To achieve the highest performance of GPU resources, you can set the GPU clock based on your business requirements, control GPU power consumption and heat generation, and improve GPU stability.

Prerequisites

GPU-type nodes exist in clusters, such as GPU Accelerated nodes and High Performance Computing Optimized nodes. For details, see Creating node pools.

Procedure

(Optional) View information about the GPU clock.

nvidia-smi -q -d CLOCK

Set the GPU to the maximum clock frequency, that is, lock the clock frequency, to achieve the highest performance of the GPU.

nvidia-smi -pm 1
MCLOCKS=`nvidia-smi -i 0 --query-gpu=clocks.max.mem --format=csv,noheader,nounits`
GCLOCKS=`nvidia-smi -i 0 --query-gpu=clocks.max.gr --format=csv,noheader,nounits`
nvidia-smi -ac $MCLOCKS,$GCLOCKS
nvidia-smi -lgc $GCLOCKS

Validating results

Run the following command to view the GPU clock frequency.

nvidia-smi -q -d CLOCK

If the returned information shows that the clock frequency is set to the maximum value, GPU clock locking is configured.

You can also view the GPU clock rate on the cluster Prometheus monitoring dashboard. For detailed instructions, see AI resource observability.

Last updated: 2026.07.24 14:51:38

Public security network record in Beijing: No. 11010802032137