Gromacs run only on 1 processor without using all threads

GROMACS version: 2023.3
GROMACS modification: No
Hello Gromacs users,

I built Gromacs 2023.3 on Ubuntu 24.04 with pre-installed libomp-dev using such code

cmake .. -DGMX_GPU=CUDA -DGMX_PREFER_STATIC_LIBS=OFF -DGMXAPI=OFF -DGMX_BUILD_OWN_FFTW=ON -DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda -DCMAKE_INSTALL_PREFIX=/home/xxx/progs/gromacs_2023_3 -DGMX_DEFAULT_SUFFIX=OFF -DGMX_BINARY_SUFFIX=_2023_3 -DGMX_LIBS_SUFFIX=_2023_3 make -j 20 make install
The installation of Gromacs did not cause any problems. However, when I run gromacs:

gmx_2023_3 mdrun -v -deffnm press600 -bonded gpu -nb gpu -pme gpu -s topol_600.tpr -cpi press600.cpt

In the log file, I see that

PME tasks will do all aspects on the GPU
Using 1 MPI thread
Using 32 OpenMP threads

Nevertheless, the nvtop or glances only show that 1 processor run on maximum 100% performance.
Could you please suggest a solution to this problem.

can you share your system information and log file. You have offloaded all the calculations to the GPU, though

When everything runs on the GPU, only one thread will be active most of the time on the CPU, so most cores will be idle.