pbs_(gpureset)

🌸 NAME 🌸

okay so the vibe is pbs_ gpureset — and its whole job is literally resetting GPU error counts 💅

✨ SYNOPSIS ✨

#include <pbs_error.h>
#include <pbs_ifl.h>

int pbs_ gpureset(int connect, char *mom_node, int gpu_id, int ecc_perm, int ecc_vol)

💅 DESCRIPTION 💅

Bestie, this function issues a batch request for pbs_mom to reset the ECC counts on one of its Nvidia GPUs. The GPU's error count gets reset by sending a GPU Control batch request to the batch server. It's giving "middle management but make it GPU" ✨

Here's the cast of arguments, no cap:

• mom_node — this specifies the host within the cluster on which the GPU is located. The argument is the name of a host that is a member of the cluster of hosts managed by the server 🖥️
• gpu_id — this specifies the ID of the GPU on the MOM node 👑
• ecc_perm — this specifies whether or not to reset the GPU's permanent ECC error count. Value of 1 resets, value of 0 does not 🔥
• ecc_vol — this specifies whether or not to reset the GPU's volatile ECC error count. Value of 1 resets, value of 0 does not 🌈

🫶 heads up, bestie: this call requires PBS Operator or Manager privilege. It also requires that Torque be configured with --enable-nvidia-gpu.

🔥 SEE ALSO 🔥

• qgpureset(1B) — go peek at this one too, iykyk 💖

🎀 DIAGNOSTICS 🎀

When the batch request generated by the pbs_ gpureset() function has been completed successfully by a batch server, the routine will return 0 (zero). Otherwise, a non zero error is returned. The error number is also set in pbs_errno. So yeah, 0 is the slay outcome, anything else is the plot twist 😭

xoxo, this was Local manual section 3B, page pbs_(gpureset) — mwah 💋✨