CUDA

• Use -DCP2K_USE_ACCEL=CUDA to generally enable support for NVIDIA GPUs

• Specify the CUDA compute capability with -DCMAKE_CUDA_ARCHITECTURES, for example 100 for B200. The deprecated -DCP2K_WITH_GPU selector remains available with the values K20X, K40, K80, P100, V100, A100, H100, B200, GB10, and A40.

• Use -DCP2K_USE_NVHPC=ON when building with the NVHPC kit.

• Use -DCP2K_WITH_GPU_PROFILING to turn on NVIDIA Tools Extensions. It requires to link -lnvToolsExt.

• Link to a BLAS/ScaLAPACK library that accelerates large DGEMMs (e.g., libsci_acc)

• Use -DCP2K_ENABLE_GRID_GPU=OFF to disable the GPU backend of the grid library.

• Use -DCP2K_ENABLE_DBM_GPU=OFF to disable the GPU backend of the sparse tensor library.

• Use -DCP2K_ENABLE_PW_GPU=OFF to disable the GPU backend of FFTs and associated gather/scatter operations.

• Use -DCP2K_ENABLE_LIBXC_GPU=OFF to build without the LibXC device/GPU path (by default the device path is compiled in when building with -DCP2K_USE_ACCEL=CUDA and a CUDA-capable libxc).

• At runtime the LibXC backend is selected with &XC/LIBXC_GPU_BACKEND = DEVICE or HOST. The default is HOST and the GPU backend is only used when DEVICE is explicitly requested. Requesting DEVICE is an error (abort) when the build has no device path or when no usable GPU is found at runtime; HOST always runs and touches no GPU.

• Use -DCP2K_DBCSR_USE_CPU_ONLY=ON to disable the GPU backend of DBCSR.