Help: RTX 5080 (Blackwell, Compute Capability 12.1) + WSL2 + CUDA 13.0 “No Kernel Image Available”

User discussions
,
1

Help: RTX 5080 (Blackwell, Compute Capability 12.1) + WSL2 + CUDA 13.0 “No Kernel Image Available” Error in GROMACS 2025.3

Post Body:

Hi GROMACS developers and users,

I am trying to run MD simulations with GROMACS 2025.3 on a new workstation:

  • GPU: NVIDIA RTX 5080 (Blackwell, Compute Capability 12.1)

  • OS: Windows 11, running Ubuntu under WSL2

  • CUDA: 13.0 toolkit and latest Windows/WSL2 NVIDIA drivers

  • GROMACS: built from source targeting -DGMX_CUDA_TARGET_SM=121

  • Simulations, gmx --version, and nvcc all confirm correct CUDA/driver setup.

However, at runtime, I consistently get:

text

CUDA error #209 (cudaErrorNoKernelImageForDevice): no kernel image is available for execution on the device. Fatal error: Cannot run short-ranged nonbonded interactions on a GPU because no GPU is detected.

Even though my GPU and driver are listed, GROMACS cannot offload to GPU.

Questions:

  • Is current support for Blackwell (sm_121, Compute Capability 12.1) officially production-ready in GROMACS 2025.3 for WSL2/Windows?

  • Are there known solutions, workarounds, or patches for this kernel image error on RTX 5080 under WSL2?

  • Should I build with special flags or change runtime options?

  • Would switching to native Linux be necessary for immediate full GPU performance?

Any advice or direct experience would be very helpful. Thank you!

2

Hi.

I believe RTX 50xx-series is CC 12.0, not CC 12.1:CUDA GPU Compute Capability | NVIDIA Developer

Have you tried using -DGMX_CUDA_TARGET_SM=120?

3

Thanks alot. It is working fine now.