---
myst:
html_meta:
"description lang=en": "Learn how to use the AMD SMI command line tool."
"keywords": "api, smi, lib, system, management, interface, example"
---
# AMD SMI CLI tool usage
This tool is a command line interface (CLI) for manipulating and monitoring the
`amdgpu` kernel; it is intended to replace and deprecate the existing `rocm_smi`
CLI tool and `gpuv-smi` tool. The AMD SMI CLI tool uses Ctypes to call the
`amd_smi_lib` API.
When using the CLI tool, you should have at least one AMD GPU and the driver
installed.
```{admonition} Disclaimer
The AMD SMI CLI tool is provided as an example code to aid the development of
telemetry tools. The [Python](./amdsmi-py-lib) or [C++
library](./amdsmi-cpp-lib) is recommended as a robust data source.
```
## Install the CLI Tool and Python library
Refer to the [installation instructions](../install/install.md).
## Get started
The `amd-smi` command provides system management and monitoring capabilities for
AMD hardware. When run without arguments, it displays the
[default summary view](#cli-ex-default) of all GPUs including version
information, GPU status, and running processes.
When run with `--help`, it reports the available subcommands:
```shell-session
~$ amd-smi --help
usage: amd-smi [-h] [--rocm-smi] ...
AMD System Management Interface | Version: 27.0.0 | ROCm version: 7.14.0 | Platform: Linux Baremetal
options:
-h, --help show this help message and exit
--rocm-smi Display GPU information in ROCm-SMI compatible format
AMD-SMI Commands:
Descriptions:
version Display version information
list List GPU information
static Gets static information about the specified GPU
firmware (ucode) Gets firmware information about the specified GPU
bad-pages Gets bad page information about the specified GPU
metric Gets metric/performance information about the specified GPU
process Lists compute process information running on the specified GPU
event Displays event information for the given GPU
topology Displays topology information of the devices
set Set options for devices
reset Reset options for devices
monitor (dmon) Monitor metrics for target devices
xgmi Displays xgmi information of the devices
partition Displays partition information of the devices
ras Retrieve RAS (CPER) entries from the driver
```
Example commands:
```shell-session
amd-smi static --gpu 0
amd-smi metric
amd-smi process --gpu 0 1
amd-smi reset --gpureset --gpu all
amd-smi --rocm-smi
```
```{note}
For command-specific help, use `amd-smi [command] --help` for see more detailed
usage information. See [Commands](#cmds).
For more detailed version information, use `amd-smi version`.
To display GPU information in the legacy ROCm-SMI format, use `amd-smi --rocm-smi`.
See [ROCm-SMI compatibility mode](#cli-ex-rocm-smi).
```
Environment variables:
You can set one or more variables in front of any `amd-smi` invocation. For example:
```shell-session
AMDSMI_GPU_METRICS_CACHE_MS=200 amd-smi metric
```
Current Variables:
```{note}
AMDSMI_GPU_METRICS_CACHE_MS - Controls the internal GPU metrics cache duration (ms). Default 100, set to 0 to disable.
AMDSMI_ASIC_INFO_CACHE_MS - Controls the internal GPU asic info cache duration (ms). Default 10000, set to 0 to disable.
```
(cmds)=
## Commands
The following are the help output for each command, providing quick reference
details for usage.
(cmd-list)=
### amd-smi list
Lists GPU information.
```{note}
`amd-smi list -e` is useful for mapping physical-to-logical GPU IDs.
The `oam_id` field identifies the physical board slot in multi-GPU OAM chassis.
The `ID` shown for `--gpu` (0, 1, 2, …) is an enumeration index assigned in discovery
order, not the device-type value returned by `amdsmi_get_gpu_id()`; `--gpu` also accepts a
BDF or UUID to select a specific card.
```
```shell-session
~$ amd-smi list --help
usage: amd-smi list [-h] [--json | --csv] [--file FILE] [--loglevel LEVEL]
[-g GPU [GPU ...] | -U CPU [CPU ...] | -O CORE [CORE ...]]
Lists all detected devices on the system.
Lists the BDF, UUID, KFD_ID, NODE_ID, and Partition ID for each GPU and/or CPUs.
In virtualization environments, it can also list VFs associated to each
GPU with some basic information for each VF.
List Arguments:
-h, --help show this help message and exit
-e, --enumeration Enumeration mapping to other features.
Includes CARD, RENDER, HSA_ID, HIP_ID, HIP_UUID, and OAM_ID.
Device Arguments:
-g, --gpu GPU [GPU ...] Select a GPU ID, BDF, or UUID from the possible choices:
ID: 0 | BDF: 0000:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 1 | BDF: 0001:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 2 | BDF: 0002:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 3 | BDF: 0003:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
all | Selects all devices
-U, --cpu CPU [CPU ...] Select a CPU ID from the possible choices:
ID: 0
ID: 1
ID: 2
ID: 3
all | Selects all devices
-O, --core CORE [CORE ...] Select a Core ID from the possible choices:
ID: 0 - 95
all | Selects all devices
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
```
(cmd-static)=
### amd-smi static
Gets static information about the specified GPU. See the [sample
output](#cli-ex-static) for `amd-smi static`.
```shell-session
~$ amd-smi static --help
usage: amd-smi static [-h] [-g GPU [GPU ...] | -U CPU [CPU ...]] [-a] [-b] [-V] [-d] [-v]
[-c] [-B] [-R] [-r] [-p] [-l] [-P] [-x] [-u] [-s] [-i]
[--json | --csv] [--file FILE] [--loglevel LEVEL]
If no GPU is specified, returns static information for all GPUs on the system.
If no static argument is provided, all static information will be displayed.
Static Arguments:
-h, --help show this help message and exit
-a, --asic All asic information
-b, --bus All bus information
-I, --ifwi All video bios\IFWI information (if available)
-d, --driver Displays driver version
-v, --vram All vram information
-c, --cache All cache information
-B, --board All board information
-R, --process-isolation The process isolation status
-r, --ras Displays RAS features information;
Sudo may be required for some features
-C, --clock [CLOCK ...] Show one or more valid clock frequency levels. Available options:
SYS, DF, DCEF, SOC, MEM, VCLK0, VCLK1, DCLK0, DCLK1, ALL
-p, --partition Partition information
-l, --limit All limit metric values (i.e. power and thermal limits)
-P, --soc-pstate The available soc pstate policy
-x, --xgmi-plpd The available XGMI per-link power down policy
-u, --numa All numa node information
CPU Arguments:
-s, --smu All SMU FW information
-i, --interface-ver Displays hsmp interface version
Device Arguments:
-g, --gpu GPU [GPU ...] Select a GPU ID, BDF, or UUID from the possible choices:
ID: 0 | BDF: 0000:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 1 | BDF: 0001:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 2 | BDF: 0002:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 3 | BDF: 0003:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
all | Selects all devices
-U, --cpu CPU [CPU ...] Select a CPU ID from the possible choices:
ID: 0
ID: 1
ID: 2
ID: 3
all | Selects all devices
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
```
(cmd-firmware)=
### amd-smi firmware
Gets firmware information about the specified GPU.
```shell-session
~$ amd-smi firmware --help
usage: amd-smi firmware [-h] [--json | --csv] [--file FILE] [--loglevel LEVEL]
[-g GPU [GPU ...] | -U CPU [CPU ...] | -O CORE [CORE ...]] [-f]
If no GPU is specified, return firmware information for all GPUs on the system.
Firmware Arguments:
-h, --help show this help message and exit
-f, --ucode-list, --fw-list All FW list information
Device Arguments:
-g, --gpu GPU [GPU ...] Select a GPU ID, BDF, or UUID from the possible choices:
ID: 0 | BDF: 0000:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 1 | BDF: 0001:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 2 | BDF: 0002:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 3 | BDF: 0003:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
all | Selects all devices
-U, --cpu CPU [CPU ...] Select a CPU ID from the possible choices:
ID: 0
ID: 1
ID: 2
ID: 3
all | Selects all devices
-O, --core CORE [CORE ...] Select a Core ID from the possible choices:
ID: 0 - 95
all | Selects all devices
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
```
(cmd-bad-pages)=
### amd-smi bad-pages
Gets bad page information about the specified GPU.
```shell-session
~$ amd-smi bad-pages --help
usage: amd-smi bad-pages [-h] [--json | --csv] [--file FILE] [--loglevel LEVEL]
[-g GPU [GPU ...] | -U CPU [CPU ...] | -O CORE [CORE ...]] [-p]
[-r] [-u]
If no GPU is specified, return bad page information for all GPUs on the system.
Bad Pages Arguments:
-h, --help show this help message and exit
-p, --pending Displays all pending retired pages
-r, --retired Displays retired pages
-u, --un-res Displays unreservable pages
Device Arguments:
-g, --gpu GPU [GPU ...] Select a GPU ID, BDF, or UUID from the possible choices:
ID: 0 | BDF: 0000:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 1 | BDF: 0001:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 2 | BDF: 0002:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 3 | BDF: 0003:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
all | Selects all devices
-U, --cpu CPU [CPU ...] Select a CPU ID from the possible choices:
ID: 0
ID: 1
ID: 2
ID: 3
all | Selects all devices
-O, --core CORE [CORE ...] Select a Core ID from the possible choices:
ID: 0 - 95
all | Selects all devices
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
```
(cmd-metric)=
### amd-smi metric
Gets metrics and performance information about the specified GPU.
```shell-session
~$ amd-smi metric --help
usage: amd-smi metric [-h] [-g GPU [GPU ...] | -U CPU [CPU ...] | -O CORE [CORE ...]]
[-w INTERVAL] [-W TIME] [-i ITERATIONS] [-m] [-u] [-p] [-c] [-t]
[-P] [-e] [-k] [-f] [-C] [-o] [-l] [-x] [-E] [-X] [--cpu-power-metrics]
[--cpu-prochot] [--cpu-freq-metrics] [--cpu-c0-res]
[--cpu-lclk-dpm-level NBIOID] [--cpu-pwr-svi-telemetry-rails]
[--cpu-io-bandwidth IO_BW LINKID_NAME]
[--cpu-xgmi-bandwidth XGMI_BW LINKID_NAME] [--cpu-metrics-ver]
[--cpu-metrics-table] [--cpu-socket-energy] [--cpu-ddr-bandwidth]
[--cpu-temp] [--cpu-dimm-temp-range-rate DIMM_ADDR]
[--cpu-dimm-pow-consumption DIMM_ADDR]
[--cpu-dimm-thermal-sensor DIMM_ADDR] [--core-boost-limit]
[--core-curr-active-freq-core-limit] [--core-energy]
[--json | --csv] [--file FILE] [--loglevel LEVEL]
If no GPU is specified, returns metric information for all GPUs on the system.
If no metric argument is provided, all metric information will be displayed.
Metric arguments:
-h, --help show this help message and exit
-m, --mem-usage Memory usage per block
-u, --usage Displays engine usage information
-p, --power Current power usage
-c, --clock Average, max, and current clock frequencies
-t, --temperature Current temperatures
-P, --pcie Current PCIe speed, width, and replay count
-e, --ecc Total number of ECC errors
-k, --ecc-blocks Number of ECC errors per block
-V, --voltage GPU voltage
-f, --fan Current fan speed
-C, --voltage-curve Display voltage curve
-o, --overdrive Current GFX and MEM clock overdrive level
-l, --perf-level Current DPM performance level
-x, --xgmi-err XGMI error information since last read
-E, --energy Amount of energy consumed
-v, --violation Displays throttle accumulators;
Only available for MI300 or newer ASICs
-X, --partition Switch temperature, clock, and usage to partition-scoped
(XCP/AID/MID) data sources; combine with those flags to scope it;
Only available for MI300 or newer ASICs
Watch Arguments:
-w, --watch INTERVAL Reprint the command in a loop of INTERVAL seconds
-W, --watch_time TIME The total duration of TIME to watch the command
-i, --iterations ITERATIONS The total number of ITERATIONS to repeat the command
CPU Arguments:
--cpu-power-metrics CPU power metrics
--cpu-prochot Displays prochot status
--cpu-freq-metrics Displays currentFclkMemclk frequencies and cclk frequency limit
--cpu-c0-res Displays C0 residency
--cpu-lclk-dpm-level NBIOID Displays lclk dpm level range. Requires socket ID and NBOID as inputs
--cpu-pwr-svi-telemetry-rails Displays svi based telemetry for all rails
--cpu-io-bandwidth IO_BW LINKID_NAME Displays current IO bandwidth for the selected CPU.
input parameters are bandwidth type(1) and link ID encodings
i.e. P2, P3, G0 - G7
--cpu-xgmi-bandwidth XGMI_BW LINKID_NAME Displays current XGMI bandwidth for the selected CPU
input parameters are bandwidth type(1,2,4) and link ID encodings
i.e. P2, P3, G0 - G7
--cpu-pwr-eff-mode Displays current power efficiency mode.
For Family 1Ah Models 50h-57h onwards and MODE= 4 or 5, displays utilization percentage and PPT limit in Watts.
--cpu-metrics-ver Displays metrics table version
--cpu-metrics-table Displays metric table
--cpu-socket-energy Displays socket energy for the selected CPU socket
--cpu-ddr-bandwidth Displays per socket max ddr bw, current utilized bw,
and current utilized ddr bw in percentage
--cpu-temp Displays cpu socket temperature
--cpu-dimm-temp-range-rate DIMM_ADDR Displays dimm temperature range and refresh rate
--cpu-dimm-pow-consumption DIMM_ADDR Displays dimm power consumption
--cpu-dimm-thermal-sensor DIMM_ADDR Displays dimm thermal sensor
--cpu-xgmi-pstate-range Displays XGMI pstate range (min and max values) for the selected CPU
--cpu-railisofreq-policy Displays CPU rail isolated frequency policy
--cpu-dfcstate-ctrl Displays DFCState control status
--cpu-pc6-enable Displays PC6 enable control
--cpu-cc6-enable Displays CC6 enable control
--cpu-dimm-sb-reg Read DIMM sideband register.Requires DIMM_ADDR, LID(0x2->TS0,0x6->TS1,0x9->PMIC0,0xA->SPDHub),
REG_OFFSET (hex), REG_SPACE (REGSPACE:0->Volatile,1->NVM)
--cpu-tdelta Displays CPU thermal delta (TDELTA) value for the selected CPU socket
--cpu-svi3-vr-controller-temp TYPE [RAIL_INDEX ...]
Get SVI3 VR controller temperature. TYPE: 0=HottestRail, 1=IndividualRail.
If TYPE=1, RAIL_INDEX: (RAIL_INDEX:0->VDDCR_CPU0,1->VDDCR_CPU1,2->VDDCR_SOC,3->VDDIO,4->VDDIO_MEM_S3) must be specified
--cpu-enabled-commands Displays HSMP enabled commands bit masks (Read/Write EnabledCommandsBitMask0-2)
--cpu-sdps-limit Displays CPU SDPS limit for the selected CPU socket (in Watts)
CPU Core Arguments:
--core-boost-limit Get boost limit for the selected cores
--core-curr-active-freq-core-limit Get Current CCLK limit set per Core
--core-energy Displays core energy for the selected core
--core-ccd-power Displays CCD (Core Complex Die) power consumption for the selected core
--core-floor-limit Get floor limit frequency for the selected core (MHz)
--core-eff-floor-limit Get effective floor limit frequency for the selected core (MHz)
Device Arguments:
-g, --gpu GPU [GPU ...] Select a GPU ID, BDF, or UUID from the possible choices:
ID: 0 | BDF: 0000:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 1 | BDF: 0001:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 2 | BDF: 0002:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 3 | BDF: 0003:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
all | Selects all devices
-U, --cpu CPU [CPU ...] Select a CPU ID from the possible choices:
ID: 0
ID: 1
ID: 2
ID: 3
all | Selects all devices
-O, --core CORE [CORE ...] Select a Core ID from the possible choices:
ID: 0 - 95
all | Selects all devices
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
```
(cmd-process)=
### amd-smi process
Lists compute process information running on the specified GPU. See the [sample
output](#cli-ex-process) for `amd-smi process`.
```shell-session
~$ amd-smi process --help
usage: amd-smi process [-h] [--json | --csv] [--file FILE] [--loglevel LEVEL]
[-g GPU [GPU ...] | -U CPU [CPU ...] | -O CORE [CORE ...]]
[-w INTERVAL] [-W TIME] [-i ITERATIONS] [-G] [-e] [-p PID]
[-n NAME]
If no GPU is specified, returns information for all GPUs on the system.
If no process argument is provided, all process information will be displayed.
Process arguments:
Process arguments:
-h, --help show this help message and exit
-G, --general pid, process name, memory usage
-e, --engine All engine usages
-p, --pid PID Gets all process information about the specified process based on Process ID
-n, --name NAME Gets all process information about the specified process based on Process Name.
If multiple processes have the same name, information is returned for all of them.
Watch Arguments:
-w, --watch INTERVAL Reprint the command in a loop of INTERVAL seconds
-W, --watch_time TIME The total duration of TIME to watch the command
-i, --iterations ITERATIONS The total number of ITERATIONS to repeat the command
Device Arguments:
-g, --gpu GPU [GPU ...] Select a GPU ID, BDF, or UUID from the possible choices:
ID: 0 | BDF: 0000:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 1 | BDF: 0001:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 2 | BDF: 0002:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 3 | BDF: 0003:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
all | Selects all devices
-U, --cpu CPU [CPU ...] Select a CPU ID from the possible choices:
ID: 0
ID: 1
ID: 2
ID: 3
all | Selects all devices
-O, --core CORE [CORE ...] Select a Core ID from the possible choices:
ID: 0 - 95
all | Selects all devices
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
```
(cmd-event)=
### amd-smi event
Displays event information for the given GPU.
```shell-session
~$ amd-smi event --help
usage: amd-smi event [-h] [--json | --csv] [--file FILE] [--loglevel LEVEL]
[-g GPU [GPU ...] | -U CPU [CPU ...] | -O CORE [CORE ...]]
If no GPU is specified, returns event information for all GPUs on the system.
Event Arguments:
-h, --help show this help message and exit
Device Arguments:
-g, --gpu GPU [GPU ...] Select a GPU ID, BDF, or UUID from the possible choices:
ID: 0 | BDF: 0000:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 1 | BDF: 0001:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 2 | BDF: 0002:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 3 | BDF: 0003:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
all | Selects all devices
-U, --cpu CPU [CPU ...] Select a CPU ID from the possible choices:
ID: 0
ID: 1
ID: 2
ID: 3
all | Selects all devices
-O, --core CORE [CORE ...] Select a Core ID from the possible choices:
ID: 0 - 95
all | Selects all devices
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
```
(cmd-topology)=
### amd-smi topology
Displays topology information of the specified devices.
```shell-session
~$ amd-smi topology --help
usage: amd-smi topology [-h] [--json | --csv] [--file FILE] [--loglevel LEVEL]
[-g GPU [GPU ...] | -U CPU [CPU ...] | -O CORE [CORE ...]] [-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
-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 coherent (or non-coherent) 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
Device Arguments:
-g, --gpu GPU [GPU ...] Select a GPU ID, BDF, or UUID from the possible choices:
ID: 0 | BDF: 0000:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 1 | BDF: 0001:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 2 | BDF: 0002:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 3 | BDF: 0003:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
all | Selects all devices
-U, --cpu CPU [CPU ...] Select a CPU ID from the possible choices:
ID: 0
ID: 1
ID: 2
ID: 3
all | Selects all devices
-O, --core CORE [CORE ...] Select a Core ID from the possible choices:
ID: 0 - 95
all | Selects all devices
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
```
#### Interpreting hops and weight
The following descriptions apply to AMD Instinct GPUs up to and including MI355X.
**Hops (`-o, --hops`)** — The hops table reports an *abstracted topology step count*, not
the number of physical xGMI links between devices. The possible values are:
| Hops | Meaning |
|------|---------|
| 1 | The two GPUs are reachable over xGMI, regardless of the number of physical xGMI links on the route. |
| 2 | The two GPUs communicate over PCIe within the same CPU NUMA node. |
| 3 | The two GPUs communicate over PCIe across different CPU NUMA nodes. |
| 4 | Fallback when the inter-CPU io_link weight cannot be read. |
Two GPUs on the same xGMI fabric always report `1`, even when data physically crosses
multiple xGMI links. To determine the literal number of physical xGMI links between two
devices, read the value from the `amdgpu` driver:
```bash
cat /sys/class/drm/card*/device/xgmi_num_hops
```
**Weight (`-w, --weight`)** — The weight table reports a qualitative cost metric derived
from the KFD io_link `weight` property (lower = closer/faster), analogous to the NUMA
distances reported by `numactl`. Each physical xGMI hop contributes 15 to the weight
(for example, a single-hop xGMI connection has weight 15). PCIe routes are summed across
segments (GPU→CPU + CPU→CPU + CPU→GPU); each GPU-to-CPU segment is typically 20, while
the CPU-to-CPU segment uses the actual io_link weight, or a fallback of 10 if that weight
cannot be read.
(cmd-set)=
### amd-smi set
Set options for specified devices.
```shell-session
~$ amd-smi set --help
usage: amd-smi set [-h] (-g GPU [GPU ...] | -U CPU [CPU ...] | -O CORE [CORE ...]) [-f %]
[-l LEVEL] [-P SETPROFILE] [-d SCLKMAX] [-C PARTITION] [-M PARTITION]
[-a MODE] [-o WATTS] [-p POLICY_ID] [-x POLICY_ID] [-R STATUS]
[--cpu-pwr-limit PWR_LIMIT] [--cpu-xgmi-link-width MIN_WIDTH MAX_WIDTH]
[--cpu-lclk-dpm-level NBIOID MIN_DPM MAX_DPM] [--cpu-pwr-eff-mode MODE [UTIL PPT_LIMIT]]
[--cpu-gmi3-link-width MIN_LW MAX_LW] [--cpu-pcie-link-rate LINK_RATE]
[--cpu-df-pstate-range MAX_PSTATE MIN_PSTATE] [--cpu-enable-apb]
[--cpu-disable-apb DF_PSTATE] [--soc-boost-limit BOOST_LIMIT]
[--core-boost-limit BOOST_LIMIT] [--json | --csv] [--file FILE]
[--loglevel LEVEL] [--cpu-xgmi-pstate-range MIN_PSTATE MAX_PSTATE] [--cpu-railisofreq-policy VALUE]
[--cpu-dfcstate-ctrl VALUE] [--cpu-pc6-enable VALUE] [--cpu-cc6-enable VALUE]
[--cpu-floor-limit FLOOR_LIMIT] [--cpu-msr-floor-limit MSR_FLOOR_LIMIT]
[--core-floor-limit FLOOR_LIMIT] [--core-msr-floor-limit MSR_FLOOR_LIMIT]
[--cpu-dimm-sb-reg DIMM_ADDR LID REG_OFFSET REG_SPACE WRITE_DATA] [--cpu-sdps-limit SDPS_LIMIT]
If no GPU is specified, will select all GPUs on the system.
A set argument must be provided; Multiple set arguments are accepted.
Requires 'sudo' privileges.
Set Arguments:
-h, --help show this help message and exit
-f, --fan % Set GPU fan speed :
GPU 0: 0-255 or 0-100%
GPU 1: 20-100 or 0-100%
-l, --perf-level LEVEL Set one of the following performance levels:
AUTO, LOW, HIGH, MANUAL, STABLE_STD, STABLE_PEAK, STABLE_MIN_MCLK, STABLE_MIN_SCLK, DETERMINISM
-P, --profile PROFILE_LEVEL Set power profile level (#) or choose one of available profiles:
CUSTOM_MASK, VIDEO_MASK, POWER_SAVING_MASK, COMPUTE_MASK, VR_MASK, THREE_D_FULL_SCR_MASK, BOOTUP_DEFAULT
-d, --perf-determinism SCLKMAX Enable performance determinism mode and set GFXCLK softmax limit (in MHz)
-C, --compute-partition, --accelerator-partition TYPE/INDEX
Set one of the following the accelerator TYPE or profile INDEX:
N/A.
Use `sudo amd-smi partition --accelerator` to find acceptable values.
-M, --memory-partition PARTITION Set one of the following the memory partition modes:
NPS1, NPS2, NPS4, NPS8
-a, --compute-partition-mem-alloc-mode MODE Set compute partition memory allocation mode (requires sudo):
CAPPING - each XCP is capped to an even share of partition memory
ALL - each XCP may use the full partition memory
-o, --power-cap WATTS Set power capacity limit:
min cap: 0 W, max cap: 550 W
-p, --soc-pstate POLICY_ID Set the GPU soc pstate policy using policy id, an integer. Valid id's include:
N/A
-x, --xgmi-plpd POLICY_ID Set the GPU XGMI per-link power down policy using policy id, an integer. Valid id's include:
N/A
-c, --clk-level CLK_TYPE [FREQ_LEVELS ...] Set one or more sclk (aka gfxclk), mclk, fclk, pcie, or socclk frequency levels.
Use `amd-smi static --clock` to find acceptable levels.
-L, --clk-limit CLK_TYPE LIM_TYPE VALUE Sets the sclk (aka gfxclk), mclk, or fclk minimum and maximum frequencies.
ex: amd-smi set -L (sclk | mclk | fclk) (min | max) value
-R, --process-isolation STATUS Enable or disable the GPU process isolation on a per partition basis: 0 for disable and 1 for enable.
--ptl-status STATUS Enable or disable the PTL on a GPU processor: 0 for disable and 1 for enable
--ptl-format FRMT1,FRMT2 Set the PTL format on a GPU processor. For example, --ptl-format I8,F32
CPU Arguments:
--cpu-pwr-limit PWR_LIMIT Set power limit for the given socket. Input parameter is power limit value.
--cpu-xgmi-link-width MIN_WIDTH MAX_WIDTH Set max and Min linkwidth. Input parameters are min and max link width values
--cpu-lclk-dpm-level NBIOID MIN_DPM MAX_DPM Sets the max and min dpm level on a given NBIO.
Input parameters are die_index, min dpm, max dpm.
--cpu-pwr-eff-mode MODE [UTIL PPT_LIMIT] [MODE [UTIL PPT_LIMIT] ...]
Sets the power efficiency mode policy. Input parameters,
MODE(0=HighPerformance, 1=PowerEfficiency, 2=IOPerformance, 3=BalancedMemory, 4=BalancedCore, 5=BalancedCoreMemory),
For Family 1Ah Models 50h-57h onwards, UTIL(%)(0-100) and PPT_limit (in mW) required if MODE= 4 or 5
--cpu-gmi3-link-width MIN_LW MAX_LW Sets max and min gmi3 link width range
--cpu-pcie-link-rate LINK_RATE Sets pcie link rate
--cpu-df-pstate-range MAX_PSTATE MIN_PSTATE Sets max and min df-pstates
--cpu-enable-apb Enables the DF p-state performance boost algorithm
--cpu-disable-apb DF_PSTATE Disables the DF p-state performance boost algorithm. Input parameter is DFPstate (0-3)
--soc-boost-limit BOOST_LIMIT Sets the boost limit for the given socket. Input parameter is socket BOOST_LIMIT value
--cpu-xgmi-pstate-range MIN_PSTATE MAX_PSTATE Sets min and max for xgmi pstate range (MAX <= MIN)
--cpu-railisofreq-policy VALUE Sets the CPU rail isolated frequency policy. Input parameter is VALUE (0-1)
--cpu-dfcstate-ctrl VALUE Sets the DFCState control for the given socket. Input parameter is VALUE (0-1)
--cpu-pc6-enable VALUE Sets PC6 enable control. Input parameter is value (0-1)
--cpu-cc6-enable VALUE Sets CC6 enable control. Input parameter is value (0-1)
--cpu-floor-limit FLOOR_LIMIT Sets the floor limit for the given CPU socket. Input parameter is CPU FLOOR_LIMIT value MHz
--cpu-msr-floor-limit MSR_FLOOR_LIMIT Sets the CPU MSR floor limit frequency for the given socket. Input parameter is MSR_FLOOR_LIMIT value in MHz
--cpu-dimm-sb-reg DIMM_ADDR LID REG_OFFSET REG_SPACE WRITE_DATA
Write data to DIMM sideband register. Requires DIMM_ADDR, LID(0x2->TS0,0x6->TS1,0x9->PMIC0,0xA->SPDHub)
REG_OFFSET (hex), REG_SPACE (REGSPACE:0->Volatile,1->NVM), WRITE_DATA (hex)
--cpu-sdps-limit SDPS_LIMIT Set CPU SDPS limit for the given socket. Input parameter is SDPS limit value in milliwatts (mW).
CPU Core Arguments:
--core-boost-limit BOOST_LIMIT Sets the boost limit for the given core. Input parameter is core BOOST_LIMIT value
Device Arguments:
-g, --gpu GPU [GPU ...] Select a GPU ID, BDF, or UUID from the possible choices:
ID: 0 | BDF: 0000:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 1 | BDF: 0001:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 2 | BDF: 0002:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 3 | BDF: 0003:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
all | Selects all devices
-U, --cpu CPU [CPU ...] Select a CPU ID from the possible choices:
ID: 0
ID: 1
ID: 2
ID: 3
all | Selects all devices
-O, --core CORE [CORE ...] Select a Core ID from the possible choices:
ID: 0 - 95
all | Selects all devices
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
```
(cmd-reset)=
### amd-smi reset
Reset options for specified devices.
```{warning}
* On systems with XGMI/Infinity Fabric (for example, AMD Instinct MI Series), resetting one
GPU resets all GPUs in the same XGMI hive. Use `amd-smi xgmi` or `amd-smi topology` to find the XGMI link connected GPUs or check `/sys/class/drm/card*/device/xgmi_info/xgmi_hive_id` to identify GPUs having the same hive id, before issuing a reset.
* On systems where GPUs share an upstream PCIe switch, resetting one GPU may trigger a
Secondary Bus Reset (SBR) on the upstream switch port, affecting all GPUs behind the same
switch, independent of XGMI hive membership.
* Any process with an open `/dev/kfd` handle will be terminated when a GPU reset occurs,
even if that process is not using the GPU being reset. GPU isolation techniques using the
environment variables `ROCR_VISIBLE_DEVICES` and `HIP_VISIBLE_DEVICES` do not
prevent this.
See [GPU reset behavior on XGMI systems](/conceptual/gpu-reset-behavior.md) for
more information.
```
```shell-session
~$ amd-smi reset --help
usage: amd-smi reset [-h] [--json | --csv] [--file FILE] [--loglevel LEVEL]
(-g GPU [GPU ...] | -U CPU [CPU ...] | -O CORE [CORE ...]) [-G] [-c]
[-f] [-p] [-x] [-d] [-C] [-M] [-o] [-l]
If no GPU is specified, will select all GPUs on the system.
A reset argument must be provided; Multiple reset arguments are accepted.
Requires 'sudo' privileges.
Reset Arguments:
-h, --help show this help message and exit
-G, --gpureset Reset the specified GPU
-c, --clocks Reset clocks and overdrive to default
-f, --fans Reset fans to automatic (driver) control
-p, --profile Reset power profile back to default
-x, --xgmierr Reset XGMI error counts
-d, --perf-determinism Disable performance determinism
-o, --power-cap Reset power capacity limit to max capable
-l, --clean-local-data Clean up local data in LDS/GPRs on a per partition basis
Device Arguments:
-g, --gpu GPU [GPU ...] Select a GPU ID, BDF, or UUID from the possible choices:
ID: 0 | BDF: 0000:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 1 | BDF: 0001:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 2 | BDF: 0002:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 3 | BDF: 0003:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
all | Selects all devices
-U, --cpu CPU [CPU ...] Select a CPU ID from the possible choices:
ID: 0
ID: 1
ID: 2
ID: 3
all | Selects all devices
-O, --core CORE [CORE ...] Select a Core ID from the possible choices:
ID: 0 - 95
all | Selects all devices
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
```
(cmd-monitor)=
### amd-smi monitor
Monitor metrics for target devices. See the [sample output](#cli-ex-monitor)
for `amd-smi monitor`.
```shell-session
~$ amd-smi monitor --help
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] [-e] [-v] [-r] [-q]
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
-p, --power-usage Monitor power usage and power cap 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)
-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 bandwidth in Mb/s
-q, --process Enable Process information table below monitor output
-V, --violation Monitor power and thermal violation status (%);
Only available for MI300 or newer ASICs
Watch Arguments:
-w, --watch INTERVAL Reprint the command in a loop of INTERVAL seconds
-W, --watch_time TIME The total duration of TIME to watch the command
-i, --iterations ITERATIONS The total number of ITERATIONS to repeat the command
Device Arguments:
-g, --gpu GPU [GPU ...] Select a GPU ID, BDF, or UUID from the possible choices:
ID: 0 | BDF: 0000:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 1 | BDF: 0001:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 2 | BDF: 0002:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 3 | BDF: 0003:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
all | Selects all devices
-U, --cpu CPU [CPU ...] Select a CPU ID from the possible choices:
ID: 0
ID: 1
ID: 2
ID: 3
all | Selects all devices
-O, --core CORE [CORE ...] Select a Core ID from the possible choices:
ID: 0 - 95
all | Selects all devices
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
```
(cmd-xgmi)=
### amd-smi xgmi
Displays XGMI information of specified devices.
```shell-session
~$ amd-smi xgmi --help
usage: amd-smi xgmi [-h] [--json | --csv] [--file FILE] [--loglevel LEVEL]
[-g GPU [GPU ...] | -U CPU [CPU ...] | -O CORE [CORE ...]] [-m]
If no GPU is specified, returns information for all GPUs on the system.
If no xgmi argument is provided, all xgmi information will be displayed.
XGMI arguments:
-h, --help show this help message and exit
-m, --metric Metric XGMI information
-l, --link-status XGMI Link Status information
Device Arguments:
-g, --gpu GPU [GPU ...] Select a GPU ID, BDF, or UUID from the possible choices:
ID: 0 | BDF: 0000:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 1 | BDF: 0001:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 2 | BDF: 0002:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 3 | BDF: 0003:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
all | Selects all devices
-U, --cpu CPU [CPU ...] Select a CPU ID from the possible choices:
ID: 0
ID: 1
ID: 2
ID: 3
all | Selects all devices
-O, --core CORE [CORE ...] Select a Core ID from the possible choices:
ID: 0 - 95
all | Selects all devices
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
```
(cmd-partition)=
### amd-smi partition
Displays partition information of the devices.
```shell-session
~$ amd-smi partition --help
usage: amd-smi partition [-h] [-g GPU [GPU ...]] [-c] [-m] [-a] [--json | --csv]
[--file FILE] [--loglevel LEVEL]
If no GPU is specified, returns information for all GPUs on the system.
If no partition argument is provided, all partition information will be displayed.
Partition arguments:
-h, --help show this help message and exit
-c, --current display the current partition information
-m, --memory display the current memory partition mode and capabilities
-a, --accelerator display accelerator partition information
Device Arguments:
-g, --gpu GPU [GPU ...] Select a GPU ID, BDF, or UUID from the possible choices:
ID: 0 | BDF: 0000:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 1 | BDF: 0001:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 2 | BDF: 0002:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 3 | BDF: 0003:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
all | Selects all devices
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:
```
### amd-smi ras
Displays RAS information of specified devices.
```shell-session
~$ amd-smi ras --help
usage: amd-smi ras [-h] (--cper | --afid) [--severity SEVERITY [SEVERITY ...]]
[--folder FOLDER] [--file-limit FILE_LIMIT] [--follow]
[--cper-file CPER_FILE] [-g GPU [GPU ...]] [--json | --csv]
[--file FILE] [--overwrite] [--append] [--loglevel LEVEL]
Retrieve and decode RAS (CPER) entries from the kernel driver.
Supports filtering by severity, exporting to different formats, and continuous monitoring.
This command accepts options only; no positional arguments are required.
RAS arguments:
-h, --help show this help message and exit
--cper Trigger current CPER data retrieval
--afid Generate an AFID (AMD Field ID) given a CPER record file or folder
CPER Arguments:
--severity SEVERITY [SEVERITY ...] Set the SEVERITY filters from the following:
nonfatal-uncorrected, fatal, nonfatal-corrected, all
--folder FOLDER With --cper: folder to dump current CPER report files (created if missing).
With --afid: existing folder of CPER records to decode.
--file-limit FILE_LIMIT Maximum number of current CPER files in target folder
Older files beyond limit will be deleted
--follow Continuously monitor for new CPER entries
AFID Arguments:
--cper-file CPER_FILE Full path of a retrieved CPER record file to generate the AFID
Device Arguments:
-g, --gpu GPU [GPU ...] Select a GPU ID, BDF, or UUID from the possible choices:
ID: 0 | BDF: 0000:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 1 | BDF: 0001:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 2 | BDF: 0002:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 3 | BDF: 0003:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
all | Selects all devices
-U, --cpu CPU [CPU ...] Select a CPU ID from the possible choices:
ID: 0
ID: 1
ID: 2
ID: 3
all | Selects all devices
-O, --core CORE [CORE ...] Select a Core ID from the possible choices:
ID: 0 - 95
all | Selects all devices
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 fabric
Displays fabric (UALoE/UALink over Ethernet) information of the devices.
```{note}
The `fabric` subcommand is registered only when the amdgpu driver is initialized.
On systems without IFoE/UALoE fabric hardware the fabric queries report `N/A` /
not supported.
```
```shell-session
~$ amd-smi fabric --help
usage: amd-smi fabric [-h] [-t] [-i] [-g GPU [GPU ...]] [--json | --csv]
[--file FILE] [--loglevel LEVEL]
If no GPU is specified, returns information for all GPUs on the system.
If no fabric argument is provided, all fabric information will be displayed.
Fabric arguments:
-h, --help show this help message and exit
-t, --topology Display fabric topology data (counters per category, instance, and item)
-i, --info Display fabric device configuration (BDF, bandwidth, latency, vPoD/pPoD, accelerator state)
Device Arguments:
-g, --gpu GPU [GPU ...] Select a GPU ID, BDF, or UUID from the possible choices:
ID: 0 | BDF: 0000:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 1 | BDF: 0001:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 2 | BDF: 0002:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
ID: 3 | BDF: 0003:01:00.0 | UUID: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX
all | Selects all devices
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
```
## Interpreting the output
When you run an `amd-smi` command, the tool presents detailed information
across various categories, each containing specific fields and their current
values.
(cli-ex-default)=
### Example output from amd-smi (default)
The following block is example output from running `amd-smi` without any
subcommand. This is the default view that displays a summary of version
information, GPU status, and running processes.
```bash
~$ amd-smi
+------------------------------------------------------------------------------+
| AMD-SMI 27.0.0 |
| amdgpu Version: 6.19.4 |
| ROCm Version: 7.14.0 |
| Platform: Linux Baremetal |
|-------------------------------------+----------------------------------------|
| BDF GPU-Name | Mem-Uti Temp UEC Power-Usage |
| GPU HIP-ID OAM-ID Partition-Mode | GFX-Uti Fan Mem-Usage |
|=====================================+========================================|
| 0000:01:00.0 AMD Instinct MI300A | 0 % 27 °C 0 49 W |
| 0 0 0 SPX/NPS1 | 6 % N/A 25/385728 MB |
|-------------------------------------+----------------------------------------|
| 0001:01:00.0 AMD Instinct MI300A | 0 % 29 °C 0 42 W |
| 1 1 1 SPX/NPS1 | 2 % N/A 25/385728 MB |
|-------------------------------------+----------------------------------------|
| 0002:01:00.0 AMD Instinct MI300A | 0 % 30 °C 0 33 W |
| 2 2 2 SPX/NPS1 | 2 % N/A 25/385728 MB |
|-------------------------------------+----------------------------------------|
| 0003:01:00.0 AMD Instinct MI300A | 0 % 26 °C 0 40 W |
| 3 3 3 SPX/NPS1 | 2 % N/A 25/385728 MB |
+-------------------------------------+----------------------------------------+
+------------------------------------------------------------------------------+
| Processes: |
| GPU PID Process Name GTT_MEM VRAM_MEM MEM_USAGE CU % SDMA |
|==============================================================================|
| No running processes found |
+------------------------------------------------------------------------------+
```
The default output includes the following sections:
- **Version header**: AMD-SMI version, amdgpu driver version, ROCm version,
and platform information.
- **GPU table** (first row per GPU): BDF address, GPU market name, OAM ID,
memory utilization, hotspot temperature, uncorrectable ECC error count, and
current/maximum power usage.
- **GPU table** (second row per GPU): GPU index, HIP ID, OAM ID,
partition mode (compute/memory), GFX utilization, fan speed, and VRAM usage.
- **Process table**: Lists all running GPU processes with GPU ID, PID, process
name, GTT memory, VRAM memory, total memory usage, compute unit occupancy
percentage, and SDMA usage.
```{note}
Process Name may require elevated permissions. If running without `sudo`,
process names may appear as `N/A`.
```
The default output also supports JSON and CSV formatting:
```shell-session
amd-smi --json
amd-smi --csv
```
(cli-ex-rocm-smi)=
### Example output from amd-smi --rocm-smi
The `--rocm-smi` flag provides a compatibility mode that displays GPU
information in a format similar to the legacy `rocm-smi` tool. This is useful
for users migrating from `rocm-smi` who rely on scripts or workflows that parse
the original concise output format.
```bash
~$ amd-smi --rocm-smi
========================================== ROCm System Management Interface ==========================================
==================================================== Concise Info ====================================================
Device Node IDs Temp Power Partitions SCLK MCLK Fan Perf PwrCap VRAM% GPU%
(DID, GUID) (Junction) (Avg) (Mem, Compute, ID)
======================================================================================================================
0 7 0x74a0, XXXXX 27.0°C 49.0W NPS1, SPX, 0 44Mhz 900Mhz 0% auto 550.0W 0% 2%
1 7 0x74a0, XXXXX 29.0°C 42.0W NPS1, SPX, 0 42Mhz 900Mhz 0% auto 550.0W 0% 1%
2 7 0x74a0, XXXXX 30.0°C 33.0W NPS1, SPX, 0 42Mhz 900Mhz 0% auto 550.0W 0% 1%
3 7 0x74a0, XXXXX 26.0°C 40.0W NPS1, SPX, 0 35Mhz 900Mhz 0% auto 550.0W 0% 1%
================================================ End of ROCm SMI Log =================================================
```
```{note}
The `--rocm-smi` flag is a top-level option (not a subcommand). It cannot be
combined with other subcommands. The temperature label (Edge, Junction, or
Memory) is automatically detected based on the first available sensor.
```
(cli-output-na)=
### About N/A values
`N/A` typically indicates that the data for a specific field is either unavailable or irrelevant in the current context for your device and its software environment. The exact reason may vary depending on the field and the GPU. In general, you can interpret `N/A` to mean one of the following:
**Not Applicable**: The feature or parameter does not apply to your specific AMD hardware or its current configuration. Examples include display-related clocks on a headless compute card and partition details when the GPU is not partitioned.
**Not Available**: The information cannot be retrieved by the `amd-smi` tool at this time. This could be due to one of the following reasons:
- The hardware component does not report the specific metric.
- The currently installed `amdgpu` driver version does not support querying
this particular piece of information through `amd-smi-lib`.
- The `amdgpu` driver reports a newer `gpu_metrics` version than the installed
AMD SMI supports. `gpu_metrics` is a versioned structure supplied by the
driver, and AMD SMI needs explicit support for each version's layout. When the
driver is newer than your AMD SMI (or ROCm) release by a release cycle, AMD SMI
can't parse the newer layout, so fields sourced from `gpu_metrics` (violations,
`SOCKET_POWER`, engine usage, and so on) read N/A. Upgrade AMD SMI or ROCm to a
release that supports your driver's `gpu_metrics` version to resolve this. (ROCm
7.13 added support for the dynamic `gpu_metrics` layout introduced in v1.9,
which handles current and future versions, so releases from 7.13 onward are no
longer affected by this mismatch.)
(cli-ex-static)=
### Example output from amd-smi static
The following block is example output from the `amd-smi static` command without additional modifiers.
```bash
~$ amd-smi static
CPU: 0
SMU:
FW_VERSION: 85.103.0
INTERFACE_VERSION:
PROTO VERSION: 6
...
GPU: 0
ASIC:
MARKET_NAME: AMD Instinct MI300A
VENDOR_ID: 0x1002
VENDOR_NAME: Advanced Micro Devices, Inc. [AMD/ATI]
SUBVENDOR_ID: 0x1002
DEVICE_ID: 0x74a0
SUBSYSTEM_ID: 0x74a0
REV_ID: 0x00
ASIC_SERIAL: 0xXXXXXXXXXXXXXXXX
OAM_ID: 0
NUM_COMPUTE_UNITS: 228
TARGET_GRAPHICS_VERSION: gfx942
FLAGS: 17
BUS:
BDF: 0000:01:00.0
MAX_PCIE_WIDTH: 16
MAX_PCIE_SPEED: 32 GT/s
PCIE_LEVELS: N/A
PCIE_INTERFACE_VERSION: Gen 5
SLOT_TYPE: PCIE
IFWI:
NAME: N/A
BUILD_DATE: N/A
PART_NUMBER: N/A
VERSION: N/A
LIMIT:
PPT0:
MAX_POWER_LIMIT: 550 W
MIN_POWER_LIMIT: 0 W
SOCKET_POWER_LIMIT: 0 W
PPT1:
MAX_POWER_LIMIT: N/A
MIN_POWER_LIMIT: N/A
SOCKET_POWER_LIMIT: N/A
SLOWDOWN_EDGE_TEMPERATURE: N/A
SLOWDOWN_HOTSPOT_TEMPERATURE: 100 °C
SLOWDOWN_VRAM_TEMPERATURE: 105 °C
SHUTDOWN_EDGE_TEMPERATURE: N/A
SHUTDOWN_HOTSPOT_TEMPERATURE: 110 °C
SHUTDOWN_VRAM_TEMPERATURE: 115 °C
PTL_STATE: N/A
PTL_FORMAT: N/A
DRIVER:
NAME: amdgpu
VERSION: 6.19.4
OS_KERNEL_VERSION: 5.15.0-generic
BOARD:
MODEL_NUMBER: N/A
PRODUCT_SERIAL: N/A
FRU_ID: N/A
PRODUCT_NAME: Aqua Vanjaram [Instinct MI300A]
MANUFACTURER_NAME: Advanced Micro Devices, Inc. [AMD/ATI]
RAS:
EEPROM_VERSION: 0x30000
BAD_PAGE_THRESHOLD: N/A
BAD_PAGE_THRESHOLD_EXCEEDED: N/A
PARITY_SCHEMA: DISABLED
SINGLE_BIT_SCHEMA: DISABLED
DOUBLE_BIT_SCHEMA: DISABLED
POISON_SCHEMA: ENABLED
ECC_BLOCK_STATE:
UMC: DISABLED
SDMA: ENABLED
GFX: ENABLED
MMHUB: ENABLED
ATHUB: DISABLED
PCIE_BIF: DISABLED
HDP: DISABLED
XGMI_WAFL: DISABLED
DF: DISABLED
SMN: DISABLED
SEM: DISABLED
MP0: DISABLED
MP1: DISABLED
FUSE: DISABLED
MCA: DISABLED
VCN: DISABLED
JPEG: DISABLED
IH: DISABLED
MPIO: DISABLED
SOC_PSTATE: N/A
XGMI_PLPD: N/A
PROFILE: AMDSMI_STATUS_NOT_SUPPORTED - Feature not supported
PROCESS_ISOLATION: Disabled
NUMA:
NODE: 0
AFFINITY: 0
CPU_AFFINITY:
CPU_LIST_0:
BITMASK: 0000000000FFFFFF
CPU_CORES_AFFINITY: 0-23
CPU_LIST_1:
BITMASK: 00FFFFFF00000000
CPU_CORES_AFFINITY: 96-119
CPU_LIST_2:
BITMASK: 0000000000000000
CPU_CORES_AFFINITY: N/A
SOCKET_AFFINITY:
CPU_LIST_0:
BITMASK: 0000000000FFFFFF
CPU_CORES_AFFINITY: 0-23
CPU_LIST_1:
BITMASK: 00FFFFFF00000000
CPU_CORES_AFFINITY: 96-119
CPU_LIST_2:
BITMASK: 0000000000000000
CPU_CORES_AFFINITY: N/A
VRAM:
TYPE: HBM
VENDOR: UNKNOWN
SIZE: 96432 MB
BIT_WIDTH: 8192
MAX_BANDWIDTH: 5325 GB/s
CACHE_INFO:
CACHE_0:
CACHE_PROPERTIES: DATA_CACHE, SIMD_CACHE
CACHE_SIZE: 32 KB
CACHE_LEVEL: 1
MAX_NUM_CU_SHARED: 1
NUM_CACHE_INSTANCE: 228
CACHE_1:
CACHE_PROPERTIES: INST_CACHE, SIMD_CACHE
CACHE_SIZE: 64 KB
CACHE_LEVEL: 1
MAX_NUM_CU_SHARED: 2
NUM_CACHE_INSTANCE: 108
CACHE_2:
CACHE_PROPERTIES: INST_CACHE, SIMD_CACHE
CACHE_SIZE: 64 KB
CACHE_LEVEL: 1
MAX_NUM_CU_SHARED: 1
NUM_CACHE_INSTANCE: 12
CACHE_3:
CACHE_PROPERTIES: DATA_CACHE, SIMD_CACHE
CACHE_SIZE: 16 KB
CACHE_LEVEL: 1
MAX_NUM_CU_SHARED: 2
NUM_CACHE_INSTANCE: 108
CACHE_4:
CACHE_PROPERTIES: DATA_CACHE, SIMD_CACHE
CACHE_SIZE: 16 KB
CACHE_LEVEL: 1
MAX_NUM_CU_SHARED: 1
NUM_CACHE_INSTANCE: 12
CACHE_5:
CACHE_PROPERTIES: DATA_CACHE, SIMD_CACHE
CACHE_SIZE: 4096 KB
CACHE_LEVEL: 2
MAX_NUM_CU_SHARED: 228
NUM_CACHE_INSTANCE: 1
CACHE_6:
CACHE_PROPERTIES: DATA_CACHE, SIMD_CACHE
CACHE_SIZE: 262144 KB
CACHE_LEVEL: 3
MAX_NUM_CU_SHARED: 228
NUM_CACHE_INSTANCE: 1
MEM_CARVEOUT: N/A (UMA carveout is not supported on this ASIC/VBIOS)
CLOCK:
SYS:
CURRENT_LEVEL: 0
FREQUENCY_LEVELS:
LEVEL 0: 44 MHz
LEVEL 1: 500 MHz
LEVEL 2: 2100 MHz
MEM:
CURRENT_LEVEL: 0
FREQUENCY_LEVELS:
LEVEL 0: 900 MHz
LEVEL 1: 1100 MHz
LEVEL 2: 1200 MHz
LEVEL 3: 1300 MHz
DF:
CURRENT_LEVEL: 0
FREQUENCY_LEVELS:
LEVEL 0: 1200 MHz
LEVEL 1: 1600 MHz
LEVEL 2: 1900 MHz
LEVEL 3: 2000 MHz
SOC:
CURRENT_LEVEL: 0
FREQUENCY_LEVELS:
LEVEL 0: 27 MHz
LEVEL 1: 800 MHz
LEVEL 2: 1000 MHz
LEVEL 3: 1143 MHz
DCEF: N/A
VCLK0:
CURRENT_LEVEL: 0
FREQUENCY_LEVELS:
LEVEL 0: 0 MHz
LEVEL 1: 914 MHz
VCLK1: N/A
DCLK0:
CURRENT_LEVEL: 0
FREQUENCY_LEVELS:
LEVEL 0: 7 MHz
LEVEL 1: 711 MHz
DCLK1: N/A
...
```
(cli-ex-process)=
### Example output from amd-smi process
The following block is example output from the `amd-smi process` command without
additional modifiers. When no GPU is specified, it returns process information
for all GPUs on the system.
```bash
~$ amd-smi process
GPU: 0
PROCESS_INFO:
NAME: python3
PID: 12345
MEMORY_USAGE:
GTT_MEM: 128.0 MB
CPU_MEM: 0.0 B
VRAM_MEM: 4.0 GB
MEM: 4.13 GB
USAGE:
GFX: 52345678 ns
ENC: 0 ns
CU_OCCUPANCY: 114
EVICTED_TIME: 0 ms
GPU: 1
PROCESS_INFO:
No running processes detected
...
```
You can filter processes by GPU, PID, or process name:
```shell-session
amd-smi process --gpu 0
amd-smi process --pid 12345
amd-smi process --name python3
```
Use the `--general` flag to display only pid, process name, and memory usage, or
`--engine` flag to display engine usage:
```shell-session
amd-smi process --gpu 0 --general
amd-smi process --gpu 0 --engine
```
The `process` command also supports watch mode to continuously display process
information:
```shell-session
amd-smi process --watch 2
amd-smi process --watch 2 --watch_time 60
amd-smi process --watch 2 --iterations 10
```
(cli-ex-monitor)=
### Example output from amd-smi monitor
The following block is example output from the `amd-smi monitor` command without
additional modifiers. When no arguments are provided, all default monitor
metrics are enabled.
```bash
~$ amd-smi monitor
GPU XCP POWER GPU_T MEM_T GFX_CLK GFX% MEM% ENC% DEC% GTT_USAGE
0 0 49 W 27 °C 29 °C 43 MHz 2 % 0 % N/A 0 % 0.0/376.7 GB
1 0 42 W 29 °C 30 °C 47 MHz 2 % 0 % N/A 0 % 0.0/376.7 GB
2 0 33 W 30 °C 31 °C 49 MHz 2 % 0 % N/A 0 % 0.0/376.7 GB
3 0 40 W 26 °C 28 °C 40 MHz 2 % 0 % N/A 0 % 0.0/376.7 GB
```
The final memory column is chosen dynamically per device. On APUs (such as the
MI300A above), the tool reports the larger of the VRAM and GTT pools, so the
column header reads `GTT_USAGE`. On discrete GPUs it reports the dedicated VRAM
pool, so the header reads `VRAM_USAGE`. The example below is from a discrete
MI300X in `CPX`/`NPS4` mode; only the primary XCP of each partition reports
per-engine sensors, while the other XCPs share the physical device and report
`N/A` for those fields:
```bash
~$ amd-smi monitor
GPU XCP POWER GPU_T MEM_T GFX_CLK GFX% MEM% ENC% DEC% VRAM_USAGE
0 0 183 W 49 °C 48 °C 1427 MHz 0 % 0 % N/A 0 % 0.3/192.0 GB
1 1 N/A N/A N/A N/A N/A N/A N/A N/A 0.5/ 48.0 GB
2 2 N/A N/A N/A N/A N/A N/A N/A N/A 0.5/ 48.0 GB
3 3 N/A N/A N/A N/A N/A N/A N/A N/A 0.5/ 48.0 GB
4 0 182 W 50 °C 46 °C 1423 MHz 0 % 0 % N/A 0 % 0.3/192.0 GB
5 1 N/A N/A N/A N/A N/A N/A N/A N/A 0.5/ 48.0 GB
6 2 N/A N/A N/A N/A N/A N/A N/A N/A 0.5/ 48.0 GB
7 3 N/A N/A N/A N/A N/A N/A N/A N/A 0.5/ 48.0 GB
...
```
You can select specific metrics to monitor:
```shell-session
amd-smi monitor --power-usage --temperature
amd-smi monitor --gfx --mem
amd-smi monitor --vram-usage --ecc
amd-smi monitor --pcie
```
Use the `--process` flag to include a process information table below the
monitor output:
```shell-session
amd-smi monitor --process
```
The `monitor` command supports watch mode for continuous monitoring:
```shell-session
amd-smi monitor --watch 1
amd-smi monitor --watch 1 --watch_time 120
amd-smi monitor --watch 1 --iterations 30
```
### Listing CPER entries using amd-smi
This example code shows how to list CPER entries for all GPUs into files
```bash
~$ sudo amd-smi ras --cper --severity all --folder /tmp/cper_dump/
timestamp gpu_id severity file_name list of afids
2000/06/27 10:45:13 0 FATAL fatal-1.cper 30
2000/06/27 10:45:13 1 FATAL fatal-2.cper 30
2000/06/27 10:45:13 2 FATAL fatal-3.cper 30
2000/06/27 10:45:13 3 FATAL fatal-4.cper 30
2000/06/27 10:45:13 4 FATAL fatal-5.cper 30
2000/06/27 10:45:13 5 FATAL fatal-6.cper 30
2000/06/27 10:45:13 6 FATAL fatal-7.cper 30
2000/06/27 10:45:13 7 FATAL fatal-8.cper 30
```
This example code shows how to list CPER entries for a given GPU into files
```bash
~$ sudo amd-smi ras --cper --severity all --folder /tmp/cper_dump/ --gpu 1
timestamp gpu_id severity file_name list of afids
2000/06/27 10:45:13 1 FATAL fatal-1.cper 30
```
This example code shows how to list CPER entries and their JSON data for a given GPU into files
```bash
~$ sudo amd-smi ras --cper --severity all --folder /tmp/cper_dump/ --gpu 1 --json
timestamp gpu_id severity file_name list of afids
2000/06/27 10:45:13 1 FATAL fatal-1.cper 30
~$ ls -alh /tmp/cper_dump/
total 12K
drwxr-xr-x 2 root root 46 Sep 16 21:12 .
drwxrwxrwt 1 root root 4.0K Sep 16 18:03 ..
-rw-r--r-- 1 root root 376 Sep 16 21:12 fatal-1.cper
-rw-r--r-- 1 root root 347 Sep 16 21:12 fatal-1.json
~$ cat /tmp/cper_dump/fatal-1.json
{
"error_severity": "fatal",
"notify_type": "MCE",
"timestamp": "2000/06/27 10:45:13",
"signature": "CPER",
"revision": 256,
"signature_end": "0xffffffff",
"sec_cnt": 1,
"record_length": 376,
"platform_id": "111102-G40307-0C",
"creator_id": "136c692517001839",
"record_id": "f0000031",
"flags": 0,
"persistence_info": 0
}
```
This example code shows how to continuously list CPER entries without exiting
```bash
~$ sudo amd-smi ras --cper --follow --severity all --folder /tmp/cper_dump
Press CTRL + C to stop.
timestamp gpu_id severity file_name list of afids
2000/06/27 10:45:13 0 FATAL fatal-1.cper 30
2000/06/27 10:45:13 1 FATAL fatal-2.cper 30
2000/06/27 10:45:13 2 FATAL fatal-3.cper 30
2000/06/27 10:45:13 3 FATAL fatal-4.cper 30
2000/06/27 10:45:13 4 FATAL fatal-5.cper 30
2000/06/27 10:45:13 5 FATAL fatal-6.cper 30
2000/06/27 10:45:13 6 FATAL fatal-7.cper 30
2000/06/27 10:45:13 7 FATAL fatal-8.cper 30
...
```
This example code shows how to list CPER entries with a limited number of entries
```bash
~$ sudo amd-smi ras --cper --severity all --folder /tmp/cper_dump --file-limit 5
timestamp gpu_id severity file_name list of afids
2000/06/27 10:45:13 0 FATAL fatal-1.cper 30
2000/06/27 10:45:13 1 FATAL fatal-2.cper 30
2000/06/27 10:45:13 2 FATAL fatal-3.cper 30
2000/06/27 10:45:13 3 FATAL fatal-4.cper 30
2000/06/27 10:45:13 4 FATAL fatal-5.cper 30
```
This example code shows how to list a specific severity of CPER entries only
```bash
~$ sudo amd-smi ras --cper --severity fatal --folder /tmp/cper_dump/
timestamp gpu_id severity file_name list of afids
2000/06/27 10:45:13 0 FATAL fatal-1.cper 30
```
This example code shows how to dump AFID errors in a CPER file
```bash
~$ sudo amd-smi ras --afid --cper-file /tmp/cper_dump/fatal-1.cper
```
Refer to
[amd_smi_cper_example.py](https://github.com/ROCm/rocm-systems/blob/develop/projects/amdsmi/example/amd_smi_cper_example.py)
and
[amd_smi_afid_example.py](https://github.com/ROCm/rocm-systems/blob/develop/projects/amdsmi/example/amd_smi_afid_example.py)
for API examples.
## Memory tuning: UMA carveout and GTT
`amd-smi static --mem-carveout` / `amd-smi set --mem-carveout INDEX` and
`amd-smi node --gtt` / `amd-smi set --gtt GB` / `amd-smi reset --gtt` let
users inspect and tune the BIOS VRAM carveout and the TTM `pages_limit`
(shared GTT) respectively. Both features talk directly to kernel UAPI
interfaces (sysfs / modprobe.d) and do **not** require libdrm.
### Supported ASICs
| Feature | Hardware | Status |
|---|---|---|
| `--mem-carveout` (UMA carveout) | Strix and later APUs (gfx1150, gfx1151, gfx1152) whose VBIOS exposes ATCS 0xA | Supported |
| `--mem-carveout` (UMA carveout) | Radeon dGPUs, Instinct MI-series (MI100, MI200, MI300, MI300A) | Not supported — reported as `MEM_CARVEOUT: N/A (UMA carveout is not supported on this ASIC/VBIOS)` |
| `--gtt` (TTM `pages_limit`) | Any amdgpu system, including Instinct MI300A (`amdttm` / `amd-ttm`) and Ryzen APUs (`ttm`) | Supported |
### Prerequisites
- **UMA carveout:** Linux kernel >= 7.0 (upstream commit [`685b711`](https://github.com/torvalds/linux/commit/685b711); some distros backport it to earlier kernels), an APU VBIOS that advertises ATCS 0xA + IGP info table v2.3, root, and a reboot after changing the index.
- **GTT (TTM `pages_limit`):** root (to write `/etc/modprobe.d/.conf`), optionally `dracut` (the tool will rebuild the initramfs automatically when `dracut` is present), and a reboot to apply the new limit. amd-smi auto-detects the TTM kernel module name (`ttm`, `amdttm`, or `amd-ttm`) and writes the matching `.conf`.
### Troubleshooting: `MEM_CARVEOUT: N/A`
On MI300A (and every non-APU / pre-ATCS-0xA platform) the kernel does not
create `/sys/class/drm//device/uma/`, so `amd-smi static --mem-carveout`
prints
```text
MEM_CARVEOUT: N/A (UMA carveout is not supported on this ASIC/VBIOS)
```
This is expected. Use `amd-smi node --gtt` / `amd-smi set --gtt` to tune
shared GPU memory on those platforms instead.