|
|
最近要公布的一个gmx自动运行小脚本，其中对并行参数定义的解释，参考：
- md_nt='32' # Number of threads used when running mdrun (excluding 'em' and 'rerun'). It cannot be greater than the number of CPU logical cores of the current host.
- multi_gpu='0' # Whether to use multiple GPUs to run 1 mdrun task (excluding 'em' and 'rerun'). If enabled, the 'md_nt' option will be ignored.
- ntmpi='8' # Number of MPI threads (RANKs) used when running 1 mdrun task with multiple GPUs.
- ntomp='6' # Number of OpenMP threads used when running 1 mdrun task with multiple GPUs. The value of [ntmpi*ntomp] cannot be greater than the number of CPU logical cores of the current host. It is generally not recommended to use the extra logical cores obtained by Hyper Threading (HT) technology.
- gputasks='' # MPI RANKs of each GPU. When this option is enabled at the same time as 'gpu_id', this option will be ignored. Note: the last RANK is PME RANK and the rest are PP RANKs. This option can be used for manual control of GPU load balancing, for example: ntmpi=8, gputasks=00000111, means The first 5 RANKs run with GPU 0, and the last 3 RANKs run with GPU 1.
- gpu_id='0' # ID of GPU which used in MD simulation tasks。If more than 1 is used, write out the GPU IDs side by side. For example, '01'. Note: if the current host has multiple GPUs with different architectures, the GPU ID detected by GMX may not be consistent with the GPU ID in the GPU driver, instead, it is sorted by the value of "compute_cap" from large to small according to the "fastest first" rule.
- update_gpu='1' # Whether to add the '-update gpu' option in mdrun (excluding 'em' and 'rerun') command when running 1 mdrun task with 1 GPU. This option is not supported if a 4-point water model is used.
- pinoffset='0' # Value of '-pinoffset' option in mdrun (including 'em' and 'rerun') command, default value is 0. When you need to run 2 or more GMX tasks in parallel in the current host, you can define other values by yourself. For example, in a host with dual 24c&48t CPUs and dual GPUs, for task 1: md_nt=24, multi_gpu=0, gpu_id=0, pinoffset=0; for task 2: md_nt=24, multi_gpu=0, gpu_id=1, pinoffset=48.
- pinstride='0' # Value of '-pinstride' option in mdrun (including 'em' and 'rerun') command, default value is 0, GMX will determine the appropriate value by itself. If there are no special requirements, it is strongly recommended not to change this option to other values!
|
|
[复制链接 Copy URL]
发表于 Post on 2022-3-5 00:00:08
