I am writing to request assistance with an issue I am experiencing while attempting to enable GPU acceleration for VASP 6.4.2. During runtime, the application fails to utilize GPU acceleration and returns an error that indicates an unsupported compute capability version.
System and Software Details:
VASP Version: 6.4.2
GPU Model: NVIDIA RTX 5090
GPU compute capability:
“$ nvidia-smi --query-gpu=compute_cap --format=csv
compute_cap
12.0”
nvfortran:
“$ nvfortran --help -gpu
…supported values: 35 50 60 61 70 75 80 86 89 90…”
Driver Version: 572.47
Operating System: Ubuntu 24.04 LTS running under WSL2
NV HPC SDK Version: 25.1. with CUDA 12.6
Build Environment: NVFORTRAN and NVC++ provided by the NVIDIA HPC SDK
Error Description:
When running VASP with GPU acceleration enabled, the application fails to execute the GPU-dependent routines. The error message specifically suggests that the compute capability version of my GPU (cc120 suggested by VASP error message) exceeds the maximum supported version by nvfortran (cc89). I have verified that the GPU is supported for CUDA applications through “nvidia-smi and nvcc”, but this mismatch prevents VASP from leveraging GPU acceleration.
However, the compilation of VASP 6.4.2 is successful with “makefile.include” setting:
FC = mpif90 -acc -gpu=cc89,cuda11.0
FCL = mpif90 -acc -gpu=cc89,cuda11.0 -c++libs
VASP error message:
“Accelerator Fatal Error: This file was compiled: -acc=gpu -gpu=cc80 -gpu=cc89 -acc=host or -acc=multicore
Rebuild this file with -gpu=cc120 to use NVIDIA Tesla GPU 0
File: /home/zhyin/vasp.6.4.2/build/std/./fft3dsimple.F
Function: fftwav:221
Line: 243”
Steps I Have Taken:
Verified that the latest compatible NVIDIA drivers are installed (25.1).
Ensured that the CUDA toolkit and NVIDIA HPC SDK were properly configured ($nvhpc exported to PATH, nvfortran, and nvcc can be found).
Compiled VASP with GPU acceleration flags enabled with “cc89” and recompiled VASP with CPU only with nvhpc toolkit (CPU version VASP runs fine).
Could you please assist me in diagnosing this issue?
Thank you in advance for your time and assistance. I am happy to provide any additional logs or configuration details if needed.