Denis

No no this was just a tst with the serial option, i want to use HIP and to run on AMD GPU

Strange this is not even configuring properly now: ``` [dbertini@lxbk1131 /lustre/rz/dbertini/gpu/picongpu_v0.7.0/sim/lwfa]$ pic-build -b hip:gfx908 -c '-DBOOST_ROOT=/usr -DMPI_CXX_WORKS=ON -DMPI_CXX_VERSION=0 -DCMAKE_CXX_FLAGS=-stdlib=libc++' build directory: .build cmake command: cmake -DCMAKE_INSTALL_PREFIX=/lustre/rz/dbertini/gpu/picongpu_v0.7.0/sim/lwfa -DPIC_EXTENSION_PATH=/lustre/rz/dbertini/gpu/picongpu_v0.7.0/sim/lwfa -DBOOST_ROOT=/usr -DMPI_CXX_WORKS=ON -DMPI_CXX_VERSION=0...

- **Container setup:** ``` [dbertini@lxbk1131 ~]$ hipcc --version HIP version: 5.7.31921-1949b1621 AMD clang version 17.0.0 (https://github.com/RadeonOpenCompute/llvm-project roc-5.7.1 23382 f3e174a1d286158c06e4cc8276366b1d4bc0c914) Target: x86_64-unknown-linux-gnu Thread model: posix InstalledDir: /opt/rocm-5.7.1/llvm/bin [dbertini@lxbk1131 ~]$ gcc --version...

nothing new on that unfortunately

@DrDaveD the point in changing `FATAL` to `INFO` is that when trying to create again an already existing instance, apptainer of course complains that the instance exists already and goes...

you mean to lock the `list` and `start` together in the same scope?

I also do not understand this effect, but it is real. SLURM jobs shows COMPLETED and finished properly. We first thought that it is linked to the `slurmd` daemon which...

we also checked for hanging processes after slurm job completion, without success.

could one think of a problem linked to kernel (4.18) namespace counter ?

OK thanks for the infos !