CPU Monitoring#
Overview#
DCGM supports CPU monitoring for NVIDIA datacenter CPUs. In general, CPU fields are treated the same way as other NVIDIA hardware entities. This section includes additional information to help you get started.
Requirements#
The features described here are only supported for NVIDIA Grace CPUs, and support for these features begins with DCGM 3.3.
Introduction#
The two entity groups for CPUs in DCGM are DCGM_FE_CPU and DCGM_FE_CPU_CORE, corresponding to the CPUs and cores in Grace. Some fields are only measured at the level of CPUs, while others have measurements for each core.
CPU And Core Fields#
Field Name |
Description |
Field ID |
Measured at |
|---|---|---|---|
DCGM_FI_DEV_CPU_UTIL_TOTAL |
The total percentage of time in use |
1100 |
Core |
DCGM_FI_DEV_CPU_UTIL_USER |
The percentage of time in use by the user |
1101 |
Core |
DCGM_FI_DEV_CPU_UTIL_NICE |
The percentage of time in use by low priority (high nice score) programs |
1102 |
Core |
DCGM_FI_DEV_CPU_UTIL_SYS |
The percentage of time in use by the system |
1103 |
Core |
DCGM_FI_DEV_CPU_UTIL_IRQ |
The percentage of time in use by interrupts |
1104 |
Core |
DCGM_FI_DEV_CPU_TEMP_CURRENT |
Instantaneous temperature (Celsius) |
1110 |
CPU |
DCGM_FI_DEV_CPU_CLOCK_CURRENT |
Instantaneous clock speed (KHz) |
1120 |
Core |
DCGM_FI_DEV_CPU_POWER_UTIL_CURRENT |
Instantaneous power usage (watts) |
1130 |
CPU |
DCGM_FI_DEV_CPU_POWER_LIMIT |
Instantaneous power limit (watts) |
1131 |
CPU |
DCGM_FI_DEV_CPU_VENDOR |
The name of the vendor |
1140 |
CPU |
DCGM_FI_DEV_CPU_MODEL |
The name of the model |
1141 |
CPU |
Examples#
Create a group and add some CPUs and cores to it
$ dcgmi group -c cpu_group1 -a cpu:{0-3},core:{0-287}
Watch field temperatures using dcgmi dmon
$ dcgmi dmon --entity-id cpu:{0-3} -e 1130