CUDA PTX JIT Compilation Failure in StreaMD on NVIDIA RTX 5070

[dsnbkprasanth]

streamd_bash_4EY7_M_BZC6__30-08-2026-18-16-23.log

Dear StreaMD developers,

I am encountering a GPU execution error while running StreaMD molecular dynamics simulations and would appreciate your guidance.

The workflow completes ligand preparation and energy minimization successfully. The failure occurs when starting the NVT equilibration step.

The generated command is:

gmx mdrun -deffnm nvt -s nvt.tpr -nt 20 -nb gpu -update gpu -pme gpu -bonded gpu -pmefft gpu -ntmpi 1 -ntomp 20

The error is:

CUDA error #218 (cudaErrorInvalidPtx): a PTX JIT compilation failed.

Specifically:

GPU kernel (k_pruneonly) failed to launch: CUDA error #218 (cudaErrorInvalidPtx): a PTX JIT compilation failed.

My hardware and software configuration is:

• GPU: NVIDIA GeForce RTX 5070
• NVIDIA driver: 595.84
• CUDA version reported by NVIDIA-SMI: 13.2
• Operating system: Linux
• Python: 3.11
• StreaMD: installed from the GitHub repository
• GROMACS used by the StreaMD-generated workflow: 2025.2-conda_forge
• Conda CUDA toolkit: 12.6

The simulation log shows that StreaMD is using the Conda GROMACS executable:

/home/balu/miniconda3/envs/md_tpr138/bin.AVX2_256/gmx

Energy minimization completes successfully. The failure occurs immediately after the NVT run starts.

I also have a system installation of GROMACS 2026.1 compiled with CUDA 13.1, but the StreaMD-generated workflow still appears to use the Conda GROMACS 2025.2 executable.

Could you please advise:

1. Whether the current GPU environment is compatible with an NVIDIA RTX 5070.
2. Whether there is a recommended CUDA/GROMACS version for RTX 50-series GPUs.
3. Whether StreaMD supports using a system-installed GROMACS instead of the Conda GROMACS executable.
4. Whether the GPU arguments generated by StreaMD can be modified safely for this hardware.
5. Whether there is an updated GPU environment YAML that you recommend for this configuration.

I have attached the complete StreaMD log for reference.

Thank you for your help.

Best regards,
Prasanth

[avnikonenko]

Hello,
Have you installed env_gpy.yml version (wget -O env_gpu.yml https://raw.githubusercontent.com/ci-lab-cz/streamd/refs/heads/master/env_gpu.yml)?
StreaMD doesn't support using a system-installed GROMACS as it could create compatibility problems between other packages used for the analysis.
As for GPU usage I would recommend to search information from Gromacs documentation itself https://manual.gromacs.org/documentation/current/user-guide/mdrun-performance.html.

Best,
Aleksandra