5060 with GROMACS

User discussions
1

GROMACS version:2025.3-conda_forge
GROMACS modification: Yes/No
Here post your question

Hello GROMACS community,

I’m encountering a CUDA error when trying to run GROMACS 2025.3 on an NVIDIA Blackwell GPU (compute capability 12.0). The simulation fails during kernel execution with a PTX compilation error.

System Configuration:

  • GPU: NVIDIA Blackwell (compute capability 12.0)

  • GROMACS: 2025.3-conda_forge (installed via conda)

  • CUDA runtime: 12.60, CUDA driver: 13.0

  • Build: nompi_cuda_h7ac747b_0

  • Installation: conda install conda-forge::gromacs=2025.3=nompi_cuda_h7ac747b_0

Error Details:

text

Assertion failed:
Condition: stat == cudaSuccess
GPU kernel (k_pruneonly) failed to launch: CUDA error #218
(cudaErrorInvalidPtx): a PTX JIT compilation failed.

Command Used:

bash

gmx mdrun -v -deffnm md -gpu_id 0 -nb gpu -pme gpu -pin on

What I’ve Verified:

✅ GROMACS detects GPU and CUDA support correctly
✅ .tpr file loads without issues
✅ GPU task mapping shows: PP:0, PME:0
✅ CUDA versions appear compatible (runtime 12.60, driver 13.0)

Troubleshooting Attempts:

  1. -nb gpu -pme cpu → Same error in k_pruneonly kernel

  2. Various GPU flag combinations → Persistent PTX error

  3. Reinstalled different conda-forge CUDA builds

  4. The GPU works fine with other CUDA applications

Specific Questions:

  1. Does the conda-forge build include full support for compute capability 12.0?

  2. Are there specific flags or settings for Blackwell architecture?

  3. Would you recommend manual compilation with specific CC 12.0 flags?

  4. Is this a known issue with PTX kernels on new architectures?

  5. Are there workarounds for immediate use while waiting for full support?

Additional Information:

Full version output:

text

[Include the complete output of 'gmx mdrun -version' here]

Partial log showing GPU detection:

text

1 GPU selected for this run.
Mapping of GPU IDs to the 2 GPU tasks in the 1 rank on this node: PP:0,PME:0
PP tasks will do (non-perturbed) short-ranged interactions on the GPU

The system runs correctly in CPU-only mode. Any guidance on resolving this PTX compatibility issue would be greatly appreciated.

Thank you for your help!

2

Please don’t open duplicate topics.

Discussion continued in CUDA error #218