number of Threads Per Block = 1

Matrix
0.0225126 5.64585 1.94304 8.09741
5.86009 4.80873 3.51291 8.96962
8.2384 7.47605 1.75108 8.59943
7.11501 5.14535 3.04995 0.159846

Vector :
0.924029 3.65452 1.48313 1.66899

Kernel Result :
0 0 0 0

cuBLAS Result :
1.41087e-311 1.41087e-311 1.41087e-311 1.41087e-311

==19608== Profiling application: D:\CUDA\a.exe
==19608== Profiling result:
Type Time(%) Time Calls Avg Min Max Name
GPU activities: 51.43% 1.1520us 2 576ns 192ns 960ns [CUDA memcpy HtoD]
48.57% 1.0880us 1 1.0880us 1.0880us 1.0880us [CUDA memcpy DtoH]
API calls: 85.79% 86.283ms 4 21.571ms 3.0000us 86.271ms cudaMalloc
13.54% 13.619ms 1 13.619ms 13.619ms 13.619ms cuDevicePrimaryCtxRelease
0.32% 320.20us 3 106.73us 36.700us 202.90us cudaMemcpy
0.19% 195.40us 4 48.850us 2.7000us 164.60us cudaFree
0.11% 113.60us 1 113.60us 113.60us 113.60us cuLibraryLoadData
0.02% 23.600us 114 207ns 100ns 1.6000us cuDeviceGetAttribute
0.01% 5.7000us 3 1.9000us 300ns 5.0000us cuDeviceGetCount
0.01% 5.3000us 1 5.3000us 5.3000us 5.3000us cuLibraryUnload
0.00% 2.7000us 1 2.7000us 2.7000us 2.7000us cuModuleGetLoadingMode
0.00% 1.9000us 2 950ns 200ns 1.7000us cuDeviceGet
0.00% 1.5000us 1 1.5000us 1.5000us 1.5000us cuDeviceTotalMem
0.00% 800ns 1 800ns 800ns 800ns cuDeviceGetName
0.00% 600ns 1 600ns 600ns 600ns cuDeviceGetLuid
0.00% 600ns 1 600ns 600ns 600ns cudaLaunchKernel
0.00% 300ns 1 300ns 300ns 300ns cuDeviceGetUuid

Does anyone know how to make it run with Nsight?
I need it running asap.

2

You will need to state which Nsight Compute version, CUDA driver version and GPU you are using. Please note that only Volta-architecture GPUs or newer are supported by Nsight Compute.

3

Nsight Compute version: 2022.4.1.0

CUDA driver version:
C:\Windows\system32>nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2023 NVIDIA Corporation
Built on Fri_Jan__6_19:04:39_Pacific_Standard_Time_2023
Cuda compilation tools, release 12.0, V12.0.140
Build cuda_12.0.r12.0/compiler.32267302_0

GPU: NVIDIA GeForce 940MX
C:\Windows\system32>nvidia-smi
Wed Feb 22 01:58:14 2023
±----------------------------------------------------------------------------+
| NVIDIA-SMI 528.33 Driver Version: 528.33 CUDA Version: 12.0

This is the information, is there anyway to run Nsight compute on my platform?
Apparently, my architecture is Maxwell, so is there any other Nsight version I can use for my architecture?

4

Thanks for the information. This card is a Maxwell GPU, which is not supported by Nsight Compute. The recommended tool for profiling on Maxwell remains nvprof.