CUDA error in 10ns run

User discussions
1

GROMACS version: 2024.1
GROMACS modification: No

2 NVIDIA GeForce RTX 2070
Driver Version: 545.23.08 CUDA Version: 12.3
AMD Ryzen 9 3900x 12-core processor x24
UBUNTU 22.04.3 LTS

The installation was made with GPU support and MPI default on. The simulations randomly end before completing the task. The task was divided in 4 MPI as pp0, pp0 , pp1, pp1. Could this be a hardware issue or could it be a code one?

Assertion failed:
Condition: stat == cudaSuccess
Cuda error #700 an illegal memory access

Assertion failed:
Condition: stat == cudaSuccess
cudaStreamSynchronize failed. CUDA error #719 (cudaErrorLaunchFailure):
unspecified launch failure.
Source file: src/gromacs/gpu_utils/device_stream.cu (line 107)
Function: DeviceStream::synchronize() const::<lambda()>

2

what you are trying to do? is it pulling simulation?

3

Yes indeed, it is a pulling simulation. The first sample of the umbrella sampling.

4

Did you manage to solve this? I’m encountering the same issue