Man Linux: Main Page and Category List

NAME

       nvidia-smi - NVIDIA System Management Interface program

SYNOPSIS

       nvidia-smi [options]

       Options: [-h, --help] [-x, --xml-format]
                [-f filename, --filename=filename]
                [-l, --loop-continuously] [-i sec, --interval=sec]
                [-t unit-number, --toggle-led=unit-number]
                [-a, --all-gpus ] [-d, --discrete-gpus ]
                [-L, --list-gpus ] [-g gpu-id, --gpu=gpu-id]
                [-c ruleset, --compute-mode-rules=ruleset]
                [-s, --show-compute-mode-rules ] [-e, --ecc-config=config]
                [-r, --report-ecc-config ] [-q, --query-gpu-info]
                [-p, --reset-persistent-ecc-error-counts]
                [-v, --reset-volatile-ecc-error-counts]
                [-P level, --set-power-limit=level]
                [-F filename, --fault-log=filename]

DESCRIPTION

       The  NVIDIA  System  Management Interface (nvidia-smi) program provides
       system state and diagnostic information for  the  NVIDIA  Tesla  Visual
       Computing  Systems.  It provides these data in either plain text or XML
       format, printed to stdout or optionally written  to  a  user  specified
       file.

OPTIONS

       -h, --help
              Print usage information and exit.

       -x, --xml-format
              Produce XML output. Unless an output file is specified using the
              -f option, all data are output to stdout.

       -f filename
       --filename=filename
              Log to the named file filename, rather than to stdout.

       -l, --loop-continuously
              Probe continuously, outputting to stdout unless an  output  file
              is  specified  using  the  -f  option.  If  an  output  file  is
              specified, it is updated on each loop iteration.

       -i sec
       --interval=sec
              Probe once every sec  seconds  when  probing  continuously.  The
              default and minimum interval is 1 second.

       -t unit-number
       --toggle-led=unit-number
              Toggle  the  LED  state  (ON  ->  OFF  or  OFF  ->  ON) for unit
              unit-number. The unit number of an NVIDIA Tesla Visual Computing
              System  can  be  found in the log file. Each NVIDIA Tesla Visual
              Computing System unit attached to a given system  has  a  unique
              unit number within that system.

       -a, --all-gpus
              Include  discrete  GPUs  when listings and consider them for GPU
              specific options if no GPU is specified  using  the  -g  option.
              Discrete  GPUs are GPUs not installed in QuadroPlex and/or Tesla
              computing systems.

       -d, --discrete-gpus
              Include  only  discrete  GPUs  in  listings  and  consider  them
              exclusively  for  GPU  specific  options  if no GPU is specified
              using the -g option.

       -L, --list-gpus
              Print a list of installed GPUs and exit.  By  default,  discrete
              GPUs  are  not  listed. This behavior can be changed with the -a
              and -d options.

       -g gpu-id
       --gpu=gpu-id
              Specify the GPU to be  targeted  when  specifying  GPU  specific
              options,  such  as  --show-compute-mode-rules.   The appropriate
              gpu-id identifier can be obtained using the -L option.

       -q, --query-gpu-info
              Print information about a given GPU or set  of  GPUs,  including
              the  product name, PCI ID, temperature, and ECC error counts. No
              information about any Visial Computing Systems attached  to  the
              host  is  listed.  Unless  the  -g  option  is  used to select a
              specific GPU, information for a set  of  GPUs  is  reported.  By
              default,  this  set  only  includes GPUs installed in QuadroPlex
              and/or Tesla Visual Computing Systems. The -a and -d options can
              be used to change the set.

       -c ruleset
       --compute-mode-rules=ruleset
              Set  the  compute  mode  rules  to ruleset for the GPU specified
              using the -g option.

              ruleset may be one of the following:
                   0 - Normal mode
                   1 - COMPUTE exclusive mode (only  one  COMPUTE  context  is
              allowed to run on the GPU)
                   2  -  COMPUTE  prohibited  mode  (no  COMPUTE  contexts are
              allowed to run on the GPU)

       -s, --show-compute-mode-rules
              Display the current compute mode ruleset for the  GPU  specified
              using the -g option. If no GPU is chosen with the -g option, the
              rules for all GPUs are shown.

       -e config
       --ecc-config=config
              Set the ECC configuration to config for the GPU specified  using
              the  -g  option.  Changes take effect with the next system boot.
              The -q option reports ECC error statistics if ECC  is  supported
              and enabled.

              config may be one of the following:
                   0 - Disabled
                   1  -  Enabled  (ECC  will be used for video memory and some
              engines)

       -r, --report-ecc-config
              Report  both  the  current  and  future   (after   reboot)   ECC
              configurations  for the GPU specified using the -g option. If no
              GPU is chosen with the -g option, the configuration for all GPUs
              is reported.

       -p, --reset-persistent-ecc-error-counts
              Reset  persistent  ECC error count. If no GPU is chosen with the
              -g option, the counts for all GPUs are reset.

       -v, --reset-volatile-ecc-error-counts
              Reset volatile ECC error count. If no GPU is chosen with the  -g
              option, the counts for all GPUs are reset.

       -P level
       --set-power-limit=level
              Set  the  input  power consumption limit, that the system is not
              allowed to exceed, to equal level in Watts.

       -F filename
       --fault-log=filename
              Read fault records, log them to  the  file  named  filename  and
              clear the fault records.

EXAMPLES

       nvidia-smi
              Probe once and log all relevant data to stdout.

       nvidia-smi -l
              Probe continuously and log to stdout.

       nvidia-smi -x -f newfile
              Probe once and log to the file newfile in XML format.

       nvidia-smi -g 0 -s
              Report the current COMPUTE ruleset for GPU 0.

AUTHOR

       Mayank Kaushik
       NVIDIA Corporation

COPYRIGHT

       Copyright (C) 2010 NVIDIA Corporation.