SELECT GPU METRICS

The SELECT gpu_metrics utility function allows you to check your cluster’s GPU usage within a specific time frame. This is useful to ensure that GPU usage stays within the allowed license limits.

An empty result indicates no usage deviation during the specified time. If the GPU usage exceeds the quota, the result will show the highest deviation in this format: Year-Month-Day, GPU deviated usage, and GPU quota limit as per your license plan.

Syntax

SELECT gpu_metrics(['monthly'] | ['daily'], <'start-date'>, <'end-date'>)

Parameters

Parameter	State	Description
`monthly` or `daily`	Mandatory	Specifies the time frame within which data was read
`start-date`	Mandatory	The starting date for the data retrieval period
`end-date`	Mandatory	The ending date for the data retrieval period

Output

Parameter	Description
`date`	Date and time of highest GPU usage deviation
`actual_number_of_gpus`	GPU usage deviation
`data_read_limit_license_value`	GPU quota limit as per license plan

Examples

Daily GPU usage:

SELECT gpu_metrics('daily','2023-05-01', '2023-05-05);

Output

 date         | actual_number_of_gpus   | data_read_limit_license_value
--------------+-------------------------+---------------------------------
 2023-May-01  | 2                       | 1
 2023-May-02  | 3                       | 1
 2023-May-03  | 3                       | 1

Monthly GPU usage:

SELECT gpu_metrics('monthly', '2023-04-01', '2023-06-05');

Output

 date         | actual_number_of_gpus   | data_read_limit_license_value
--------------+-------------------------+---------------------------------
 2023 Apr     | 2                       | 1

Permissions

Using the SELECT gpu_metrics command requires SUPERUSER permissions.