Changelog for AMD SMI Library#
Full documentation for amd_smi_lib is available at https://rocm.docs.amd.com/projects/amdsmi.
All information listed below is for reference and subject to change.
amd_smi_lib for ROCm 6.3.1#
Added#
Changed#
Changed
amd-smi monitor: No longer displayENC_CLOCK/DEC_CLOCKbutVCLOCKandDCLOCK.
Due to fix mentioned inResolved Issues, this change was needed.
Reason: Navi products use vclk and dclk for both encode and decode. On MI products, only decode is supported.
Before:$ amd-smi monitor -n -d GPU ENC_UTIL ENC_CLOCK DEC_UTIL DEC_CLOCK 0 0.0 % 29 MHz N/A 22 MHz 1 0.0 % 29 MHz N/A 22 MHz 2 0.0 % 29 MHz N/A 22 MHz 3 0.0 % 29 MHz N/A 22 MHz 4 0.0 % 29 MHz N/A 22 MHz 5 0.0 % 29 MHz N/A 22 MHz 6 0.0 % 29 MHz N/A 22 MHz 7 0.0 % 29 MHz N/A 22 MHz
After:
$ amd-smi monitor -n -d GPU ENC_UTIL DEC_UTIL VCLOCK DCLOCK 0 N/A 0.0 % 29 MHz 22 MHz 1 N/A 0.0 % 29 MHz 22 MHz 2 N/A 0.0 % 29 MHz 22 MHz 3 N/A 0.0 % 29 MHz 22 MHz 4 N/A 0.0 % 29 MHz 22 MHz 5 N/A 0.0 % 29 MHz 22 MHz 6 N/A 0.0 % 29 MHz 22 MHz 7 N/A 0.0 % 29 MHz 22 MHz
Removed#
Optimized#
Resolved issues#
Fixed
amd-smi monitor’s encode/decode:ENC_UTIL,DEC_UTIL, and now associateVCLOCK/DCLOCKwith both.
Navi products use vclk and dclk for both encode and decode. On MI products, only decode is supported.Navi products cannot support displaying ENC_UTIL % at this time.
Before:
$ amd-smi monitor -n -d GPU ENC_UTIL ENC_CLOCK DEC_UTIL DEC_CLOCK 0 0.0 % 29 MHz N/A 22 MHz 1 0.0 % 29 MHz N/A 22 MHz 2 0.0 % 29 MHz N/A 22 MHz 3 0.0 % 29 MHz N/A 22 MHz 4 0.0 % 29 MHz N/A 22 MHz 5 0.0 % 29 MHz N/A 22 MHz 6 0.0 % 29 MHz N/A 22 MHz 7 0.0 % 29 MHz N/A 22 MHz
After:
$ amd-smi monitor -n -d GPU ENC_UTIL DEC_UTIL VCLOCK DCLOCK 0 N/A 0.0 % 29 MHz 22 MHz 1 N/A 0.0 % 29 MHz 22 MHz 2 N/A 0.0 % 29 MHz 22 MHz 3 N/A 0.0 % 29 MHz 22 MHz 4 N/A 0.0 % 29 MHz 22 MHz 5 N/A 0.0 % 29 MHz 22 MHz 6 N/A 0.0 % 29 MHz 22 MHz 7 N/A 0.0 % 29 MHz 22 MHz
Upcoming changes#
Known issues#
amd_smi_lib for ROCm 6.3.0#
Added#
Added support for
amd-smi metric --ecc&amd-smi metric --ecc-blockson Guest VMs.
Guest VMs now support getting current ECC counts and ras information from the Host cards.Added support for GPU metrics 1.6 to
amdsmi_get_gpu_metrics_info().
Updatedamdsmi_get_gpu_metrics_info()and structureamdsmi_gpu_metrics_tto include new fields for PVIOL / TVIOL, XCP (Graphics Compute Partitions) stats, and pcie_lc_perf_other_end_recovery:uint64_t accumulation_counter- used for all throttled calculationsuint64_t prochot_residency_acc- Processor hot accumulatoruint64_t ppt_residency_acc- Package Power Tracking (PPT) accumulator (used in PVIOL calculations)uint64_t socket_thm_residency_acc- Socket thermal accumulator - (used in TVIOL calculations)uint64_t vr_thm_residency_acc- Voltage Rail (VR) thermal accumulatoruint64_t hbm_thm_residency_acc- High Bandwidth Memory (HBM) thermal accumulatoruint16_t num_partition- corresponds to the current total number of partitionsstruct amdgpu_xcp_metrics_t xcp_stats[MAX_NUM_XCP]- for each partition associated with current GPU, provides gfx busy & accumulators, jpeg, and decoder (VCN) engine utilizationsuint32_t gfx_busy_inst[MAX_NUM_XCC]- graphic engine utilization (%)uint16_t jpeg_busy[MAX_NUM_JPEG_ENGS]- jpeg engine utilization (%)uint16_t vcn_busy[MAX_NUM_VCNS]- decoder (VCN) engine utilization (%)uint64_t gfx_busy_acc[MAX_NUM_XCC]- graphic engine utilization accumulated (%)
uint32_t pcie_lc_perf_other_end_recovery- corresponds to the pcie other end recovery counter
Added new violation status outputs and APIs:
amdsmi_status_t amdsmi_get_violation_status(),amd-smi metric --throttle, andamd-smi monitor --violation.
Only available for MI300+ ASICs.
Users can now retrieve violation status’ through either our Python or C++ APIs. Additionally, we have added capability to view these outputs conviently throughamd-smi metric --throttleandamd-smi monitor --violation.
Example outputs are listed below (below is for reference, output is subject to change):
$ amd-smi metric --throttle
GPU: 0
THROTTLE:
ACCUMULATION_COUNTER: 3808991
PROCHOT_ACCUMULATED: 0
PPT_ACCUMULATED: 585613
SOCKET_THERMAL_ACCUMULATED: 2190
VR_THERMAL_ACCUMULATED: 0
HBM_THERMAL_ACCUMULATED: 0
PROCHOT_VIOLATION_STATUS: NOT ACTIVE
PPT_VIOLATION_STATUS: NOT ACTIVE
SOCKET_THERMAL_VIOLATION_STATUS: NOT ACTIVE
VR_THERMAL_VIOLATION_STATUS: NOT ACTIVE
HBM_THERMAL_VIOLATION_STATUS: NOT ACTIVE
PROCHOT_VIOLATION_ACTIVITY: 0 %
PPT_VIOLATION_ACTIVITY: 0 %
SOCKET_THERMAL_VIOLATION_ACTIVITY: 0 %
VR_THERMAL_VIOLATION_ACTIVITY: 0 %
HBM_THERMAL_VIOLATION_ACTIVITY: 0 %
GPU: 1
THROTTLE:
ACCUMULATION_COUNTER: 3806335
PROCHOT_ACCUMULATED: 0
PPT_ACCUMULATED: 586332
SOCKET_THERMAL_ACCUMULATED: 18010
VR_THERMAL_ACCUMULATED: 0
HBM_THERMAL_ACCUMULATED: 0
PROCHOT_VIOLATION_STATUS: NOT ACTIVE
PPT_VIOLATION_STATUS: NOT ACTIVE
SOCKET_THERMAL_VIOLATION_STATUS: NOT ACTIVE
VR_THERMAL_VIOLATION_STATUS: NOT ACTIVE
HBM_THERMAL_VIOLATION_STATUS: NOT ACTIVE
PROCHOT_VIOLATION_ACTIVITY: 0 %
PPT_VIOLATION_ACTIVITY: 0 %
SOCKET_THERMAL_VIOLATION_ACTIVITY: 0 %
VR_THERMAL_VIOLATION_ACTIVITY: 0 %
HBM_THERMAL_VIOLATION_ACTIVITY: 0 %
...
$ amd-smi monitor --violation
GPU PVIOL TVIOL PHOT_TVIOL VR_TVIOL HBM_TVIOL
0 0 % 0 % 0 % 0 % 0 %
1 0 % 0 % 0 % 0 % 0 %
2 0 % 0 % 0 % 0 % 0 %
3 0 % 0 % 0 % 0 % 0 %
4 0 % 0 % 0 % 0 % 0 %
5 0 % 0 % 0 % 0 % 0 %
6 0 % 0 % 0 % 0 % 0 %
7 0 % 0 % 0 % 0 % 0 %
8 0 % 0 % 0 % 0 % 0 %
9 0 % 0 % 0 % 0 % 0 %
10 0 % 0 % 0 % 0 % 0 %
11 0 % 0 % 0 % 0 % 0 %
12 0 % 0 % 0 % 0 % 0 %
13 0 % 0 % 0 % 0 % 0 %
14 0 % 0 % 0 % 0 % 0 %
15 0 % 0 % 0 % 0 % 0 %
...
Added ability to view XCP (Graphics Compute Partition) activity within
amd-smi metric --usage.
Partition specific features are only available on MI300+ ASICs
Users can now retrieve graphic utilization statistic on a per-XCP (per-partition) basis. Here all XCP activities will be listed, but the current XCP is the partition id listed under bothamd-smi listandamd-smi static --partition.
Example outputs are listed below (below is for reference, output is subject to change):
$ amd-smi metric --usage
GPU: 0
USAGE:
GFX_ACTIVITY: 0 %
UMC_ACTIVITY: 0 %
MM_ACTIVITY: N/A
VCN_ACTIVITY: [0 %, N/A, N/A, N/A]
JPEG_ACTIVITY: [0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A]
GFX_BUSY_INST:
XCP_0: [0 %, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_1: [0 %, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_2: [0 %, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_3: [0 %, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_4: [0 %, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_5: [0 %, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_6: [0 %, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_7: [0 %, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
JPEG_BUSY:
XCP_0: [0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A]
XCP_1: [0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A]
XCP_2: [0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A]
XCP_3: [0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A]
XCP_4: [0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A]
XCP_5: [0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A]
XCP_6: [0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A]
XCP_7: [0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A]
VCN_BUSY:
XCP_0: [0 %, N/A, N/A, N/A]
XCP_1: [0 %, N/A, N/A, N/A]
XCP_2: [0 %, N/A, N/A, N/A]
XCP_3: [0 %, N/A, N/A, N/A]
XCP_4: [0 %, N/A, N/A, N/A]
XCP_5: [0 %, N/A, N/A, N/A]
XCP_6: [0 %, N/A, N/A, N/A]
XCP_7: [0 %, N/A, N/A, N/A]
GFX_BUSY_ACC:
XCP_0: [N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_1: [N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_2: [N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_3: [N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_4: [N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_5: [N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_6: [N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_7: [N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
GPU: 1
USAGE:
GFX_ACTIVITY: 0 %
UMC_ACTIVITY: 0 %
MM_ACTIVITY: N/A
VCN_ACTIVITY: [0 %, N/A, N/A, N/A]
JPEG_ACTIVITY: [0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A]
GFX_BUSY_INST:
XCP_0: [0 %, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_1: [0 %, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_2: [0 %, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_3: [0 %, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_4: [0 %, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_5: [0 %, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_6: [0 %, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_7: [0 %, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
JPEG_BUSY:
XCP_0: [0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A]
XCP_1: [0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A]
XCP_2: [0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A]
XCP_3: [0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A]
XCP_4: [0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A]
XCP_5: [0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A]
XCP_6: [0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A]
XCP_7: [0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A,
N/A, N/A, N/A]
VCN_BUSY:
XCP_0: [0 %, N/A, N/A, N/A]
XCP_1: [0 %, N/A, N/A, N/A]
XCP_2: [0 %, N/A, N/A, N/A]
XCP_3: [0 %, N/A, N/A, N/A]
XCP_4: [0 %, N/A, N/A, N/A]
XCP_5: [0 %, N/A, N/A, N/A]
XCP_6: [0 %, N/A, N/A, N/A]
XCP_7: [0 %, N/A, N/A, N/A]
GFX_BUSY_ACC:
XCP_0: [N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_1: [N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_2: [N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_3: [N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_4: [N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_5: [N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_6: [N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
XCP_7: [N/A, N/A, N/A, N/A, N/A, N/A, N/A, N/A]
...
Added
LC_PERF_OTHER_END_RECOVERYCLI output toamd-smi metric --pcieand updatedamdsmi_get_pcie_info()to include this value.
Feature is only available on MI300+ ASICs
Users can now retrieve both throughamdsmi_get_pcie_info()which has an updated structure:
typedef struct {
...
struct pcie_metric_ {
uint16_t pcie_width; //!< current PCIe width
uint32_t pcie_speed; //!< current PCIe speed in MT/s
uint32_t pcie_bandwidth; //!< current instantaneous PCIe bandwidth in Mb/s
uint64_t pcie_replay_count; //!< total number of the replays issued on the PCIe link
uint64_t pcie_l0_to_recovery_count; //!< total number of times the PCIe link transitioned from L0 to the recovery state
uint64_t pcie_replay_roll_over_count; //!< total number of replay rollovers issued on the PCIe link
uint64_t pcie_nak_sent_count; //!< total number of NAKs issued on the PCIe link by the device
uint64_t pcie_nak_received_count; //!< total number of NAKs issued on the PCIe link by the receiver
uint32_t pcie_lc_perf_other_end_recovery_count; //!< PCIe other end recovery counter
uint64_t reserved[12];
} pcie_metric;
uint64_t reserved[32];
} amdsmi_pcie_info_t;
Example outputs are listed below (below is for reference, output is subject to change):
$ amd-smi metric --pcie
GPU: 0
PCIE:
WIDTH: 16
SPEED: 32 GT/s
BANDWIDTH: 18 Mb/s
REPLAY_COUNT: 0
L0_TO_RECOVERY_COUNT: 0
REPLAY_ROLL_OVER_COUNT: 0
NAK_SENT_COUNT: 0
NAK_RECEIVED_COUNT: 0
CURRENT_BANDWIDTH_SENT: N/A
CURRENT_BANDWIDTH_RECEIVED: N/A
MAX_PACKET_SIZE: N/A
LC_PERF_OTHER_END_RECOVERY: 0
GPU: 1
PCIE:
WIDTH: 16
SPEED: 32 GT/s
BANDWIDTH: 18 Mb/s
REPLAY_COUNT: 0
L0_TO_RECOVERY_COUNT: 0
REPLAY_ROLL_OVER_COUNT: 0
NAK_SENT_COUNT: 0
NAK_RECEIVED_COUNT: 0
CURRENT_BANDWIDTH_SENT: N/A
CURRENT_BANDWIDTH_RECEIVED: N/A
MAX_PACKET_SIZE: N/A
LC_PERF_OTHER_END_RECOVERY: 0
...
Added retrieving a set of GPUs that are nearest to a given device at a specific link type level.
Added
amdsmi_get_link_topology_nearest()function to amd-smi C and Python Libraries.
Added more supported utilization count types to
amdsmi_get_utilization_count().Added
amd-smi set -L/--clk-limit ...command.
Equivalent to rocm-smi’s ‘–extremum’ command which sets sclk’s or mclk’s soft minimum or soft maximum clock frequency.Added unittest functionality to test amdsmi API calls in Python.
Changed the
powerparameter inamdsmi_get_energy_count()toenergy_accumulator.Changes propagate forwards into the python interface as well, however we are maintaing backwards compatibility and keeping the
powerfield in the python API until ROCm 6.4.
Added GPU memory overdrive percentage to
amd-smi metric -o.Added
amdsmi_get_gpu_mem_overdrive_level()function to amd-smi C and Python Libraries.
Added retrieving connection type and P2P capabilities between two GPUs.
Added
amdsmi_topo_get_p2p_status()function to amd-smi C and Python Libraries.Added retrieving P2P link capabilities to CLI
amd-smi topology.
$ amd-smi topology -h
usage: amd-smi topology [-h] [--json | --csv] [--file FILE] [--loglevel LEVEL]
[-g GPU [GPU ...]] [-a] [-w] [-o] [-t] [-b]
If no GPU is specified, returns information for all GPUs on the system.
If no topology argument is provided all topology information will be displayed.
Topology arguments:
-h, --help show this help message and exit
-g, --gpu GPU [GPU ...] Select a GPU ID, BDF, or UUID from the possible choices:
ID: 0 | BDF: 0000:0c:00.0 | UUID: <redacted>
ID: 1 | BDF: 0000:22:00.0 | UUID: <redacted>
ID: 2 | BDF: 0000:38:00.0 | UUID: <redacted>
ID: 3 | BDF: 0000:5c:00.0 | UUID: <redacted>
ID: 4 | BDF: 0000:9f:00.0 | UUID: <redacted>
ID: 5 | BDF: 0000:af:00.0 | UUID: <redacted>
ID: 6 | BDF: 0000:bf:00.0 | UUID: <redacted>
ID: 7 | BDF: 0000:df:00.0 | UUID: <redacted>
all | Selects all devices
-a, --access Displays link accessibility between GPUs
-w, --weight Displays relative weight between GPUs
-o, --hops Displays the number of hops between GPUs
-t, --link-type Displays the link type between GPUs
-b, --numa-bw Display max and min bandwidth between nodes
-c, --coherent Display cache coherant (or non-coherant) link capability between nodes
-n, --atomics Display 32 and 64-bit atomic io link capability between nodes
-d, --dma Display P2P direct memory access (DMA) link capability between nodes
-z, --bi-dir Display P2P bi-directional link capability between nodes
Command Modifiers:
--json Displays output in JSON format (human readable by default).
--csv Displays output in CSV format (human readable by default).
--file FILE Saves output into a file on the provided path (stdout by default).
--loglevel LEVEL Set the logging level from the possible choices:
DEBUG, INFO, WARNING, ERROR, CRITICAL
$ amd-smi topology -cndz
CACHE COHERANCY TABLE:
0000:0c:00.0 0000:22:00.0 0000:38:00.0 0000:5c:00.0 0000:9f:00.0 0000:af:00.0 0000:bf:00.0 0000:df:00.0
0000:0c:00.0 SELF C NC NC C C C NC
0000:22:00.0 C SELF NC C C C NC C
0000:38:00.0 NC NC SELF C C NC C NC
0000:5c:00.0 NC C C SELF NC C NC NC
0000:9f:00.0 C C C NC SELF NC NC C
0000:af:00.0 C C NC C NC SELF C C
0000:bf:00.0 C NC C NC NC C SELF NC
0000:df:00.0 NC C NC NC C C NC SELF
ATOMICS TABLE:
0000:0c:00.0 0000:22:00.0 0000:38:00.0 0000:5c:00.0 0000:9f:00.0 0000:af:00.0 0000:bf:00.0 0000:df:00.0
0000:0c:00.0 SELF 64,32 64,32 64 32 32 N/A 64,32
0000:22:00.0 64,32 SELF 64 32 32 N/A 64,32 64,32
0000:38:00.0 64,32 64 SELF 32 N/A 64,32 64,32 64,32
0000:5c:00.0 64 32 32 SELF 64,32 64,32 64,32 32
0000:9f:00.0 32 32 N/A 64,32 SELF 64,32 32 32
0000:af:00.0 32 N/A 64,32 64,32 64,32 SELF 32 N/A
0000:bf:00.0 N/A 64,32 64,32 64,32 32 32 SELF 64,32
0000:df:00.0 64,32 64,32 64,32 32 32 N/A 64,32 SELF
DMA TABLE:
0000:0c:00.0 0000:22:00.0 0000:38:00.0 0000:5c:00.0 0000:9f:00.0 0000:af:00.0 0000:bf:00.0 0000:df:00.0
0000:0c:00.0 SELF T T F F T F T
0000:22:00.0 T SELF F F T F T T
0000:38:00.0 T F SELF T F T T T
0000:5c:00.0 F F T SELF T T T F
0000:9f:00.0 F T F T SELF T F F
0000:af:00.0 T F T T T SELF F T
0000:bf:00.0 F T T T F F SELF F
0000:df:00.0 T T T F F T F SELF
BI-DIRECTIONAL TABLE:
0000:0c:00.0 0000:22:00.0 0000:38:00.0 0000:5c:00.0 0000:9f:00.0 0000:af:00.0 0000:bf:00.0 0000:df:00.0
0000:0c:00.0 SELF T T F F T F T
0000:22:00.0 T SELF F F T F T T
0000:38:00.0 T F SELF T F T T T
0000:5c:00.0 F F T SELF T T T F
0000:9f:00.0 F T F T SELF T F F
0000:af:00.0 T F T T T SELF F T
0000:bf:00.0 F T T T F F SELF F
0000:df:00.0 T T T F F T F SELF
Legend:
SELF = Current GPU
ENABLED / DISABLED = Link is enabled or disabled
N/A = Not supported
T/F = True / False
C/NC = Coherant / Non-Coherant io links
64,32 = 64 bit and 32 bit atomic support
<BW from>-<BW to>
Created new amdsmi_kfd_info_t and added information under
amd-smi list.Due to fixes needed to properly enumerate all logical GPUs in CPX, new device identifiers were added in to a new
amdsmi_kfd_info_twhich gets populated via the APIamdsmi_get_gpu_kfd_info().This info has been added to the
amd-smi list.These new fields are only available for BM/Guest Linux devices at this time.
typedef struct {
uint64_t kfd_id; //< 0xFFFFFFFFFFFFFFFF if not supported
uint32_t node_id; //< 0xFFFFFFFF if not supported
uint32_t current_partition_id; //< 0xFFFFFFFF if not supported
uint32_t reserved[12];
} amdsmi_kfd_info_t;
$ amd-smi list
GPU: 0
BDF: 0000:23:00.0
UUID: <redacted>
KFD_ID: 45412
NODE_ID: 1
PARTITION_ID: 0
GPU: 1
BDF: 0000:26:00.0
UUID: <redacted>
KFD_ID: 59881
NODE_ID: 2
PARTITION_ID: 0
Added Subsystem Device ID to
amd-smi static --asic.No underlying changes to amdsmi_get_gpu_asic_info
$ amd-smi static --asic
GPU: 0
ASIC:
MARKET_NAME: MI308X
VENDOR_ID: 0x1002
VENDOR_NAME: Advanced Micro Devices Inc. [AMD/ATI]
SUBVENDOR_ID: 0x1002
DEVICE_ID: 0x74a2
SUBSYSTEM_ID: 0x74a2
REV_ID: 0x00
ASIC_SERIAL: <redacted>
OAM_ID: 5
NUM_COMPUTE_UNITS: 20
TARGET_GRAPHICS_VERSION: gfx942
Added Target_Graphics_Version to
amd-smi static --asicandamdsmi_get_gpu_asic_info().
typedef struct {
char market_name[AMDSMI_256_LENGTH];
uint32_t vendor_id; //< Use 32 bit to be compatible with other platform.
char vendor_name[AMDSMI_MAX_STRING_LENGTH];
uint32_t subvendor_id; //< The subsystem vendor id
uint64_t device_id; //< The device id of a GPU
uint32_t rev_id;
char asic_serial[AMDSMI_NORMAL_STRING_LENGTH];
uint32_t oam_id; //< 0xFFFF if not supported
uint32_t num_of_compute_units; //< 0xFFFFFFFF if not supported
uint64_t target_graphics_version; //< 0xFFFFFFFFFFFFFFFF if not supported
uint32_t reserved[15];
} amdsmi_asic_info_t;
$ amd-smi static --asic
GPU: 0
ASIC:
MARKET_NAME: MI308X
VENDOR_ID: 0x1002
VENDOR_NAME: Advanced Micro Devices Inc. [AMD/ATI]
SUBVENDOR_ID: 0x1002
DEVICE_ID: 0x74a2
SUBSYSTEM_ID: 0x74a2
REV_ID: 0x00
ASIC_SERIAL: <redacted>
OAM_ID: 5
NUM_COMPUTE_UNITS: 20
TARGET_GRAPHICS_VERSION: gfx942
Changed#
Improvement: Users now have the ability to set and reset without providing
-g allusing AMD SMI CLI.
Users can now provide set and reset without-g all. Previously, users were required to provide:
sudo amd-smi set -g all <set arguments>orsudo amd-smi reset -g all <set arguments>
This update allows users to set or reset without providing-g allarguments. Allowing commands:
sudo amd-smi set <set arguments>orsudo amd-smi reset <set arguments>
This action will default to try to set/reset for all AMD GPUs on the user’s system.Improvement:
amd-smi set --memory-partitionnow includes a warning banner and progress bar.
For devices which support dynamically changing memory partitions, we now provide a warning for users. We provide this warning to provide users knowledge that this action requires users to quit any gpu workloads. Also to let them know this process will trigger an AMD GPU driver reload. Since this process takes time to complete, a progress bar has been provided until actions can verified as a successful change. Otherwise, AMD SMI will report any errors to users and what actions can be taken. See example below:
$ sudo amd-smi set -M NPS1
****** WARNING ******
Setting Dynamic Memory (NPS) partition modes require users to quit all GPU workloads.
AMD SMI will then attempt to change memory (NPS) partition mode.
Upon a successful set, AMD SMI will then initiate an action to restart amdgpu driver.
This action will change all GPU's in the hive to the requested memory (NPS) partition mode.
Please use this utility with caution.
Do you accept these terms? [Y/N] y
Updating memory partition for gpu 0: [████████████████████████████████████████] 40/40 secs remain
GPU: 0
MEMORYPARTITION: Successfully set memory partition to NPS1
GPU: 1
MEMORYPARTITION: Successfully set memory partition to NPS1
GPU: 2
MEMORYPARTITION: Successfully set memory partition to NPS1
...
Updated
amdsmi_get_gpu_accelerator_partition_profileto provide driver memory partition capablities.
Driver now has the ability to report what the user can set memory partition modes to. User can now see available memory partition modes upon an invalid argument return from memory partition mode set (amdsmi_set_gpu_memory_partition). This change also updatesamd-smi partition,amd-smi partition --memory, andamd-smi partition --accelerator(*see note below)
*Note: Subject to change for ROCm 6.4Updated
amdsmi_set_gpu_memory_partitionto not return until a successful restart of AMD GPU Driver.
This change keeps checking for ~ up to 40 seconds for a successful restart of the AMD GPU driver. Additionally, the API call continues to check if memory partition (NPS) SYSFS files are successfully updated to reflect the user’s requested memory partition (NPS) mode change. Otherwise, reports an error back to the user. Due to these changes, we have updated AMD SMI’s CLI to reflect the maximum wait of 40 seconds, while a memory partition change is in progress.All APIs now have the ability to catch driver reporting invalid arguments.
Now AMD SMI APIs can show AMDSMI_STATUS_INVAL when driver returns EINVAL.
For example, if user tries to set to NPS8, but the memory partition mode is not an available mode to set to. Commonly referred to asCAPS(seeamd-smi partition --memory), provided byamdsmi_get_gpu_accelerator_partition_profile(*see note below).
*Note: Subject to change for ROCm 6.4Updated BDF commands to look use KFD SYSFS for BDF:
amdsmi_get_gpu_device_bdf().
This aligns BDF output with ROCm SMI. See below for overview as seen fromrsmi_dev_pci_id_get()now provides partition ID. See API for better detail. Previously these bits were reserved bits (right before domain) and partition id was within function.bits [63:32] = domain
bits [31:28] = partition id
bits [27:16] = reserved
bits [15: 0] = pci bus/device/function
Moved python tests directory path install location.
/opt/<rocm-path>/share/amd_smi/pytest/..to/opt/<rocm-path>/share/amd_smi/tests/python_unittest/..On amd-smi-lib-tests uninstall, the amd_smi tests folder is removed
Removed pytest dependency, our python testing now only depends on the unittest framework.
Updated Partition APIs and struct information and added and partition_id to
amd-smi static --partition.As part of an overhaul to partition information, some partition information will be made available in the
amdsmi_accelerator_partition_profile_t.This struct will be filled out by a new API,
amdsmi_get_gpu_accelerator_partition_profile().Future data from these APIs wil will eventually get added to
amd-smi partition.
#define AMDSMI_MAX_ACCELERATOR_PROFILE 32
#define AMDSMI_MAX_CP_PROFILE_RESOURCES 32
#define AMDSMI_MAX_ACCELERATOR_PARTITIONS 8
/**
* @brief Accelerator Partition. This enum is used to identify
* various accelerator partitioning settings.
*/
typedef enum {
AMDSMI_ACCELERATOR_PARTITION_INVALID = 0,
AMDSMI_ACCELERATOR_PARTITION_SPX, //!< Single GPU mode (SPX)- All XCCs work
//!< together with shared memory
AMDSMI_ACCELERATOR_PARTITION_DPX, //!< Dual GPU mode (DPX)- Half XCCs work
//!< together with shared memory
AMDSMI_ACCELERATOR_PARTITION_TPX, //!< Triple GPU mode (TPX)- One-third XCCs
//!< work together with shared memory
AMDSMI_ACCELERATOR_PARTITION_QPX, //!< Quad GPU mode (QPX)- Quarter XCCs
//!< work together with shared memory
AMDSMI_ACCELERATOR_PARTITION_CPX, //!< Core mode (CPX)- Per-chip XCC with
//!< shared memory
} amdsmi_accelerator_partition_type_t;
/**
* @brief Possible Memory Partition Modes.
* This union is used to identify various memory partitioning settings.
*/
typedef union {
struct {
uint32_t nps1_cap :1; // bool 1 = true; 0 = false; Max uint32 means unsupported
uint32_t nps2_cap :1; // bool 1 = true; 0 = false; Max uint32 means unsupported
uint32_t nps4_cap :1; // bool 1 = true; 0 = false; Max uint32 means unsupported
uint32_t nps8_cap :1; // bool 1 = true; 0 = false; Max uint32 means unsupported
uint32_t reserved :28;
} amdsmi_nps_flags_t;
uint32_t nps_cap_mask;
} amdsmi_nps_caps_t;
typedef struct {
amdsmi_accelerator_partition_type_t profile_type; // SPX, DPX, QPX, CPX and so on
uint32_t num_partitions; // On MI300X, SPX: 1, DPX: 2, QPX: 4, CPX: 8, length of resources array
uint32_t profile_index;
amdsmi_nps_caps_t memory_caps; // Possible memory partition capabilities
uint32_t num_resources; // length of index_of_resources_profile
uint32_t resources[AMDSMI_MAX_ACCELERATOR_PARTITIONS][AMDSMI_MAX_CP_PROFILE_RESOURCES];
uint64_t reserved[6];
} amdsmi_accelerator_partition_profile_t;
$ amd-smi static --partition
GPU: 0
PARTITION:
COMPUTE_PARTITION: CPX
MEMORY_PARTITION: NPS4
PARTITION_ID: 0
Removed#
Removed
amd-smi reset --compute-partitionand... --memory-partitionand associated APIs.This change is part of the partition redesign. Reset functionality will be reintroduced in a later update.
associated APIs include
amdsmi_reset_gpu_compute_partition()andamdsmi_reset_gpu_memory_partition()
Removed usage of _validate_positive in Parser and replaced with _positive_int and _not_negative_int as appropriate.
This will allow 0 to be a valid input for several options in setting CPUs where appropriate (for example, as a mode or NBIOID)
Optimized#
Adjusted ordering of gpu_metrics calls to ensure that pcie_bw values remain stable in
amd-smi metric&amd-smi monitor.With this change additional padding was added to PCIE_BW
amd-smi monitor --pcie
Resolved issues#
Fixed
amdsmi_get_gpu_asic_info’starget_graphics_versionandamd-smi --asicnot displaying properly for MI2x or Navi 3x ASICs.Fixed
amd-smi resetcommands showing an AttributeError.Improved Offline install process & lowered dependency for PyYAML.
Fixed CPX not showing total number of logical GPUs.
Updates were made to
amdsmi_init()andamdsmi_get_gpu_bdf_id(..). In order to display all logical devices, we needed a way to provide order to GPU’s enumerated. This was done by adding a partition_id within the BDF optional pci_id bits.Due to driver changes in KFD, some devices may report bits [31:28] or [2:0]. With the newly added
amdsmi_get_gpu_bdf_id(..), we provided this fallback to properly retreive partition ID. We plan to eventually remove partition ID from the function portion of the BDF (Bus Device Function). See below for PCI ID description.bits [63:32] = domain
bits [31:28] or bits [2:0] = partition id
bits [27:16] = reserved
bits [15:8] = Bus
bits [7:3] = Device
bits [2:0] = Function (partition id maybe in bits [2:0]) <– Fallback for non SPX modes
Previously in non-SPX modes (ex. CPX/TPX/DPX/etc) some MI3x ASICs would not report all logical GPU devices within AMD SMI.
$ amd-smi monitor -p -t -v
GPU POWER GPU_TEMP MEM_TEMP VRAM_USED VRAM_TOTAL
0 248 W 55 °C 48 °C 283 MB 196300 MB
1 247 W 55 °C 48 °C 283 MB 196300 MB
2 247 W 55 °C 48 °C 283 MB 196300 MB
3 247 W 55 °C 48 °C 283 MB 196300 MB
4 221 W 50 °C 42 °C 283 MB 196300 MB
5 221 W 50 °C 42 °C 283 MB 196300 MB
6 222 W 50 °C 42 °C 283 MB 196300 MB
7 221 W 50 °C 42 °C 283 MB 196300 MB
8 239 W 53 °C 46 °C 283 MB 196300 MB
9 239 W 53 °C 46 °C 283 MB 196300 MB
10 239 W 53 °C 46 °C 283 MB 196300 MB
11 239 W 53 °C 46 °C 283 MB 196300 MB
12 219 W 51 °C 48 °C 283 MB 196300 MB
13 219 W 51 °C 48 °C 283 MB 196300 MB
14 219 W 51 °C 48 °C 283 MB 196300 MB
15 219 W 51 °C 48 °C 283 MB 196300 MB
16 222 W 51 °C 47 °C 283 MB 196300 MB
17 222 W 51 °C 47 °C 283 MB 196300 MB
18 222 W 51 °C 47 °C 283 MB 196300 MB
19 222 W 51 °C 48 °C 283 MB 196300 MB
20 241 W 55 °C 48 °C 283 MB 196300 MB
21 241 W 55 °C 48 °C 283 MB 196300 MB
22 241 W 55 °C 48 °C 283 MB 196300 MB
23 240 W 55 °C 48 °C 283 MB 196300 MB
24 211 W 51 °C 45 °C 283 MB 196300 MB
25 211 W 51 °C 45 °C 283 MB 196300 MB
26 211 W 51 °C 45 °C 283 MB 196300 MB
27 211 W 51 °C 45 °C 283 MB 196300 MB
28 227 W 51 °C 49 °C 283 MB 196300 MB
29 227 W 51 °C 49 °C 283 MB 196300 MB
30 227 W 51 °C 49 °C 283 MB 196300 MB
31 227 W 51 °C 49 °C 283 MB 196300 MB
Fixed incorrect implementation of the Python API
amdsmi_get_gpu_metrics_header_info().amdsmitstTestGpuMetricsRead now prints metric in correct units.
Upcoming changes#
Python API for
amdsmi_get_energy_count()will deprecate thepowerfield in ROCm 6.4 and useenergy_accumulatorfield instead.New memory and compute partition APIs incoming for ROCm 6.4.
These APIs will be updated to fully populate the CLI and allowing compute (accelerator) partitions to be set by profile ID.
One API will be provided, to reset both memory and compute (accelerator).
There are dependencies regarding available compute partitions when in other memory modes.
Driver will be providing these default modes
Memory partition resets (for BM) require driver reloads - this will allow us to notify users before taking this action, then change to the default compute partition modes.
The following APIs will remain:
amdsmi_status_t
amdsmi_set_gpu_compute_partition(amdsmi_processor_handle processor_handle,
amdsmi_compute_partition_type_t compute_partition);
amdsmi_status_t
amdsmi_get_gpu_compute_partition(amdsmi_processor_handle processor_handle,
char *compute_partition, uint32_t len);
amdsmi_status_t
amdsmi_get_gpu_memory_partition(amdsmi_processor_handle processor_handle,
char *memory_partition, uint32_t len);
amdsmi_status_t
amdsmi_set_gpu_memory_partition(amdsmi_processor_handle processor_handle,
amdsmi_memory_partition_type_t memory_partition);
amd-smi set –compute-partition “SPX/DPX/CPX…” will no longer be supported for ROCm 6.4.
This is due to aligning with Host setups and providing more robust partition information through the APIs outlined above. Furthermore, new APIs which will be available on both BM/Host can set by profile ID. (functionality coming soon!)
Added preliminary
amd-smi partitioncommand.The new partition command can be used to display GPU information, including memory and accelerator partition information.
The command will be at full functionality once additional partition information from
amdsmi_get_gpu_accelerator_partition_profile()has been implemented.
amd_smi_lib for ROCm 6.2.1#
Added#
Removed
amd-smi metric --ecc&amd-smi metric --ecc-blockson Guest VMs. Guest VMs do not support getting current ECC counts from the Host cards.Added
amd-smi static --rason Guest VMs. Guest VMs can view enabled/disabled ras features that are on Host cards.
Resolved issues#
Fixed TypeError in
amd-smi process -G.Updated CLI error strings to handle empty and invalid GPU/CPU inputs.
Fixed Guest VM showing passthrough options.
Fixed firmware formatting where leading 0s were missing.
amd_smi_lib for ROCm 6.2.0#
Added#
amd-smi dmonis now available as an alias toamd-smi monitor.Added optional process table under
amd-smi monitor -q.
The monitor subcommand within the CLI Tool now has the-qoption to enable an optional process table underneath the original monitored output.
$ amd-smi monitor -q
GPU POWER GPU_TEMP MEM_TEMP GFX_UTIL GFX_CLOCK MEM_UTIL MEM_CLOCK ENC_UTIL ENC_CLOCK DEC_UTIL DEC_CLOCK SINGLE_ECC DOUBLE_ECC PCIE_REPLAY VRAM_USED VRAM_TOTAL PCIE_BW
0 199 W 103 °C 84 °C 99 % 1920 MHz 31 % 1000 MHz N/A 0 MHz N/A 0 MHz 0 0 0 1235 MB 16335 MB N/A Mb/s
PROCESS INFO:
GPU NAME PID GTT_MEM CPU_MEM VRAM_MEM MEM_USAGE GFX ENC
0 rvs 1564865 0.0 B 0.0 B 1.1 GB 0.0 B 0 ns 0 ns
Added Handling to detect VMs with passthrough configurations in CLI Tool.
CLI Tool had only allowed a restricted set of options for Virtual Machines with passthrough GPUs. Now we offer an expanded set of functions availble to passthrough configured GPUs.Added Process Isolation and Clear SRAM functionality to the CLI Tool for VMs.
VMs now have the ability to set the process isolation and clear the sram from the CLI tool. Using the following commands
amd-smi set --process-isolation <0 or 1>
amd-smi reset --clean_local_data
Added macros that were in
amdsmi.hto the amdsmi Python libraryamdsmi_interface.py.
Added macros to reference max size limitations for certain amdsmi functions such as max dpm policies and max fanspeed.Added Ring Hang event.
AddedAMDSMI_EVT_NOTIF_RING_HANGto the possible events in theamdsmi_evt_notification_type_tenum.
Optimized#
Updated CLI error strings to specify invalid device type queried
$ amd-smi static --asic --gpu 123123
Can not find a device: GPU '123123' Error code: -3
Removed elevated permission requirements for
amdsmi_get_gpu_process_list().
Previously if a processes with elevated permissions was running amd-smi would required sudo to display all output. Now amd-smi will populate all process data and return N/A for elevated process names instead. However if ran with sudo you will be able to see the name like so:
$ amd-smi process
GPU: 0
PROCESS_INFO:
NAME: N/A
PID: 1693982
MEMORY_USAGE:
GTT_MEM: 0.0 B
CPU_MEM: 0.0 B
VRAM_MEM: 10.1 GB
MEM_USAGE: 0.0 B
USAGE:
GFX: 0 ns
ENC: 0 ns
$ sudo amd-smi process
GPU: 0
PROCESS_INFO:
NAME: TransferBench
PID: 1693982
MEMORY_USAGE:
GTT_MEM: 0.0 B
CPU_MEM: 0.0 B
VRAM_MEM: 10.1 GB
MEM_USAGE: 0.0 B
USAGE:
GFX: 0 ns
ENC: 0 ns
Updated naming for
amdsmi_set_gpu_clear_sram_data()toamdsmi_clean_gpu_local_data().
Changed the naming to be more accurate to what the function was doing. This change also extends to the CLI where we changed theclear-sram-datacommand toclean_local_data.Updated
amdsmi_clk_info_tstruct in amdsmi.h and amdsmi_interface.py to align with host/guest.
Changed cur_clk to clk, changed sleep_clk to clk_deep_sleep, and added clk_locked value. New struct will be in the following format:
typedef struct {
+ uint32_t clk;
uint32_t min_clk;
uint32_t max_clk;
+ uint8_t clk_locked;
+ uint8_t clk_deep_sleep;
uint32_t reserved[4];
} amdsmi_clk_info_t;
Multiple structure updates in amdsmi.h and amdsmi_interface.py to align with host/guest.
Multiple structures used by APIs were changed for alignment unification:Changed
amdsmi_vram_info_tvram_size_mbfield changed to tovram_sizeUpdated
amdsmi_vram_type_tstruct updated to include new enums and addedAMDSMIprefixUpdated
amdsmi_status_tsome enums were missing theAMDSMI_STATUSprefixAdded
AMDSMI_PROCESSOR_TYPEprefix toprocessor_type_tenumsRemoved the fields structure definition in favor for an anonymous definition in
amdsmi_bdf_t
Added
AMDSMIprefix in amdsmi.h and amdsmi_interface.py to align with host/guest.
Multiple structures used by APIs were changed for alignment unification.AMDSMIprefix was added to the following structures:Added AMDSMI prefix to
amdsmi_container_types_tenumsAdded AMDSMI prefix to
amdsmi_clk_type_tenumsAdded AMDSMI prefix to
amdsmi_compute_partition_type_tenumsAdded AMDSMI prefix to
amdsmi_memory_partition_type_tenumsAdded AMDSMI prefix to
amdsmi_clk_type_tenumsAdded AMDSMI prefix to
amdsmi_temperature_type_tenumsAdded AMDSMI prefix to
amdsmi_fw_block_tenums
Changed dpm_policy references to soc_pstate.
The file structure referenced to dpm_policy changed to soc_pstate and we have changed the APIs and CLI tool to be inline with the current structure.amdsmi_get_dpm_policy()andamdsmi_set_dpm_policy()is no longer valid with the new API beingamdsmi_get_soc_pstate()andamdsmi_set_soc_pstate(). The CLI tool has been changed from--policyto--soc-pstateUpdated
amdsmi_get_gpu_board_info()product_name to fallback to pciids.
Previously on devices without a FRU we would not populate the product name in theamdsmi_board_info_tstructure, now we will fallback to using the name listed according to the pciids file if available.Updated CLI voltage curve command output.
The output foramd-smi metric --voltage-curvenow splits the frequency and voltage output by curve point or outputs N/A for each curve point if not applicable
GPU: 0
VOLTAGE_CURVE:
POINT_0_FREQUENCY: 872 Mhz
POINT_0_VOLTAGE: 736 mV
POINT_1_FREQUENCY: 1354 Mhz
POINT_1_VOLTAGE: 860 mV
POINT_2_FREQUENCY: 1837 Mhz
POINT_2_VOLTAGE: 1186 mV
Updated
amdsmi_get_gpu_board_info()now has larger structure sizes foramdsmi_board_info_t.
Updated sizes that work for retreiving relavant board information across AMD’s ASIC products. This requires users to update any ABIs using this structure.
Resolved issues#
Fixed Leftover Mutex deadlock when running multiple instances of the CLI tool.
When runningamd-smi reset --gpureset --gpu alland then running an instance ofamd-smi static(or any other subcommand that access the GPUs) a mutex would lock and not return requiring either a clear of the mutex in /dev/shm or rebooting the machine.Fixed multiple processes not being registered in
amd-smi processwith json and csv format.
Multiple process outputs in the CLI tool were not being registered correctly. The json output did not handle multiple processes and is now in a new valid json format:
[
{
"gpu": 0,
"process_list": [
{
"process_info": {
"name": "TransferBench",
"pid": 420157,
"mem_usage": {
"value": 0,
"unit": "B"
}
}
},
{
"process_info": {
"name": "rvs",
"pid": 420315,
"mem_usage": {
"value": 0,
"unit": "B"
}
}
}
]
}
]
Removed
throttle-statusfromamd-smi monitoras it is no longer reliably supported.
Throttle status may work for older ASICs, but will be replaced with PVIOL and TVIOL metrics for future ASIC support. It remains a field in the gpu_metrics API and inamd-smi metric --power.amdsmi_get_gpu_board_info()no longer returns junk char strings.
Previously if there was a partial failure to retrieve character strings, we would return garbage output to users using the API. This fix intends to populate as many values as possible. Then any failure(s) found along the way,\0is provided toamdsmi_board_info_tstructures data members which cannot be populated. Ensuring empty char string values.Fixed parsing of
pp_od_clk_voltagewithinamdsmi_get_gpu_od_volt_info.
The parsing ofpp_od_clk_voltagewas not dynamic enough to work with the dropping of voltage curve support on MI series cards. This propagates down to correcting the CLI’s outputamd-smi metric --voltage-curveto N/A if voltage curve is not enabled.
Known issues#
amdsmi_get_gpu_process_isolationandamdsmi_clean_gpu_local_datacommands do no currently work and will be supported in a future release.
amd_smi_lib for ROCm 6.1.2#
Added#
Added process isolation and clean shader APIs and CLI commands.
Added APIs CLI and APIs to address LeftoverLocals security issues. Allowing clearing the sram data and setting process isolation on a per GPU basis. New APIs:amdsmi_get_gpu_process_isolation()amdsmi_set_gpu_process_isolation()amdsmi_set_gpu_clear_sram_data()
Added
MIN_POWERto output ofamd-smi static --limit.
This change helps users identify the range to which they can change the power cap of the GPU. The change is added to simplify why a device supports (or does not support) power capping (also known as overdrive). Seeamd-smi set -g all --power-cap <value in W>oramd-smi reset -g all --power-cap.
$ amd-smi static --limit
GPU: 0
LIMIT:
MAX_POWER: 203 W
MIN_POWER: 0 W
SOCKET_POWER: 203 W
SLOWDOWN_EDGE_TEMPERATURE: 100 °C
SLOWDOWN_HOTSPOT_TEMPERATURE: 110 °C
SLOWDOWN_VRAM_TEMPERATURE: 100 °C
SHUTDOWN_EDGE_TEMPERATURE: 105 °C
SHUTDOWN_HOTSPOT_TEMPERATURE: 115 °C
SHUTDOWN_VRAM_TEMPERATURE: 105 °C
GPU: 1
LIMIT:
MAX_POWER: 213 W
MIN_POWER: 213 W
SOCKET_POWER: 213 W
SLOWDOWN_EDGE_TEMPERATURE: 109 °C
SLOWDOWN_HOTSPOT_TEMPERATURE: 110 °C
SLOWDOWN_VRAM_TEMPERATURE: 100 °C
SHUTDOWN_EDGE_TEMPERATURE: 114 °C
SHUTDOWN_HOTSPOT_TEMPERATURE: 115 °C
SHUTDOWN_VRAM_TEMPERATURE: 105 °C
Optimized#
Updated
amd-smi monitor --pcieoutput.
The source for pcie bandwidth monitor output was a legacy file we no longer support and was causing delays within the monitor command. The output is no longer using TX/RX but instantaneous bandwidth from gpu_metrics instead; updated output:
$ amd-smi monitor --pcie
GPU PCIE_BW
0 26 Mb/s
amdsmi_get_power_cap_infonow returns values in uW instead of W.
amdsmi_get_power_cap_infowill return in uW as originally reflected by driver. Previouslyamdsmi_get_power_cap_inforeturned W values, this conflicts with our sets and modifies values retrieved from driver. We decided to keep the values returned from driver untouched (in original units, uW). Then in CLI we will convert to watts (as previously done - no changes here). Additionally, driver made updates to min power cap displayed for devices when overdrive is disabled which prompted for this change (in this case min_power_cap and max_power_cap are the same).Updated Python Library return types for amdsmi_get_gpu_memory_reserved_pages & amdsmi_get_gpu_bad_page_info.
Previously calls were returning “No bad pages found.” if no pages were found, now it only returns the list type and can be empty.Updated
amd-smi metric --ecc-blocksoutput.
The ecc blocks argument was outputing blocks without counters available, updated the filtering show blocks that counters are available for:
$ amd-smi metric --ecc-block
GPU: 0
ECC_BLOCKS:
UMC:
CORRECTABLE_COUNT: 0
UNCORRECTABLE_COUNT: 0
DEFERRED_COUNT: 0
SDMA:
CORRECTABLE_COUNT: 0
UNCORRECTABLE_COUNT: 0
DEFERRED_COUNT: 0
GFX:
CORRECTABLE_COUNT: 0
UNCORRECTABLE_COUNT: 0
DEFERRED_COUNT: 0
MMHUB:
CORRECTABLE_COUNT: 0
UNCORRECTABLE_COUNT: 0
DEFERRED_COUNT: 0
PCIE_BIF:
CORRECTABLE_COUNT: 0
UNCORRECTABLE_COUNT: 0
DEFERRED_COUNT: 0
HDP:
CORRECTABLE_COUNT: 0
UNCORRECTABLE_COUNT: 0
DEFERRED_COUNT: 0
XGMI_WAFL:
CORRECTABLE_COUNT: 0
UNCORRECTABLE_COUNT: 0
DEFERRED_COUNT: 0
Removed
amdsmi_get_gpu_process_infofrom Python library.
amdsmi_get_gpu_process_info was removed from the C library in an earlier build, but the API was still in the Python interface.
Resolved issues#
Fixed
amd-smi metric --powernow provides power output for Navi2x/Navi3x/MI1x.
These systems use an older version of gpu_metrics in amdgpu. This fix only updates what CLI outputs. No change in any of our APIs.
$ amd-smi metric --power
GPU: 0
POWER:
SOCKET_POWER: 11 W
GFX_VOLTAGE: 768 mV
SOC_VOLTAGE: 925 mV
MEM_VOLTAGE: 1250 mV
POWER_MANAGEMENT: ENABLED
THROTTLE_STATUS: UNTHROTTLED
GPU: 1
POWER:
SOCKET_POWER: 17 W
GFX_VOLTAGE: 781 mV
SOC_VOLTAGE: 806 mV
MEM_VOLTAGE: 1250 mV
POWER_MANAGEMENT: ENABLED
THROTTLE_STATUS: UNTHROTTLED
Fixed
amdsmitstReadWrite.TestPowerCapReadWritetest for Navi3X, Navi2X, MI100.
Updates requiredamdsmi_get_power_cap_infoto return in uW as originally reflected by driver. Previouslyamdsmi_get_power_cap_inforeturned W values, this conflicts with our sets and modifies values retrieved from driver. We decided to keep the values returned from driver untouched (in original units, uW). Then in CLI we will convert to watts (as previously done - no changes here). Additionally, driver made updates to min power cap displayed for devices when overdrive is disabled which prompted for this change (in this case min_power_cap and max_power_cap are the same).Fixed Python interface call amdsmi_get_gpu_memory_reserved_pages & amdsmi_get_gpu_bad_page_info.
Previously Python interface calls to populated bad pages resulted in aValueError: NULL pointer access. This fixes the bad-pages subcommand CLI subcommand as well.
amd_smi_lib for ROCm 6.1.1#
Changed#
Updated metrics –clocks.
Output foramd-smi metric --clockis updated to reflect each engine and bug fixes for the clock lock status and deep sleep status.
$ amd-smi metric --clock
GPU: 0
CLOCK:
GFX_0:
CLK: 113 MHz
MIN_CLK: 500 MHz
MAX_CLK: 1800 MHz
CLK_LOCKED: DISABLED
DEEP_SLEEP: ENABLED
GFX_1:
CLK: 113 MHz
MIN_CLK: 500 MHz
MAX_CLK: 1800 MHz
CLK_LOCKED: DISABLED
DEEP_SLEEP: ENABLED
GFX_2:
CLK: 112 MHz
MIN_CLK: 500 MHz
MAX_CLK: 1800 MHz
CLK_LOCKED: DISABLED
DEEP_SLEEP: ENABLED
GFX_3:
CLK: 113 MHz
MIN_CLK: 500 MHz
MAX_CLK: 1800 MHz
CLK_LOCKED: DISABLED
DEEP_SLEEP: ENABLED
GFX_4:
CLK: 113 MHz
MIN_CLK: 500 MHz
MAX_CLK: 1800 MHz
CLK_LOCKED: DISABLED
DEEP_SLEEP: ENABLED
GFX_5:
CLK: 113 MHz
MIN_CLK: 500 MHz
MAX_CLK: 1800 MHz
CLK_LOCKED: DISABLED
DEEP_SLEEP: ENABLED
GFX_6:
CLK: 113 MHz
MIN_CLK: 500 MHz
MAX_CLK: 1800 MHz
CLK_LOCKED: DISABLED
DEEP_SLEEP: ENABLED
GFX_7:
CLK: 113 MHz
MIN_CLK: 500 MHz
MAX_CLK: 1800 MHz
CLK_LOCKED: DISABLED
DEEP_SLEEP: ENABLED
MEM_0:
CLK: 900 MHz
MIN_CLK: 900 MHz
MAX_CLK: 1200 MHz
CLK_LOCKED: N/A
DEEP_SLEEP: DISABLED
VCLK_0:
CLK: 29 MHz
MIN_CLK: 914 MHz
MAX_CLK: 1480 MHz
CLK_LOCKED: N/A
DEEP_SLEEP: ENABLED
VCLK_1:
CLK: 29 MHz
MIN_CLK: 914 MHz
MAX_CLK: 1480 MHz
CLK_LOCKED: N/A
DEEP_SLEEP: ENABLED
VCLK_2:
CLK: 29 MHz
MIN_CLK: 914 MHz
MAX_CLK: 1480 MHz
CLK_LOCKED: N/A
DEEP_SLEEP: ENABLED
VCLK_3:
CLK: 29 MHz
MIN_CLK: 914 MHz
MAX_CLK: 1480 MHz
CLK_LOCKED: N/A
DEEP_SLEEP: ENABLED
DCLK_0:
CLK: 22 MHz
MIN_CLK: 711 MHz
MAX_CLK: 1233 MHz
CLK_LOCKED: N/A
DEEP_SLEEP: ENABLED
DCLK_1:
CLK: 22 MHz
MIN_CLK: 711 MHz
MAX_CLK: 1233 MHz
CLK_LOCKED: N/A
DEEP_SLEEP: ENABLED
DCLK_2:
CLK: 22 MHz
MIN_CLK: 711 MHz
MAX_CLK: 1233 MHz
CLK_LOCKED: N/A
DEEP_SLEEP: ENABLED
DCLK_3:
CLK: 22 MHz
MIN_CLK: 711 MHz
MAX_CLK: 1233 MHz
CLK_LOCKED: N/A
DEEP_SLEEP: ENABLED
Added deferred ecc counts.
Added deferred error correctable counts toamd-smi metric --ecc --ecc-blocks
$ amd-smi metric --ecc --ecc-blocks
GPU: 0
ECC:
TOTAL_CORRECTABLE_COUNT: 0
TOTAL_UNCORRECTABLE_COUNT: 0
TOTAL_DEFERRED_COUNT: 0
CACHE_CORRECTABLE_COUNT: 0
CACHE_UNCORRECTABLE_COUNT: 0
ECC_BLOCKS:
UMC:
CORRECTABLE_COUNT: 0
UNCORRECTABLE_COUNT: 0
DEFERRED_COUNT: 0
SDMA:
CORRECTABLE_COUNT: 0
UNCORRECTABLE_COUNT: 0
DEFERRED_COUNT: 0
...
Updated
amd-smi topology --jsonto align with host/guest.
Topology’s--jsonoutput now is changed to align with output host/guest systems. Additionally, users can select/filter specific topology details as desired (refer toamd-smi topology -hfor full list). See examples shown below.
Previous format:
$ amd-smi topology --json
[
{
"gpu": 0,
"link_accessibility": {
"gpu_0": "ENABLED",
"gpu_1": "DISABLED"
},
"weight": {
"gpu_0": 0,
"gpu_1": 40
},
"hops": {
"gpu_0": 0,
"gpu_1": 2
},
"link_type": {
"gpu_0": "SELF",
"gpu_1": "PCIE"
},
"numa_bandwidth": {
"gpu_0": "N/A",
"gpu_1": "N/A"
}
},
{
"gpu": 1,
"link_accessibility": {
"gpu_0": "DISABLED",
"gpu_1": "ENABLED"
},
"weight": {
"gpu_0": 40,
"gpu_1": 0
},
"hops": {
"gpu_0": 2,
"gpu_1": 0
},
"link_type": {
"gpu_0": "PCIE",
"gpu_1": "SELF"
},
"numa_bandwidth": {
"gpu_0": "N/A",
"gpu_1": "N/A"
}
}
]
New format:
$ amd-smi topology --json
[
{
"gpu": 0,
"bdf": "0000:01:00.0",
"links": [
{
"gpu": 0,
"bdf": "0000:01:00.0",
"weight": 0,
"link_status": "ENABLED",
"link_type": "SELF",
"num_hops": 0,
"bandwidth": "N/A",
},
{
"gpu": 1,
"bdf": "0001:01:00.0",
"weight": 15,
"link_status": "ENABLED",
"link_type": "XGMI",
"num_hops": 1,
"bandwidth": "50000-100000",
},
...
]
},
...
]
$ /opt/rocm/bin/amd-smi topology -a -t --json
[
{
"gpu": 0,
"bdf": "0000:08:00.0",
"links": [
{
"gpu": 0,
"bdf": "0000:08:00.0",
"link_status": "ENABLED",
"link_type": "SELF"
},
{
"gpu": 1,
"bdf": "0000:44:00.0",
"link_status": "DISABLED",
"link_type": "PCIE"
}
]
},
{
"gpu": 1,
"bdf": "0000:44:00.0",
"links": [
{
"gpu": 0,
"bdf": "0000:08:00.0",
"link_status": "DISABLED",
"link_type": "PCIE"
},
{
"gpu": 1,
"bdf": "0000:44:00.0",
"link_status": "ENABLED",
"link_type": "SELF"
}
]
}
]
Resolved issues#
Fix for GPU reset error on non-amdgpu cards.
Previously our reset could attempting to reset non-amd GPUS- resuting in “Unable to reset non-amd GPU” error. Fix updates CLI to target only AMD ASICs.Fix for
amd-smi static --pcieandamdsmi_get_pcie_info()Navi32/31 cards.
Updated API to includeamdsmi_card_form_factor_t.AMDSMI_CARD_FORM_FACTOR_CEM. Prevously, this would report “UNKNOWN”. This fix provides the correct boardSLOT_TYPEassociated with these ASICs (and other Navi cards).Fix for
amd-smi process.
Fixed output results when getting processes running on a device.Improved Error handling for
amd-smi process.
Fixed Attribute Error when getting process in csv format
Known issues#
amd-smi bad-pagescan results with “ValueError: NULL pointer access” with certain PM FW versions.
amd_smi_lib for ROCm 6.1.0#
Added#
Added Monitor Command.
Provides users the ability to customize GPU metrics to capture, collect, and observe. Output is provided in a table view. This aligns closer to ROCm SMIrocm-smi(no argument), additionally allows uers to customize what data is helpful for their use-case.
$ amd-smi monitor -h
usage: amd-smi monitor [-h] [--json | --csv] [--file FILE] [--loglevel LEVEL]
[-g GPU [GPU ...] | -U CPU [CPU ...] | -O CORE [CORE ...]]
[-w INTERVAL] [-W TIME] [-i ITERATIONS] [-p] [-t] [-u] [-m] [-n]
[-d] [-s] [-e] [-v] [-r]
Monitor a target device for the specified arguments.
If no arguments are provided, all arguments will be enabled.
Use the watch arguments to run continuously
Monitor Arguments:
-h, --help show this help message and exit
-g, --gpu GPU [GPU ...] Select a GPU ID, BDF, or UUID from the possible choices:
ID: 0 | BDF: 0000:01:00.0 | UUID: <redacted>
all | Selects all devices
-U, --cpu CPU [CPU ...] Select a CPU ID from the possible choices:
ID: 0
all | Selects all devices
-O, --core CORE [CORE ...] Select a Core ID from the possible choices:
ID: 0 - 23
all | Selects all devices
-w, --watch INTERVAL Reprint the command in a loop of INTERVAL seconds
-W, --watch_time TIME The total TIME to watch the given command
-i, --iterations ITERATIONS Total number of ITERATIONS to loop on the given command
-p, --power-usage Monitor power usage in Watts
-t, --temperature Monitor temperature in Celsius
-u, --gfx Monitor graphics utilization (%) and clock (MHz)
-m, --mem Monitor memory utilization (%) and clock (MHz)
-n, --encoder Monitor encoder utilization (%) and clock (MHz)
-d, --decoder Monitor decoder utilization (%) and clock (MHz)
-s, --throttle-status Monitor thermal throttle status
-e, --ecc Monitor ECC single bit, ECC double bit, and PCIe replay error counts
-v, --vram-usage Monitor memory usage in MB
-r, --pcie Monitor PCIe Tx/Rx in MB/s
Command Modifiers:
--json Displays output in JSON format (human readable by default).
--csv Displays output in CSV format (human readable by default).
--file FILE Saves output into a file on the provided path (stdout by default).
--loglevel LEVEL Set the logging level from the possible choices:
DEBUG, INFO, WARNING, ERROR, CRITICAL
$ amd-smi monitor -ptumv
GPU POWER GPU_TEMP MEM_TEMP GFX_UTIL GFX_CLOCK MEM_UTIL MEM_CLOCK VRAM_USED VRAM_TOTAL
0 171 W 32 °C 33 °C 0 % 114 MHz 0 % 900 MHz 283 MB 196300 MB
1 175 W 33 °C 34 °C 0 % 113 MHz 0 % 900 MHz 283 MB 196300 MB
2 177 W 31 °C 33 °C 0 % 113 MHz 0 % 900 MHz 283 MB 196300 MB
3 172 W 33 °C 32 °C 0 % 113 MHz 0 % 900 MHz 283 MB 196300 MB
4 178 W 32 °C 32 °C 0 % 113 MHz 0 % 900 MHz 284 MB 196300 MB
5 176 W 33 °C 35 °C 0 % 113 MHz 0 % 900 MHz 283 MB 196300 MB
6 176 W 32 °C 32 °C 0 % 113 MHz 0 % 900 MHz 283 MB 196300 MB
7 175 W 34 °C 32 °C 0 % 113 MHz 0 % 900 MHz 283 MB 196300 MB
Integrated ESMI Tool.
Users can get CPU metrics and telemetry through our API and CLI tools. This information can be seen inamd-smi staticandamd-smi metriccommands. Only available for limited target processors. As of ROCm 6.0.2, this is listed as:AMD Zen3 based CPU Family 19h Models 0h-Fh and 30h-3Fh
AMD Zen4 based CPU Family 19h Models 10h-1Fh and A0-AFh
See a few examples listed below.
$ amd-smi static -U all
CPU: 0
SMU:
FW_VERSION: 85.90.0
INTERFACE_VERSION:
PROTO VERSION: 6
$ amd-smi metric -O 0 1 2
CORE: 0
BOOST_LIMIT:
VALUE: 400 MHz
CURR_ACTIVE_FREQ_CORE_LIMIT:
VALUE: 400 MHz
CORE_ENERGY:
VALUE: N/A
CORE: 1
BOOST_LIMIT:
VALUE: 400 MHz
CURR_ACTIVE_FREQ_CORE_LIMIT:
VALUE: 400 MHz
CORE_ENERGY:
VALUE: N/A
CORE: 2
BOOST_LIMIT:
VALUE: 400 MHz
CURR_ACTIVE_FREQ_CORE_LIMIT:
VALUE: 400 MHz
CORE_ENERGY:
VALUE: N/A
$ amd-smi metric -U all
CPU: 0
POWER_METRICS:
SOCKET POWER: 102675 mW
SOCKET POWER LIMIT: 550000 mW
SOCKET MAX POWER LIMIT: 550000 mW
PROCHOT:
PROCHOT_STATUS: 0
FREQ_METRICS:
FCLKMEMCLK:
FCLK: 2000 MHz
MCLK: 1300 MHz
CCLKFREQLIMIT: 400 MHz
SOC_CURRENT_ACTIVE_FREQ_LIMIT:
FREQ: 400 MHz
FREQ_SRC: [HSMP Agent]
SOC_FREQ_RANGE:
MAX_SOCKET_FREQ: 3700 MHz
MIN_SOCKET_FREQ: 400 MHz
C0_RESIDENCY:
RESIDENCY: 4 %
SVI_TELEMETRY_ALL_RAILS:
POWER: 102673 mW
METRIC_VERSION:
VERSION: 11
METRICS_TABLE:
CPU_FAMILY: 25
CPU_MODEL: 144
RESPONSE:
MTBL_ACCUMULATION_COUNTER: 2887162626
MTBL_MAX_SOCKET_TEMPERATURE: 41.0 °C
MTBL_MAX_VR_TEMPERATURE: 39.0 °C
MTBL_MAX_HBM_TEMPERATURE: 40.0 °C
MTBL_MAX_SOCKET_TEMPERATURE_ACC: 108583340881.125 °C
MTBL_MAX_VR_TEMPERATURE_ACC: 109472702595.0 °C
MTBL_MAX_HBM_TEMPERATURE_ACC: 111516663941.0 °C
MTBL_SOCKET_POWER_LIMIT: 550.0 W
MTBL_MAX_SOCKET_POWER_LIMIT: 550.0 W
MTBL_SOCKET_POWER: 102.678 W
MTBL_TIMESTAMP_RAW: 288731677361880
MTBL_TIMESTAMP_READABLE: Tue Mar 19 12:32:21 2024
MTBL_SOCKET_ENERGY_ACC: 166127.84 kJ
MTBL_CCD_ENERGY_ACC: 3317.837 kJ
MTBL_XCD_ENERGY_ACC: 21889.147 kJ
MTBL_AID_ENERGY_ACC: 121932.397 kJ
MTBL_HBM_ENERGY_ACC: 18994.108 kJ
MTBL_CCLK_FREQUENCY_LIMIT: 3.7 GHz
MTBL_GFXCLK_FREQUENCY_LIMIT: 0.0 MHz
MTBL_FCLK_FREQUENCY: 1999.988 MHz
MTBL_UCLK_FREQUENCY: 1299.993 MHz
MTBL_SOCCLK_FREQUENCY: [35.716, 35.715, 35.714, 35.714] MHz
MTBL_VCLK_FREQUENCY: [0.0, 53.749, 53.749, 53.749] MHz
MTBL_DCLK_FREQUENCY: [7.143, 44.791, 44.791, 44.791] MHz
MTBL_LCLK_FREQUENCY: [20.872, 18.75, 35.938, 599.558] MHz
MTBL_FCLK_FREQUENCY_TABLE: [1200.0, 1600.0, 1900.0, 2000.0] MHz
MTBL_UCLK_FREQUENCY_TABLE: [900.0, 1100.0, 1200.0, 1300.0] MHz
MTBL_SOCCLK_FREQUENCY_TABLE: [800.0, 1000.0, 1142.857, 1142.857] MHz
MTBL_VCLK_FREQUENCY_TABLE: [914.286, 1300.0, 1560.0, 1720.0] MHz
MTBL_DCLK_FREQUENCY_TABLE: [711.111, 975.0, 1300.0, 1433.333] MHz
MTBL_LCLK_FREQUENCY_TABLE: [600.0, 844.444, 1150.0, 1150.0] MHz
MTBL_CCLK_FREQUENCY_ACC: [4399751656.639, 4399751656.639, 4399751656.639, 4399751656.639,
4399751656.639, 4399751656.639, 4399751656.639, 4399751656.639, 4399751656.639,
4399751656.639, 4399751656.639, 4399751656.639, 4399751656.639, 4399751656.639,
4399751656.639, 4399751656.639, 4399751656.639, 4399751656.639, 4399751656.639,
4399751656.639, 4399751656.639, 4399751656.639, 4399751656.639, 4399751656.639,
0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0] GHz
MTBL_GFXCLK_FREQUENCY_ACC: [0.0, 0.0, 250534397827.603, 251546257401.82, 250811364089.836,
249999070486.505, 251622633562.855, 251342375116.05] MHz
MTBL_GFXCLK_FREQUENCY: [0.0, 0.0, 31.091, 31.414, 31.141, 31.478, 31.32, 31.453]
MHz
MTBL_MAX_CCLK_FREQUENCY: 3.7 GHz
MTBL_MIN_CCLK_FREQUENCY: 0.4 GHz
MTBL_MAX_GFXCLK_FREQUENCY: 2100.0 MHz
MTBL_MIN_GFXCLK_FREQUENCY: 500.0 MHz
MTBL_MAX_LCLK_DPM_RANGE: 2
MTBL_MIN_LCLK_DPM_RANGE: 0
MTBL_XGMI_WIDTH: 0.0
MTBL_XGMI_BITRATE: 0.0 Gbps
MTBL_XGMI_READ_BANDWIDTH_ACC: [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0] Gbps
MTBL_XGMI_WRITE_BANDWIDTH_ACC: [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0] Gbps
MTBL_SOCKET_C0_RESIDENCY: 4.329 %
MTBL_SOCKET_GFX_BUSY: 0.0 %
MTBL_HBM_BANDWIDTH_UTILIZATION: 0.001 %
MTBL_SOCKET_C0_RESIDENCY_ACC: 311523106.34
MTBL_SOCKET_GFX_BUSY_ACC: 84739.281
MTBL_HBM_BANDWIDTH_ACC: 33231180.073 Gbps
MTBL_MAX_HBM_BANDWIDTH: 5324.801 Gbps
MTBL_DRAM_BANDWIDTH_UTILIZATION_ACC: 612843.699
MTBL_PCIE_BANDWIDTH_ACC: [0.0, 0.0, 0.0, 0.0] Gbps
MTBL_PROCHOT_RESIDENCY_ACC: 0
MTBL_PPT_RESIDENCY_ACC: 2887162626
MTBL_SOCKET_THM_RESIDENCY_ACC: 2887162626
MTBL_VR_THM_RESIDENCY_ACC: 0
MTBL_HBM_THM_RESIDENCY_ACC: 2887162626
SOCKET_ENERGY:
RESPONSE: N/A
DDR_BANDWIDTH:
RESPONSE: N/A
CPU_TEMP:
RESPONSE: N/A
Added support for new metrics: VCN, JPEG engines, and PCIe errors.
Using the AMD SMI tool, users can retreive VCN, JPEG engines, and PCIe errors by callingamd-smi metric -Poramd-smi metric --usage. Depending on device support,VCN_ACTIVITYwill update for MI3x ASICs (with 4 separate VCN engine activities) for older asicsMM_ACTIVITYwith UVD/VCN engine activity (average of all engines).JPEG_ACTIVITYis a new field for MI3x ASICs, where device can support up to 32 JPEG engine activities. See our documentation for more in-depth understanding of these new fields.
$ amd-smi metric -P
GPU: 0
PCIE:
WIDTH: 16
SPEED: 16 GT/s
REPLAY_COUNT: 0
L0_TO_RECOVERY_COUNT: 1
REPLAY_ROLL_OVER_COUNT: 0
NAK_SENT_COUNT: 0
NAK_RECEIVED_COUNT: 0
CURRENT_BANDWIDTH_SENT: N/A
CURRENT_BANDWIDTH_RECEIVED: N/A
MAX_PACKET_SIZE: N/A
$ amd-smi metric --usage
GPU: 0
USAGE:
GFX_ACTIVITY: 0 %
UMC_ACTIVITY: 0 %
MM_ACTIVITY: N/A
VCN_ACTIVITY: [0 %, 0 %, 0 %, 0 %]
JPEG_ACTIVITY: [0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0
%, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %, 0 %,
0 %, 0 %, 0 %, 0 %]
Added AMDSMI Tool Version.
AMD SMI will report three versions: AMDSMI Tool, AMDSMI Library version, and ROCm version. The AMDSMI Tool version is the CLI/tool version number with commit ID appended after+sign. The AMDSMI Library version is the library package version number. The ROCm version is the system’s installed ROCm version, if ROCm is not installed it will report N/A.
$ amd-smi version
AMDSMI Tool: 23.4.2+505b858 | AMDSMI Library version: 24.2.0.0 | ROCm version: 6.1.0
Added XGMI table.
Displays XGMI information for AMD GPU devices in a table format. Only available on supported ASICs (eg. MI300). Here users can view read/write data XGMI or PCIe accumulated data transfer size (in KiloBytes).
$ amd-smi xgmi
LINK METRIC TABLE:
bdf bit_rate max_bandwidth link_type 0000:0c:00.0 0000:22:00.0 0000:38:00.0 0000:5c:00.0 0000:9f:00.0 0000:af:00.0 0000:bf:00.0 0000:df:00.0
GPU0 0000:0c:00.0 32 Gb/s 512 Gb/s XGMI
Read N/A 2 KB 2 KB 1 KB 2 KB 1 KB 2 KB 2 KB
Write N/A 1 KB 1 KB 1 KB 1 KB 1 KB 1 KB 1 KB
GPU1 0000:22:00.0 32 Gb/s 512 Gb/s XGMI
Read 0 KB N/A 2 KB 2 KB 1 KB 2 KB 1 KB 2 KB
Write 0 KB N/A 1 KB 1 KB 1 KB 1 KB 1 KB 1 KB
GPU2 0000:38:00.0 32 Gb/s 512 Gb/s XGMI
Read 0 KB 1 KB N/A 2 KB 1 KB 2 KB 0 KB 0 KB
Write 0 KB 1 KB N/A 1 KB 1 KB 1 KB 1 KB 1 KB
GPU3 0000:5c:00.0 32 Gb/s 512 Gb/s XGMI
Read 0 KB 0 KB 2 KB N/A 1 KB 0 KB 0 KB 2 KB
Write 0 KB 1 KB 1 KB N/A 1 KB 1 KB 1 KB 1 KB
GPU4 0000:9f:00.0 32 Gb/s 512 Gb/s XGMI
Read 0 KB 1 KB 0 KB 0 KB N/A 2 KB 0 KB 2 KB
Write 0 KB 1 KB 1 KB 1 KB N/A 1 KB 1 KB 1 KB
GPU5 0000:af:00.0 32 Gb/s 512 Gb/s XGMI
Read 0 KB 2 KB 0 KB 0 KB 0 KB N/A 2 KB 0 KB
Write 0 KB 1 KB 1 KB 1 KB 1 KB N/A 1 KB 1 KB
GPU6 0000:bf:00.0 32 Gb/s 512 Gb/s XGMI
Read 0 KB 0 KB 0 KB 0 KB 0 KB 0 KB N/A 0 KB
Write 0 KB 1 KB 1 KB 1 KB 1 KB 1 KB N/A 1 KB
GPU7 0000:df:00.0 32 Gb/s 512 Gb/s XGMI
Read 0 KB 0 KB 0 KB 0 KB 0 KB 0 KB 0 KB N/A
Write 0 KB 1 KB 1 KB 1 KB 1 KB 1 KB 1 KB N/A
Added units of measure to JSON output.
We added unit of measure to JSON/CSVamd-smi metric,amd-smi static, andamd-smi monitorcommands.
Ex.
amd-smi metric -p --json
[
{
"gpu": 0,
"power": {
"socket_power": {
"value": 10,
"unit": "W"
},
"gfx_voltage": {
"value": 6,
"unit": "mV"
},
"soc_voltage": {
"value": 918,
"unit": "mV"
},
"mem_voltage": {
"value": 1250,
"unit": "mV"
},
"power_management": "ENABLED",
"throttle_status": "UNTHROTTLED"
}
}
]
Changed#
Topology is now left-aligned with BDF of each device listed individual table’s row/coloumns.
We provided each device’s BDF for every table’s row/columns, then left aligned data. We want AMD SMI Tool output to be easy to understand and digest for our users. Having users scroll up to find this information made it difficult to follow, especially for devices which have many devices associated with one ASIC.
$ amd-smi topology
ACCESS TABLE:
0000:0c:00.0 0000:22:00.0 0000:38:00.0 0000:5c:00.0 0000:9f:00.0 0000:af:00.0 0000:bf:00.0 0000:df:00.0
0000:0c:00.0 ENABLED ENABLED ENABLED ENABLED ENABLED ENABLED ENABLED ENABLED
0000:22:00.0 ENABLED ENABLED ENABLED ENABLED ENABLED ENABLED ENABLED ENABLED
0000:38:00.0 ENABLED ENABLED ENABLED ENABLED ENABLED ENABLED ENABLED ENABLED
0000:5c:00.0 ENABLED ENABLED ENABLED ENABLED ENABLED ENABLED ENABLED ENABLED
0000:9f:00.0 ENABLED ENABLED ENABLED ENABLED ENABLED ENABLED ENABLED ENABLED
0000:af:00.0 ENABLED ENABLED ENABLED ENABLED ENABLED ENABLED ENABLED ENABLED
0000:bf:00.0 ENABLED ENABLED ENABLED ENABLED ENABLED ENABLED ENABLED ENABLED
0000:df:00.0 ENABLED ENABLED ENABLED ENABLED ENABLED ENABLED ENABLED ENABLED
WEIGHT TABLE:
0000:0c:00.0 0000:22:00.0 0000:38:00.0 0000:5c:00.0 0000:9f:00.0 0000:af:00.0 0000:bf:00.0 0000:df:00.0
0000:0c:00.0 0 15 15 15 15 15 15 15
0000:22:00.0 15 0 15 15 15 15 15 15
0000:38:00.0 15 15 0 15 15 15 15 15
0000:5c:00.0 15 15 15 0 15 15 15 15
0000:9f:00.0 15 15 15 15 0 15 15 15
0000:af:00.0 15 15 15 15 15 0 15 15
0000:bf:00.0 15 15 15 15 15 15 0 15
0000:df:00.0 15 15 15 15 15 15 15 0
HOPS TABLE:
0000:0c:00.0 0000:22:00.0 0000:38:00.0 0000:5c:00.0 0000:9f:00.0 0000:af:00.0 0000:bf:00.0 0000:df:00.0
0000:0c:00.0 0 1 1 1 1 1 1 1
0000:22:00.0 1 0 1 1 1 1 1 1
0000:38:00.0 1 1 0 1 1 1 1 1
0000:5c:00.0 1 1 1 0 1 1 1 1
0000:9f:00.0 1 1 1 1 0 1 1 1
0000:af:00.0 1 1 1 1 1 0 1 1
0000:bf:00.0 1 1 1 1 1 1 0 1
0000:df:00.0 1 1 1 1 1 1 1 0
LINK TYPE TABLE:
0000:0c:00.0 0000:22:00.0 0000:38:00.0 0000:5c:00.0 0000:9f:00.0 0000:af:00.0 0000:bf:00.0 0000:df:00.0
0000:0c:00.0 SELF XGMI XGMI XGMI XGMI XGMI XGMI XGMI
0000:22:00.0 XGMI SELF XGMI XGMI XGMI XGMI XGMI XGMI
0000:38:00.0 XGMI XGMI SELF XGMI XGMI XGMI XGMI XGMI
0000:5c:00.0 XGMI XGMI XGMI SELF XGMI XGMI XGMI XGMI
0000:9f:00.0 XGMI XGMI XGMI XGMI SELF XGMI XGMI XGMI
0000:af:00.0 XGMI XGMI XGMI XGMI XGMI SELF XGMI XGMI
0000:bf:00.0 XGMI XGMI XGMI XGMI XGMI XGMI SELF XGMI
0000:df:00.0 XGMI XGMI XGMI XGMI XGMI XGMI XGMI SELF
NUMA BW TABLE:
0000:0c:00.0 0000:22:00.0 0000:38:00.0 0000:5c:00.0 0000:9f:00.0 0000:af:00.0 0000:bf:00.0 0000:df:00.0
0000:0c:00.0 N/A 50000-50000 50000-50000 50000-50000 50000-50000 50000-50000 50000-50000 50000-50000
0000:22:00.0 50000-50000 N/A 50000-50000 50000-50000 50000-50000 50000-50000 50000-50000 50000-50000
0000:38:00.0 50000-50000 50000-50000 N/A 50000-50000 50000-50000 50000-50000 50000-50000 50000-50000
0000:5c:00.0 50000-50000 50000-50000 50000-50000 N/A 50000-50000 50000-50000 50000-50000 50000-50000
0000:9f:00.0 50000-50000 50000-50000 50000-50000 50000-50000 N/A 50000-50000 50000-50000 50000-50000
0000:af:00.0 50000-50000 50000-50000 50000-50000 50000-50000 50000-50000 N/A 50000-50000 50000-50000
0000:bf:00.0 50000-50000 50000-50000 50000-50000 50000-50000 50000-50000 50000-50000 N/A 50000-50000
0000:df:00.0 50000-50000 50000-50000 50000-50000 50000-50000 50000-50000 50000-50000 50000-50000 N/A
Resolved issues#
Fix for Navi3X/Navi2X/MI100
amdsmi_get_gpu_pci_bandwidth()in frequencies_read tests.
Devices which do not report (eg. Navi3X/Navi2X/MI100) we have added checks to confirm these devices return AMDSMI_STATUS_NOT_SUPPORTED. Otherwise, tests now display a return string.Fix for devices which have an older pyyaml installed.
Platforms which are identified as having an older pyyaml version or pip, we no manually update both pip and pyyaml as needed. This corrects issues identified below. Fix impacts the following CLI commands:amd-smi listamd-smi staticamd-smi firmwareamd-smi metricamd-smi topology
TypeError: dump_all() got an unexpected keyword argument 'sort_keys'
Fix for crash when user is not a member of video/render groups.
AMD SMI now uses same mutex handler for devices as rocm-smi. This helps avoid crashes when DRM/device data is inaccessable to the logged in user.
amd_smi_lib for ROCm 6.0.0#
Added#
Integrated the E-SMI (EPYC-SMI) library.
You can now query CPU-related information directly through AMD SMI. Metrics include power, energy, performance, and other system details.Added support for gfx942 metrics.
You can now query MI300 device metrics to get real-time information. Metrics include power, temperature, energy, and performance.Compute and memory partition support.
Users can now view, set, and reset partitions. The topology display can provide a more in-depth look at the device’s current configuration.
Changed#
GPU index sorting made consistent with other tools.
To ensure alignment with other ROCm software tools, GPU index sorting is optimized to use Bus:Device.Function (BDF) rather than the card number.Topology output is now aligned with GPU BDF table.
Earlier versions of the topology output were difficult to read since each GPU was displayed linearly. Now the information is displayed as a table by each GPU’s BDF, which closer resembles rocm-smi output.
Optimized#
Updated to C++17, gtest-1.14, and cmake 3.14
Resolved issues#
Fix for driver not initialized.
If driver module is not loaded, user retrieve error reponse indicating amdgpu module is not loaded.