求助：CUDA error cudaStreamSynchronize: an illegal memory access was encountered

yingi

运行NAMD-cuda加速版本的时候出现：FATAL ERROR: CUDA error cudaStreamSynchronize(stream) in file src/CudaTileListKernel.cu, function buildTileLists, line 1105
on Pe 8 (device 0 pci 0:3b:0): an illegal memory access was encountered
FATAL ERROR: CUDA error cudaStreamSynchronize(stream) in file src/CudaTileListKernel.cu, function buildTileLists, line 1105
on Pe 16 (device 1 pci 0:af:0): an illegal memory access was encountered
使用的是两张NVIDIA TITAN V,跑的时候GPU使用率都很低，一张36%，一张28%。运行环境是SEG集群上一个单节点。
提前感谢大家：）



镇楼图：

yingi

然后我刚刚使用了一张显卡（+devices 0）也不行：
FATAL ERROR: CUDA error cudaStreamSynchronize(stream) in file src/CudaTileListKernel.cu, function buildTileLists, line 1105
on Pe 16 (device 0 pci 0:3b:0): an illegal memory access was encountered

eniac

yingi 发表于 2022-6-11 15:39
然后我刚刚使用了一张显卡（+devices 0）也不行：
FATAL ERROR: CUDA error cudaStreamSynchronize(stream ...

请问您是怎么解决的？我也遇到了相同的问题
——GROMACS 2024，显卡是4090，运行mdrun报错Unexpected cudaStreamQuery failure. CUDA error #700 (cudaErrorIllegalAddress): an illegal memory access was encountered.