1

我想知道什么

does. Is it just freeing up the memory of GPU and do nothing?
4

2 回答 2

1

从nvidia-smi帮助菜单 ( man nvidia-smi)：

-r, --gpu-reset
       Trigger a reset of one or more GPUs.  Can be used to clear GPU HW and SW state in situations that would otherwise require a machine reboot.  Typically useful if a double bit ECC
       error has occurred.  Optional -i switch can be used to target one or more specific devices.  Without this option, all GPUs are reset.  Requires root.  There can't be any  appli‐
       cations  using these devices (e.g. CUDA application, graphics application like X server, monitoring application like other instance of nvidia-smi).  There also can't be any com‐
       pute applications running on any other GPU in the system.

       Starting with the NVIDIA Ampere architecture, GPUs with NVLink connections can be individually reset.  On NVSwitch systems, Fabric Manager is required to facilitate reset.

       If Fabric Manager is not running, or if any of the GPUs being reset are based on an architecture preceding the NVIDIA Ampere architecture, any GPUs with NVLink connections to  a
       GPU  being  reset  must also be reset in the same command.  This can be done either by omitting the -i switch, or using the -i switch to specify the GPUs to be reset.  If the -i
       option does not specify a complete set of NVLink GPUs to reset, this command will issue an error identifying the additional GPUs that must be included in the reset command.

       GPU reset is not guaranteed to work in all cases. It is not recommended for production environments at this time.  In some situations there may be HW  components  on  the  board
       that  fail to revert back to an initial state following the reset request.  This is more likely to be seen on Fermi-generation products vs. Kepler, and more likely to be seen if
       the reset is being performed on a hung GPU.

       Following a reset, it is recommended that the health of each reset GPU be verified before further use.  If any GPU is not healthy a complete reset should be instigated by  power
       cycling the node.

       GPU reset operation will not be supported on MIG enabled vGPU guests.

       Visit http://developer.nvidia.com/gpu-deployment-kit to download the GDK.
于 2021-11-16T17:59:14.580 回答
1

重置 GPU 状态。可用于清除双位 ECC 错误或恢复挂起的 GPU。需要 -i 切换到目标特定设备。仅在 Linux 上可用。 https://developer.download.nvidia.com/compute/DCGM/docs/nvidia-smi-367.38.pdf

于 2021-11-16T09:06:52.750 回答