Reduce number of iterations for ncu profiling

Author: OliverRietmann

.github/workflows/standalone-benchmark.yml

@@ -97,17 +97,16 @@ jobs:
           source /etc/profile.d/modules.sh
           module load ninja/fortran-v1.11.1.g9-15 Vc/1.4.5-10 boost/v1.83.0-alice2-57 fmt/11.1.2-14 CMake/v3.31.6-10 ms_gsl/4.2.1-3 Clang/v20.1.7-9 TBB/v2022.3.0-3 ROOT/v6-36-04-alice9-15 ONNXRuntime/v1.22.0-71 GLFW/3.3.2-25
           cd ${STANDALONE_DIR}
-          ncu --set none --metrics gpu__time_duration.avg --export ${{ matrix.name }} --clock-control none --force-overwrite ${TIMING_CA} --runs 42 --debug 1 --PROCdebugMarkdown 1  # Generates ${{ matrix.name }}.ncu-rep
+          ncu --set none --metrics gpu__time_duration.avg --export ${{ matrix.name }} --clock-control none --force-overwrite ${TIMING_CA} --runs 21 --debug 1 --PROCdebugMarkdown 1  # Generates ${{ matrix.name }}.ncu-rep
           ncu --import ${STANDALONE_DIR}/${{ matrix.name }}.ncu-rep --print-units base --csv > /root/${PROFILER_CSV}
           rm -rf ${STANDALONE_DIR}/events/50kHz ${STANDALONE_DIR}/build
-          python3 ${GITHUB_WORKSPACE}/.github/scripts/profiler_ncu.py --runs 42 --input /root/${PROFILER_CSV} --output /root/${PROFILER_CSV}
+          python3 ${GITHUB_WORKSPACE}/.github/scripts/profiler_ncu.py --runs 21 --input /root/${PROFILER_CSV} --output /root/${PROFILER_CSV}
 
       - name: Profiler - Nsight Systems
         if: ${{ matrix.name == 'nvidia-l40s' }}
         run: |
-          curl -fL --retry 3 -o ${STANDALONE_DIR}/nsys.rpm https://developer.nvidia.com/downloads/assets/tools/secure/nsight-systems/2026_2/NsightSystems-linux-cli-public-2026.2.1.210-3763964.rpm
-          dnf install -y ${STANDALONE_DIR}/nsys.rpm
-          rm -f ${STANDALONE_DIR}/nsys.rpm
+          dnf config-manager --add-repo "https://developer.download.nvidia.com/devtools/repos/rhel$(source /etc/os-release; echo ${VERSION_ID%%.*})/$(rpm --eval '%{_arch}' | sed s/aarch/arm/)/"
+          dnf install -y nsight-systems-cli-2026.2.1
           source /etc/profile.d/modules.sh
           module load ninja/fortran-v1.11.1.g9-15 Vc/1.4.5-10 boost/v1.83.0-alice2-57 fmt/11.1.2-14 CMake/v3.31.6-10 ms_gsl/4.2.1-3 Clang/v20.1.7-9 TBB/v2022.3.0-3 ROOT/v6-36-04-alice9-15 ONNXRuntime/v1.22.0-71 GLFW/3.3.2-25
           cd ${STANDALONE_DIR}