Main:

-h, --help

show this help message and exit

--gpureset

Reset specified GPU (One GPU must be specified). This flag will attempt to reset the GPU for a specified device. This will invoke the GPU reset through the kernel debugfs file amdgpu_gpu_recover. Note that GPU reset will not always work, depending on the manner in which the GPU is hung.

--load FILE

Load Clock, Fan, Performance and Profile settings from FILE

--save FILE

Save Clock, Fan, Performance and Profile settings to FILE

-d DEVICE [DEVICE ...], --device DEVICE [DEVICE ...]

Execute command on specified device

Display Options:

--alldevices

--showhw

Show Hardware details

-a, --showallinfo

Show Temperature, Fan and Clock values

Topology:

-i, --showid

Show GPU ID

-v, --showvbios

Show VBIOS version

-e [EVENT ...], --showevents [EVENT ...]

Show event list

--showdriverversion

Show kernel driver version. This flag will print out the AMDGPU module version for amdgpu-pro or ROCK kernels. For other kernels, it will simply print out the name of the kernel (uname).

--showtempgraph

Show Temperature Graph

--showfwinfo [BLOCK ...]

Show FW information

--showmclkrange

Show mclk range

--showmemvendor

Show GPU memory vendor

--showsclkrange

Show sclk range

--showproductname

Show SKU/Vendor name. This uses the pci.ids file to print out more information regarding the GPUs on the system. update-pciids(8) may need to be executed on the machine to get the latest PCI ID snapshot, as certain newer GPUs will not be present in the stock pci.ids file, and the file may even be absent on certain OS installation types.

--showserial

Show GPU's Serial Number. This flag will print out the serial number for the graphics card. NOTE: This is currently only supported on Vega20 server cards that support it. Consumer cards and cards older than Vega20 will not support this feature.

--showuniqueid

Show GPU's Unique ID

--showvoltagerange

Show voltage range

--showbus

Show PCI bus number

Pages information:

--showpagesinfo

Show retired, pending and unreservable pages

--showpendingpages

Show pending retired pages

--showretiredpages

Show retired pages

--showunreservablepages

Show unreservable pages. The above four flags display the different "bad pages" as reported by the kernel. The three types of pages are: Retired pages (reserved pages) - These pages are reserved and are unable to be used. Pending pages - These pages are pending for reservation, and will be reserved/retired. Unreservable pages - These pages are not reservable for some reason.

Hardware-related information:

-f, --showfan

Show current fan speed

-P, --showpower

Show current Average Graphics Package Power Consumption. "Graphics Package" refers to the GPU plus any HBM (High-Bandwidth memory) modules, if present.

-t, --showtemp

Show current temperature

-u, --showuse

Show current GPU use

--showmemuse

Show current GPU memory used. This used to indicate how busy the respective blocks are. For example, for --showuse (gpu_busy_percent sysfs file), the SMU samples every ms or so to see if any GPU block (RLC, MEC, PFP, CP) is busy. If so, that's 1 (or high). If not, that's 0 (low). If we have 5 high and 5 low samples, that means 50% utilization (50% GPU busy, or 50% GPU use). The windows and sampling vary from generation to generation, but that is how GPU and VRAM use is calculated in a generic sense. --showmeminfo (and VRAM% in concise output) will show the amount of VRAM used (visible, total, GTT), as well as the total available for those partitions. The percentage shown there indicates the amount of used memory in terms of current allocations.

--showvoltage

Show current GPU voltage.

Software-related/controlled information:

-b, --showbw

Show estimated PCIe use This shows an approximation of the number of bytes received and sent by the GPU over the last second through the PCIe bus. Note that this will not work for APUs since data for the GPU portion of the APU goes through the memory fabric and does not 'enter/exit' the chip via the PCIe interface, thus no accesses are generated, and the performance counters can't count accesses that are not generated. NOTE: It is not possible to easily grab the size of every packet that is transmitted in real time, so the kernel estimates the bandwidth by taking the maximum payload size (mps), which is the max size that a PCIe packet can be. and multiplies it by the number of packets received and sent. This means that the SMI will report the maximum estimated bandwidth, the actual usage could (and likely will be) less.

-c, --showclocks

Show current clock frequencies

Clock type	Description
DCEFCLK	DCE (Display)
FCLK	Data fabric (VG20 and later) - Data flow from XGMI, Memory, PCIe
SCLK	GFXCLK (Graphics core)
Note	SOCCLK split from SCLK as of Vega10. Pre-Vega10 they were both controlled by SCLK
MCLK	GPU Memory (VRAM)
PCLK	PCIe bus
Note	This gives 2 speeds, PCIe Gen1 x1 and the highest available based on the hardware
SOCCLK	System clock (VG10 and later) - DF, MM HUB, AT HUB, SYSTEM HUB, OSS, DFD
Note	DF split from SOCCLK as of Vega20. Pre-Vega20 they were both controlled by SOCCLK

-g, --showgpuclocks

Show current GPU clock frequencies

-l, --showprofile

Show Compute Profile attributes

-M, --showmaxpower

Show maximum graphics package power this GPU will consume. This limit is enforced by the hardware.

-m, --showmemoverdrive

Show current GPU Memory Clock OverDrive level

-o, --showoverdrive

Show current GPU Clock OverDrive level

-p, --showperflevel

Show current DPM Performance Level

-S, --showclkvolt

Show supported GPU and Memory Clocks and Voltages

-s, --showclkfrq

Show supported GPU and Memory Clock

--showmeminfo TYPE [TYPE ...]

Show Memory usage information for given block(s) TYPE This allows the user to see the amount of used and total memory for a given block (vram, vis_vram, gtt). It returns the number of bytes used and total number of bytes for each block 'all' can be passed as a field to return all blocks, otherwise a quoted-string is used for multiple values (e.g. "vram vis_vram") vram refers to the Video RAM, or graphics memory, on the specified device vis_vram refers to Visible VRAM, which is the CPU-accessible video memory on the device gtt refers to the Graphics Translation Table.

--showpids [VERBOSE]

Show current running KFD PIDs (pass details to VERBOSE for detailed information)

--showpidgpus [SHOWPIDGPUS ...]

Show GPUs used by specified KFD PIDs (all if no arg given)

--showreplaycount

Show PCIe Replay Count

--showrasinfo [SHOWRASINFO ...]

Show RAS enablement information and error counts for the specified block(s) (all if no arg given) This shows the RAS information for a given block. This includes enablement of the block (currently GFX, SDMA and UMC are the only supported blocks) and the number of errors ue - Uncorrectable errors ce - Correctable errors.

--showvc

Show voltage curve

--showxgmierr

Show XGMI error information since last read

--showtopo

Show hardware topology information

--showtopoaccess

Shows the link accessibility between GPUs

--showtopoweight

Shows the relative weight between GPUs

--showtopohops

Shows the number of hops between GPUs

--showtopotype

Shows the link type between GPUs

--showtoponuma

Shows the numa nodes

--showenergycounter

Energy accumulator that stores amount of energy consumed

--shownodesbw

Shows the numa nodes

--showcomputepartition

Shows current compute partitioning

--shownpsmode

Shows current NPS mode

Set options:

--setclock TYPE LEVEL

Set Clock Frequency Level(s) for specified clock (requires manual Perf level)

--setsclk LEVEL [LEVEL ...]

Set GPU Clock Frequency Level(s) (requires manual Perf level)

--resetperfdeterminism

Disable performance determinism

--setmclk LEVEL [LEVEL ...]

Set GPU Memory Clock Frequency Level(s) (requires manual Perf level)

The two above options allow you to set a mask for the levels. For example, if a GPU has 8 clock levels, you can set a mask to use levels 0, 5, 6 and 7 with --setsclk 0 5 6 7 . This will only use the base level, and the top 3 clock levels. This will allow you to keep the GPU at base level when there is no GPU load, and the top 3 levels when the GPU load increases.

NOTES:
The clock levels will change dynamically based on GPU load based on the default
Compute and Graphics profiles. The thresholds and delays for a custom mask cannot
be controlled through the SMI tool.

This flag automatically sets the Performance Level to "manual" as the mask is not
applied when the Performance level is set to auto.

--setclock LEVEL LEVEL

Set Clock Frequency Level(s) for specified clock (requires manual Perf level)

--setpcie LEVEL [LEVEL ...]

Set PCIE Clock Frequency Level(s) (requires manual Perf level)

--setslevel SCLKLEVEL SCLK SVOLT

Change GPU Clock frequency (MHz) and Voltage (mV) for a specific Level

--setmlevel MCLKLEVEL MCLK MVOLT

Change GPU Memory clock frequency (MHz) and Voltage for (mV) a specific Level

--setvc POINT SCLK SVOLT

Change SCLK Voltage Curve (MHz mV) for a specific point

--setsrange SCLKMIN SCLKMAX

Set min and max SCLK speed

--setmrange MCLKMIN MCLKMAX

Set min and max MCLK speed

--setfan LEVEL

Set GPU Fan Speed (Level or %). This sets the fan speed to a value ranging from 0 to maxlevel, or from 0%-100% If the level ends with a %, the fan speed is calculated as pct*maxlevel/100
(maxlevel is usually 255, but is determined by the ASIC).

NOTE: While the hardware is usually capable of overriding this value when required, it is
recommended to not set the fan level lower than the default value for extended periods
of time.

--setperflevel LEVEL

Set Performance Level This lets you use the pre-defined Performance Level values for clocks and power profile, which can include: auto (Automatically change values based on GPU workload) low (Keep values low, regardless of workload) high (Keep values high, regardless of workload) manual (Only use values defined by --setsclk and --setmclk).

--setoverdrive %

Set GPU OverDrive level (requires manual|high Perf level)

--setmemoverdrive %

Set GPU Memory Overclock OverDrive level (requires manual|high Perf level) The above two options are DEPRECATED IN NEWER KERNEL VERSIONS (use --setslevel/--setmlevel instead). This sets the percentage above maximum for the max Performance Level. For example, --setoverdrive 20 will increase the top sclk level by 20%, similarly --setmemoverdrive 20 will increase the top mclk level by 20%. Thus if the maximum clock level is 1000MHz, then --setoverdrive 20 will increase the maximum clock to 1200MHz. Note this option can be used in conjunction with the --setsclk/--setmclk mask. Operating the GPU outside of specifications can cause irreparable damage to your hardware. Please observe the warning displayed when using this option. This flag automatically sets the clock to the highest level, as only the highest level is increased by the OverDrive value.

--setpoweroverdrive WATTS

Set the maximum GPU power using Power OverDrive in Watts This allows users to change the maximum power available to a GPU package. The input value is in Watts. This limit is enforced by the hardware, and some cards allow users to set it to a higher value than the default that ships with the GPU. This Power OverDrive mode allows the GPU to run at higher frequencies for longer periods of time, though this may mean the GPU uses more power than it is allowed to use per power supply specifications. Each GPU has a model-specific maximum Power OverDrive that is will take; attempting to set a higher limit than that will cause this command to fail. Note operating the GPU outside of specifications can cause irreparable damage to your hardware. Please observe the warning displayed when using this option.

--setprofile SETPROFILE

Specify Power Profile level (#) or a quoted string of CUSTOM Profile attributes "# # # #..." (requires manual Perf level) The Compute Profile accepts 1 or n parameters, either the Profile to select (see --showprofile for a list of preset Power Profiles) or a quoted string of values for the CUSTOM profile. Note these values can vary based on the ASIC, and may include: SCLK_PROFILE_ENABLE - Whether or not to apply the 3 following SCLK settings (0=disable,1=enable). Note: this is a hidden field. If set to 0, the following 3 values are displayed as '-’.

Setting	Description
SCLK_UP_HYST	Delay before sclk is increased (in milliseconds)
SCLK_DOWN_HYST	Delay before sclk is decresed (in milliseconds)
SCLK_ACTIVE_LEVEL	Workload required before sclk levels change (in %)

MCLK_PROFILE_ENABLE - Whether or not to apply the 3 following MCLK settings (0=disable,1=enable) NOTE: This is a hidden field. If set to 0, the following 3 values are displayed as '-'.

Setting	Description
MCLK_UP_HYST	Delay before mclk is increased (in milliseconds)
MCLK_DOWN_HYST	Delay before mclk is decresed (in milliseconds)
MCLK_ACTIVE_LEVEL	Workload required before mclk levels change (in %)

Other settings:

Setting	Description
BUSY_SET_POINT	Threshold for raw activity level before levels change
FPS	Frames Per Second
USE_RLC_BUSY	When set to 1, DPM is switched up as long as RLC busy message is received
MIN_ACTIVE_LEVEL	Workload required before levels change (in %)

NOTES:
When a compute queue is detected, the COMPUTE Power Profile values will be automatically
applied to the system, provided that the Perf Level is set to "auto".

The CUSTOM Power Profile is only applied when the Performance Level is set to "manual"
so using this flag will automatically set the performance level to "manual".

It is not possible to modify the non-CUSTOM Profiles. These are hard-coded by the kernel.

--setperfdeterminism SCLK

Set clock frequency limit to get minimal performance variation

--setcomputepartition {CPX,SPX,DPX,TPX,QPX,cpx,spx,dpx,tpx,qpx}

Set compute partition

--setnpsmode {NPS1,NPS2,NPS4,NPS8,nps1,nps2,nps4,nps8}

Set nps mode

--rasenable BLOCK ERRTYPE

Enable RAS for specified block and error type

--rasdisable BLOCK ERRTYPE

Disable RAS for specified block and error type

--rasinject BLOCK

Inject RAS poison for specified block (ONLY WORKS ON UNSECURE BOARDS)

Reset options:

-r, --resetclocks

Reset clocks and OverDrive to default

--resetfans

Reset fans to automatic (driver) control

--resetprofile

Reset Power Profile back to default

--resetpoweroverdrive

Set the maximum GPU power back to the device default state

--resetxgmierr

Reset XGMI error count

--resetperfdeterminism

Disable performance determinism

--resetcomputepartition

Resets to boot compute partition state

--resetnpsmode

Resets to boot NPS mode state

Auto-response options:

--autorespond RESPONSE

Response to automatically provide for all prompts (NOT RECOMMENDED)

Output options:

--loglevel LEVEL

This will allow the user to set a logging level for the SMI's actions, one of debug/info/warning/error/critical. Currently this is only implemented for sysfs writes, but can easily be expanded upon in the future to log other things from the SMI.

--json

Print output in JSON format

--csv

Print output in CSV format