``GMX_CUDA_NB_TAB_EWALD``
force the use of tabulated Ewald kernels. Should be used only for benchmarking.
-``GMX_DISABLE_CUDALAUNCH``
- disable the use of the lower-latency cudaLaunchKernel API even when supported (CUDA >=v7.0).
- Should only be used for benchmarking purposes.
-
``GMX_DISABLE_CUDA_TIMING``
Deprecated. Use ``GMX_DISABLE_GPU_TIMING`` instead.