Hi,
System:
-
GROMACS 2024.2
-
CUDA toolkit 12.8
-
NVIDIA driver 581.66
-
GPU: NVIDIA GeForce RTX 5050 Laptop GPU
-
WSL Ubuntu 24.04
-
GCC 12.4
I compiled GROMACS with:
CC=/usr/bin/gcc-12 CXX=/usr/bin/g+±12 cmake ..
-DGMX_BUILD_OWN_FFTW=ON
-DGMX_MPI=ON
-DGMX_GPU=CUDA
-DGMX_CUDA_TARGET_SM=89
-DCMAKE_CUDA_HOST_COMPILER=/usr/bin/g+±12
-DCMAKE_CUDA_FLAGS=“–allow-unsupported-compiler --generate-code arch=compute_89,code=compute_89”
-DCMAKE_INSTALL_PREFIX=/usr/local/gromacs
Compilation succeeds, and gmx_mpi --version reports CUDA support.
However, running:
gmx_mpi mdrun -s production.tpr -deffnm production -nb gpu
fails with:
WARNING: An error occurred while sanity checking device #0.
CUDA error #209 (cudaErrorNoKernelImageForDevice): no kernel image is available for execution on the device.
Then GROMACS aborts with:
“Cannot run short-ranged nonbonded interactions on a GPU because no GPU is detected.”
I also attempted:
-DGMX_CUDA_TARGET_SM=120
with CUDA 12.8, but compilation fails during ptxas with:
“Value of threads per SM … is out of range”
Is RTX 5050 / Blackwell support expected in a newer GROMACS release, or is there a recommended workaround?