For the complete documentation index, see llms.txt. This page is also available as Markdown.

GPU

This page describes the Optimization Pack for the component type GPU.

Metrics

Metric
Unit
Description

gpu_util

percent

GPU utilization

gpu_sm_active

percent

GPU streaming multiprocessor utilization

gpu_sm_occupancy

percent

GPU streaming multiprocessor occupancy

gpu_tensor_core_active

percent

GPU tensor core utilization

gpu_dram_active

percent

GPU DRAM utilization

gpu_fb_used

megabytes

GPU framebuffer memory used

gpu_fb_total

megabytes

GPU framebuffer memory total

gpu_power_usage

watt

GPU power usage

gpu_power_limit

watt

GPU power limit

gpu_temp

celsius

GPU temperature

gpu_memory_temp

celsius

GPU memory temperature

gpu_power_violation_rate

violations per second

GPU power violation rate

gpu_thermal_violation_rate

violations per second

GPU thermal violation rate

gpu_pcie_replay_rate

replays per second

GPU PCIe replay rate

gpu_fp16_pipe_active

percent

GPU FP16 (non-tensor) pipe activity

gpu_fp32_pipe_active

percent

GPU FP32 (non-tensor) pipe activity

Parameters

The GPU component type exposes metrics only (for observation and for defining constraints or optimization goals); it does not provide tunable parameters.

Last updated

Was this helpful?