GPU Monitoring

GPU Monitoring#

AMD SMI: GPU Monitoring

Functions
amdsmi_status_t	amdsmi_get_gpu_activity (amdsmi_processor_handle processor_handle, amdsmi_engine_usage_t *info)
	Returns the current usage of the GPU engines (GFX, MM and MEM). Each usage is reported as a percentage from 0-100%. More...

amdsmi_status_t	amdsmi_get_power_info (amdsmi_processor_handle processor_handle, uint32_t sensor_ind, amdsmi_power_info_t *info)
	Returns the current power and voltage of the GPU. More...

amdsmi_status_t	amdsmi_set_power_cap (amdsmi_processor_handle processor_handle, uint32_t sensor_ind, uint64_t cap)
	Sets GPU power cap. More...

amdsmi_status_t	amdsmi_is_gpu_power_management_enabled (amdsmi_processor_handle processor_handle, bool *enabled)
	Returns is power management enabled. More...

amdsmi_status_t	amdsmi_get_clock_info (amdsmi_processor_handle processor_handle, amdsmi_clk_type_t clk_type, amdsmi_clk_info_t *info)
	Returns the measurements of the clocks in the GPU for the GFX and multimedia engines and Memory. This call reports the averages over 1s in MHz. For clk_type AMDSMI_CLK_TYPE_GFX cur_clk is expected to be larger than max_clk in some cases due to decoupled nature of master vs slave oscillator in DFLL clk_locked supported only for AMDSMI_CLK_TYPE_GFX. More...

amdsmi_status_t	amdsmi_get_temp_metric (amdsmi_processor_handle processor_handle, amdsmi_temperature_type_t sensor_type, amdsmi_temperature_metric_t metric, int64_t *temperature)
	Returns temperature measurements of the GPU. The results are in °C. More...

amdsmi_status_t	amdsmi_get_gpu_cache_info (amdsmi_processor_handle processor_handle, amdsmi_gpu_cache_info_t *info)
	Returns gpu cache info. More...

amdsmi_status_t	amdsmi_get_gpu_metrics (amdsmi_processor_handle processor_handle, uint32_t metrics_size, amdsmi_metric_t metrics)
	Returns metrics information. More...

amdsmi_status_t	amdsmi_get_soc_pstate (amdsmi_processor_handle processor_handle, amdsmi_dpm_policy_t *policy)
	Returns the soc pstate policy for the processor. More...

amdsmi_status_t	amdsmi_set_soc_pstate (amdsmi_processor_handle processor_handle, uint32_t policy_id)
	Set the soc pstate policy for the processor. More...

Detailed Description

Function Documentation

◆ amdsmi_get_gpu_activity()

amdsmi_status_t amdsmi_get_gpu_activity	(	amdsmi_processor_handle	processor_handle,
		amdsmi_engine_usage_t *	info
	)

Returns the current usage of the GPU engines (GFX, MM and MEM). Each usage is reported as a percentage from 0-100%.

Parameters

[in]	processor_handle	PF of a processor for which to query
[out]	info	Reference to the gpu engine usage structure. Must be allocated by user.

Returns: amdsmi_status_t | AMDSMI_STATUS_SUCCESS on success, non-zero on fail

◆ amdsmi_get_power_info()

amdsmi_status_t amdsmi_get_power_info	(	amdsmi_processor_handle	processor_handle,
		uint32_t	sensor_ind,
		amdsmi_power_info_t *	info
	)

Returns the current power and voltage of the GPU.

Note: amdsmi_power_info_t::socket_power metric can rarely spike above the socket power limit in some cases

Parameters

[in]	processor_handle	PF of a processor for which to query
[in]	sensor_ind	a 0-based sensor index. Normally, this will be 0. If a processor has more than one sensor, it could be greater than 0. Parameter `sensor_ind` is unused on

Platform:: host.

Parameters

[out] info Reference to the gpu power structure. Must be allocated by user.

Returns: amdsmi_status_t | AMDSMI_STATUS_SUCCESS on success, non-zero on fail

◆ amdsmi_set_power_cap()

amdsmi_status_t amdsmi_set_power_cap	(	amdsmi_processor_handle	processor_handle,
		uint32_t	sensor_ind,
		uint64_t	cap
	)

Sets GPU power cap.

Platform:

host

gpu_bm_linux

Set the power cap to the provided value cap. cap must be between the minimum (min_power_cap) and maximum (max_power_cap) power cap values, which can be obtained from amdsmi_power_cap_info_t.

Parameters

[in]	processor_handle	processor handle
[in]	sensor_ind	a 0-based sensor index. Normally, this will be 0. If a processor has more than one sensor, it could be greater than 0. Parameter `sensor_ind` is unused on

Platform:: host.

Parameters

[in] cap value representing power cap to set

Returns: amdsmi_status_t | AMDSMI_STATUS_SUCCESS on success, non-zero on fail

◆ amdsmi_is_gpu_power_management_enabled()

amdsmi_status_t amdsmi_is_gpu_power_management_enabled	(	amdsmi_processor_handle	processor_handle,
		bool *	enabled
	)

Returns is power management enabled.

Parameters

[in]	processor_handle	PF of a processor for which to query
[out]	enabled	Reference to bool. Must be allocated by user.

Returns: amdsmi_status_t | AMDSMI_STATUS_SUCCESS on success, non-zero on fail

◆ amdsmi_get_clock_info()

amdsmi_status_t amdsmi_get_clock_info	(	amdsmi_processor_handle	processor_handle,
		amdsmi_clk_type_t	clk_type,
		amdsmi_clk_info_t *	info
	)

Returns the measurements of the clocks in the GPU for the GFX and multimedia engines and Memory. This call reports the averages over 1s in MHz. For clk_type AMDSMI_CLK_TYPE_GFX cur_clk is expected to be larger than max_clk in some cases due to decoupled nature of master vs slave oscillator in DFLL clk_locked supported only for AMDSMI_CLK_TYPE_GFX.

Parameters

[in]	processor_handle	PF of a processor for which to query
[in]	clk_type	Enum representing the clock type to query.
[out]	info	Reference to the gpu clock structure. Must be allocated by user.

Returns: amdsmi_status_t | AMDSMI_STATUS_SUCCESS on success, non-zero on fail

◆ amdsmi_get_temp_metric()

amdsmi_status_t amdsmi_get_temp_metric	(	amdsmi_processor_handle	processor_handle,
		amdsmi_temperature_type_t	sensor_type,
		amdsmi_temperature_metric_t	metric,
		int64_t *	temperature
	)

Returns temperature measurements of the GPU. The results are in °C.

Parameters

[in]	processor_handle	PF of a processor for which to query
[in]	sensor_type	Enum representing GPU sensor to query.
[in]	metric	Enum representing the temperature metric to query (min, max, etc.)
[out]	temperature	Reference to the current, limit or shutdown temperature measured (depending on the metric parameter) Current temp is obtained for metric=AMDSMI_TEMP_CURRENT Limit temp is obtained for metric=AMDSMI_TEMP_CRITICAL Must be allocated by user.

Returns: amdsmi_status_t | AMDSMI_STATUS_SUCCESS on success, non-zero on fail

◆ amdsmi_get_gpu_cache_info()

amdsmi_status_t amdsmi_get_gpu_cache_info	(	amdsmi_processor_handle	processor_handle,
		amdsmi_gpu_cache_info_t *	info
	)

Returns gpu cache info.

Parameters

[in]	processor_handle	PF of a processor for which to query
[out]	info	reference to the cache info struct. Must be allocated by user.

Returns: amdsmi_status_t | AMDSMI_STATUS_SUCCESS on success, non-zero on fail

◆ amdsmi_get_gpu_metrics()

amdsmi_status_t amdsmi_get_gpu_metrics	(	amdsmi_processor_handle	processor_handle,
		uint32_t *	metrics_size,
		amdsmi_metric_t *	metrics
	)

Returns metrics information.

Parameters

[in]	processor_handle	PF of a processor for which to query
[in,out]	metrics_size	As input, the size of the provided buffer. As output, number of metrics in the buffer. Parameter must be allocated by user.
[out]	metrics	Reference to list of metrics returned by the library. Buffer must be allocated by user.

Returns: amdsmi_status_t | AMDSMI_STATUS_SUCCESS on success, non-zero on fail

◆ amdsmi_get_soc_pstate()

amdsmi_status_t amdsmi_get_soc_pstate	(	amdsmi_processor_handle	processor_handle,
		amdsmi_dpm_policy_t *	policy
	)

Returns the soc pstate policy for the processor.

Platform:

gpu_bm_linux

guest_1vf

host

Given a processor handle processor_handle, this function will write current soc pstate policy settings to policy. All the processors at the same socket will have the same policy.

Parameters

[in]	processor_handle	a processor handle
[out]	policy	the soc pstate policy for this processor. If this parameter is nullptr, this function will return AMDSMI_STATUS_INVAL

Returns: amdsmi_status_t | AMDSMI_STATUS_SUCCESS on success, non-zero on fail

◆ amdsmi_set_soc_pstate()

amdsmi_status_t amdsmi_set_soc_pstate	(	amdsmi_processor_handle	processor_handle,
		uint32_t	policy_id
	)

Set the soc pstate policy for the processor.

Platform:

gpu_bm_linux

guest_1vf

host

Given a processor handle processor_handle and a soc pstate policy policy_id, this function will set the soc pstate policy for this processor. All the processors at the same socket will be set to the same policy.

Parameters

[in]	processor_handle	a processor handle
[in]	policy_id	the soc pstate policy id to set. The id is the id in amdsmi_dpm_policy_entry_t, which can be obtained by calling amdsmi_get_soc_pstate()

Returns: amdsmi_status_t | AMDSMI_STATUS_SUCCESS on success, non-zero on fail

GPU Monitoring

GPU Monitoring#

Functions

Detailed Description

Function Documentation

◆ amdsmi_get_gpu_activity()

◆ amdsmi_get_power_info()

◆ amdsmi_set_power_cap()

◆ amdsmi_is_gpu_power_management_enabled()

◆ amdsmi_get_clock_info()

◆ amdsmi_get_temp_metric()

◆ amdsmi_get_gpu_cache_info()

◆ amdsmi_get_gpu_metrics()

◆ amdsmi_get_soc_pstate()

◆ amdsmi_set_soc_pstate()