# dcgm-exporter counters, without the PROF_* fields: DCGM's profiling module serves one client per GPU, so where # another exporter on the machine already holds it, this one would exit with "The third-party Profiling module returned # an unrecoverable error" (point Prometheus at that exporter instead: DCGM_TARGET). # Format: DCGM FIELD, Prometheus metric type, help DCGM_FI_DEV_SM_CLOCK, gauge, SM clock (MHz) DCGM_FI_DEV_MEM_CLOCK, gauge, memory clock (MHz) DCGM_FI_DEV_POWER_USAGE, gauge, power draw (W) DCGM_FI_DEV_GPU_TEMP, gauge, GPU temperature (C) DCGM_FI_DEV_GPU_UTIL, gauge, legacy utilisation = kernel residency (%) DCGM_FI_DEV_MEM_COPY_UTIL, gauge, memory controller busy (%) DCGM_FI_DEV_FB_USED, gauge, framebuffer used (MiB) DCGM_FI_DEV_CLOCK_THROTTLE_REASONS, gauge, clock throttle reason bitmask DCGM_FI_DEV_PCIE_TX_THROUGHPUT, gauge, PCIe TX (KB/s) DCGM_FI_DEV_PCIE_RX_THROUGHPUT, gauge, PCIe RX (KB/s)