amdsmi_gpu_metrics_t Struct Reference

amdsmi_gpu_metrics_t Struct Reference#

AMD SMI: amdsmi_gpu_metrics_t Struct Reference

Structure holds the gpu metrics values for a device. More...

#include <amdsmi.h>

Data Fields
amd_metrics_table_header_t	common_header

uint16_t	temperature_edge
	v1.0 Base More...

uint16_t	temperature_hotspot

uint16_t	temperature_mem

uint16_t	temperature_vrgfx

uint16_t	temperature_vrsoc

uint16_t	temperature_vrmem

uint16_t	average_gfx_activity
	Average Utilization (in %) More...

uint16_t	average_umc_activity
	memory controller

uint16_t	average_mm_activity
	UVD or VCN.

uint16_t	average_socket_power
	Power (W) /Energy (15.259uJ per 1ns)

uint64_t	energy_accumulator
	v1 mod. (32->64)

uint64_t	system_clock_counter
	Driver attached timestamp (in ns) More...

uint16_t	average_gfxclk_frequency
	Average clocks (MHz)

uint16_t	average_socclk_frequency

uint16_t	average_uclk_frequency

uint16_t	average_vclk0_frequency

uint16_t	average_dclk0_frequency

uint16_t	average_vclk1_frequency

uint16_t	average_dclk1_frequency

uint16_t	current_gfxclk
	Current clocks (MHz)

uint16_t	current_socclk

uint16_t	current_uclk

uint16_t	current_vclk0

uint16_t	current_dclk0

uint16_t	current_vclk1

uint16_t	current_dclk1

uint32_t	throttle_status
	Throttle status.

uint16_t	current_fan_speed
	Fans (RPM)

uint16_t	pcie_link_width
	Link width (number of lanes) /speed (0.1 GT/s) More...

uint16_t	pcie_link_speed
	in 0.1 GT/s; v1 mod. (8->16)

uint32_t	gfx_activity_acc
	new in v1

uint32_t	mem_activity_acc
	new in v1

uint16_t	temperature_hbm [AMDSMI_NUM_HBM_INSTANCES]
	new in v1

uint64_t	firmware_timestamp
	PMFW attached timestamp (10ns resolution)

uint16_t	voltage_soc
	In mV.

uint16_t	voltage_gfx
	In mV.

uint16_t	voltage_mem
	In mV.

uint64_t	indep_throttle_status
	Throttle status.

uint16_t	current_socket_power
	In Watts.

uint16_t	vcn_activity [AMDSMI_MAX_NUM_VCN]
	Utilization (%)

uint32_t	gfxclk_lock_status
	Clock Lock Status. Each bit corresponds to clock instance.

uint16_t	xgmi_link_width
	XGMI bus width in GB/s.

uint16_t	xgmi_link_speed
	XGMI bus bitrate in GB/s.

uint64_t	pcie_bandwidth_acc
	PCIE accumulated bandwidth (GB/sec)

uint64_t	pcie_bandwidth_inst
	PCIE instantaneous bandwidth (GB/sec)

uint64_t	pcie_l0_to_recov_count_acc
	PCIE L0 to recovery state transition accumulated count.

uint64_t	pcie_replay_count_acc
	PCIE replay accumulated count.

uint64_t	pcie_replay_rover_count_acc
	PCIE replay rollover accumulated count.

uint64_t	xgmi_read_data_acc [AMDSMI_MAX_NUM_XGMI_LINKS]
	XGMI accumulated data transfer size(KiloBytes) More...

uint64_t	xgmi_write_data_acc [AMDSMI_MAX_NUM_XGMI_LINKS]
	In KB.

uint16_t	current_gfxclks [AMDSMI_MAX_NUM_GFX_CLKS]
	XGMI current data transfer size(KiloBytes) More...

uint16_t	current_socclks [AMDSMI_MAX_NUM_CLKS]
	In KB.

uint16_t	current_vclk0s [AMDSMI_MAX_NUM_CLKS]
	In KB.

uint16_t	current_dclk0s [AMDSMI_MAX_NUM_CLKS]
	In KB.

uint16_t	jpeg_activity [AMDSMI_MAX_NUM_JPEG]
	v1.5 additions More...

uint32_t	pcie_nak_sent_count_acc
	PCIE NAK sent accumulated count.

uint32_t	pcie_nak_rcvd_count_acc
	PCIE NAK received accumulated count.

uint64_t	accumulation_counter
	v1.6 additions More...

uint64_t	prochot_residency_acc
	Accumulated throttler residencies.

uint64_t	ppt_residency_acc
	Accumulated throttler residencies. More...

uint64_t	socket_thm_residency_acc
	Accumulated throttler residencies. More...

uint64_t	vr_thm_residency_acc

uint64_t	hbm_thm_residency_acc

uint16_t	num_partition
	Number of current partition.

amdsmi_gpu_xcp_metrics_t	xcp_stats [AMDSMI_MAX_NUM_XCP]
	XCP (Graphic Cluster Partitions) metrics stats.

uint32_t	pcie_lc_perf_other_end_recovery
	PCIE other end recovery counter.

uint64_t	vram_max_bandwidth
	v1.7 additions More...

uint16_t	xgmi_link_status [AMDSMI_MAX_NUM_XGMI_LINKS]
	XGMI link status(up/down)

Detailed Description

Structure holds the gpu metrics values for a device.

This structure is extended to fit the needs of different GPU metric versions when exposing data through the structure. Depending on the version, some data members will hold data, and some will not. A good example is the set of 'current clocks': current_gfxclk, current_socclk, current_vclk0, current_dclk0. These are single-valued data members, up to version 1.3. For version 1.4 and up these are multi-valued data members (arrays) and their counterparts; current_gfxclks[], current_socclks[], current_vclk0s[], current_dclk0s[], will hold the data

Field Documentation

◆ temperature_edge

uint16_t amdsmi_gpu_metrics_t::temperature_edge

v1.0 Base

Temperature in C

◆ average_gfx_activity

uint16_t amdsmi_gpu_metrics_t::average_gfx_activity

Average Utilization (in %)

gfx

◆ system_clock_counter

uint64_t amdsmi_gpu_metrics_t::system_clock_counter

Driver attached timestamp (in ns)

v1 mod. (moved from top of struct)

◆ pcie_link_width

uint16_t amdsmi_gpu_metrics_t::pcie_link_width

Link width (number of lanes) /speed (0.1 GT/s)

v1 mod.(8->16)

◆ xgmi_read_data_acc

uint64_t amdsmi_gpu_metrics_t::xgmi_read_data_acc[AMDSMI_MAX_NUM_XGMI_LINKS]

XGMI accumulated data transfer size(KiloBytes)

In KB

◆ current_gfxclks

uint16_t amdsmi_gpu_metrics_t::current_gfxclks[AMDSMI_MAX_NUM_GFX_CLKS]

XGMI current data transfer size(KiloBytes)

In KB

◆ jpeg_activity

uint16_t amdsmi_gpu_metrics_t::jpeg_activity[AMDSMI_MAX_NUM_JPEG]

v1.5 additions

JPEG activity percent (encode/decode)

◆ accumulation_counter

uint64_t amdsmi_gpu_metrics_t::accumulation_counter

v1.6 additions

Accumulation cycle counter

◆ ppt_residency_acc

uint64_t amdsmi_gpu_metrics_t::ppt_residency_acc

Accumulated throttler residencies.

Prochot (thermal) - PPT (power) Package Power Tracking (PPT) violation % (greater than 0% is a violation); aka PVIOL

Ex. PVIOL/TVIOL calculations Where A and B are measurments recorded at prior points in time. Typically A is the earlier measured value and B is the latest measured value.

PVIOL % = (PptResidencyAcc (B) - PptResidencyAcc (A)) * 100/ (AccumulationCounter (B) - AccumulationCounter (A)) TVIOL % = (SocketThmResidencyAcc (B) - SocketThmResidencyAcc (A)) * 100 / (AccumulationCounter (B) - AccumulationCounter (A))

◆ socket_thm_residency_acc

uint64_t amdsmi_gpu_metrics_t::socket_thm_residency_acc