A top-like utility for monitoring CUDA activity on a GPU

Question

I m trying to monitor a process that uses CUDA and MPI  is there any way I could do this  something like the command  quot top quot  but that monitors the GPU too

User · Answer

You can try nvtop  which is similar to the widely-used htop tool but for NVIDIA GPUs  Here is a screenshot of nvtop of it in action

User · Answer

Download and install latest stable CUDA driver  4 2  from here  On linux  nVidia-smi 295 41 gives you just what you want  use nvidia-smi    root localhost release   nvidia-smi  Wed Sep 26 23 16 16 2012         ------------------------------------------------------                           NVIDIA-SMI 3 295 41   Driver Version  295 41                                   ------------------------------- ---------------------- ----------------------    Nb   Name                       Bus Id        Disp     Volatile ECC SB   DB     Fan   Temp   Power Usage  Cap   Memory Usage           GPU Util  Compute M                                                                                      0   Tesla C2050                 0000 05 00 0  On               0          0      30    62 C  P0    N A    N A     3    70MB   2687MB     44      Default       ------------------------------- ---------------------- ----------------------    Compute processes                                                GPU Memory      GPU  PID     Process name                                       Usage                                                                                           0   7336       align                                                 61MB     -----------------------------------------------------------------------------    EDIT  In latest NVIDIA drivers  this support is limited to Tesla Cards

User · Answer

There is Prometheus GPU Metrics Exporter  PGME  that leverages the nvidai-smi binary  You may try this out  Once you have the exporter running  you can access it via http   localhost 9101 metrics  For two GPUs  the sample result looks like this    temperature gpu gpu  TITAN X  Pascal  0    41 utilization gpu gpu  TITAN X  Pascal  0    0 utilization memory gpu  TITAN X  Pascal  0    0 memory total gpu  TITAN X  Pascal  0    12189 memory free gpu  TITAN X  Pascal  0    12189 memory used gpu  TITAN X  Pascal  0    0 temperature gpu gpu  TITAN X  Pascal  1    78 utilization gpu gpu  TITAN X  Pascal  1    95 utilization memory gpu  TITAN X  Pascal  1    59 memory total gpu  TITAN X  Pascal  1    12189 memory free gpu  TITAN X  Pascal  1    1738 memory used gpu  TITAN X  Pascal  1    10451

User · Answer

If you just want to find the process which is running on gpu  you can simply using the following command    lsof  dev nvidia    For me  nvidia-smi and watch -n 1 nvidia-smi are enough in most cases  Sometimes nvidia-smi shows no process but the gpu memory is used up so i need to use the above command to find the processes

User · Answer

You can use the  monitoring program glances with its  GPU monitoring plug-in    open source to install  sudo apt-get install -y python-pip  sudo pip install glances gpu  to launch  sudo glances     It also monitors the CPU  disk IO  disk space  network  and a few other things

User · Answer

Use argument  --query-compute-apps     nvidia-smi --query-compute-apps pid process name used memory --format csv   for further help  please follow  nvidia-smi --help-query-compute-app

User · Answer

This may not be elegant  but you can try  while true  do sleep 2  nvidia-smi  done   I also tried the method by  Edric  which works  but I prefer the original layout of nvidia-smi

User · Answer

Another useful monitoring approach is to use ps filtered on processes that consume your GPUs  I use this one a lot   ps f -o user pgrp pid pcpu pmem start time command -p  lsof -n -w -t  dev nvidia     That ll show all nvidia GPU-utilizing processes and some stats about them  lsof     retrieves a list of all processes using an nvidia GPU owned by the current user  and ps -p     shows ps results for those processes  ps f shows nice formatting for child parent process relationships   hierarchies  and -o specifies a custom formatting  That one is similar to just doing ps u but adds the process group ID and removes some other fields    One advantage of this over nvidia-smi is that it ll show process forks as well as main processes that use the GPU    One disadvantage  though  is it s limited to processes owned by the user that executes the command  To open it up to all processes owned by any user  I add a sudo before the lsof    Lastly  I combine it with watch to get a continuous update  So  in the end  it looks like   watch -n 0 1  ps f -o user pgrp pid pcpu pmem start time command -p  sudo lsof -n -w -t  dev nvidia      Which has output like   Every 0 1s  ps f -o user pgrp pid pcpu pmem start time command -p  sudo lsof -n -w -t  dev nvi     Mon Jun  6 14 03 20 2016 USER      PGRP   PID  CPU  MEM  STARTED     TIME COMMAND grisait  27294 50934  0 0  0 1   Jun 02 00 01 40  opt google chrome chrome --type gpu-process --channel 50877 0 2015482623 grisait  27294 50941  0 0  0 0   Jun 02 00 00 00      opt google chrome chrome --type gpu-broker grisait  53596 53596 36 6  1 1 13 47 06 00 05 57 python -u process examples py grisait  53596 33428  6 9  0 5 14 02 09 00 00 04     python -u process examples py grisait  53596 33773  7 5  0 5 14 02 19 00 00 04     python -u process examples py grisait  53596 34174  5 0  0 5 14 02 30 00 00 02     python -u process examples py grisait  28205 28205  905  1 5 13 30 39 04 56 09 python -u train py grisait  28205 28387  5 8  0 4 13 30 49 00 01 53     python -u train py grisait  28205 28388  5 3  0 4 13 30 49 00 01 45     python -u train py grisait  28205 28389  4 5  0 4 13 30 49 00 01 29     python -u train py grisait  28205 28390  4 5  0 4 13 30 49 00 01 28     python -u train py grisait  28205 28391  4 8  0 4 13 30 49 00 01 34     python -u train py

User · Answer

I m not aware of anything that combines this information  but you can use the nvidia-smi tool to get the raw data  like so  thanks to  jmsu for the tip on -l      nvidia-smi -q -g 0 -d UTILIZATION -l                NVSMI LOG                Timestamp                         Tue Nov 22 11 50 05 2011  Driver Version                    275 19  Attached GPUs                     2  GPU 0 1 0     Utilization         Gpu                       0           Memory                    0

User · Answer

I find gpustat very useful  In can be installed with pip install gpustat  and prints breakdown of usage by processes or users

User · Answer

In Linux Mint  and most likely Ubuntu  you can try  nvidia-smi --loop 1

User · Answer

Just use watch nvidia-smi  it will output the message by 2s interval in default   For example  as the below image     You can also use watch -n 5 nvidia-smi  -n 5 by 5s interval

User · Answer

I created a batch file with the following code in a windows machine to monitor every second  It works for me    loop cls  C  Program Files NVIDIA Corporation NVSMI nvidia-smi  timeout  T 1 goto loop   nvidia-smi exe is usually located in  C  Program Files NVIDIA Corporation  if you want to run the command only once

User · Answer

To get real-time insight on used resources  do   nvidia-smi -l 1  This will loop and call the view at every second   If you do not want to keep past traces of the looped call in the console history  you can also do   watch -n0 1 nvidia-smi  Where 0 1 is the time interval  in seconds

User · Answer

you can use nvidia-smi pmon -i 0 to monitor every process in GPU 0  including compute mode  sm usage  memory usage  encoder usage  decoder usage

[cuda] A top-like utility for monitoring CUDA activity on a GPU

Examples related to cuda

Examples related to process-monitoring

Examples related to resource-monitor