AnsweredAssumed Answered

AMD rocm-smi does not report GPU utilization. Can someone please explain why?

Question asked by colesdav on Aug 19, 2018
Latest reply on Aug 21, 2018 by colesdav

Hello,

 

I have been looking for a means to monitor and control my GPU's on Ubuntu 16.04 LTS running with amdgpu-pro-18.10-572953 drivers.
I would prefer to use a GUI to do this.

I have managed to compile Radeon Profile successfully and it seems to run o.k. but there is a problem that GPU Utilization is not reported.

 

I have also installed ROCm and  rocm-smi.

 

I looked at rocm-smi and I see that there is no way to report GPU Utilization in a similar mannmer to that reported on AMD Adrenalin Drivers for Windows.

 

Here are the options I see:

/opt/rocm/bin/rocm-smi -h

usage: rocm-smi [-h] [-d DEVICE] [-i] [-v] [-hw] [-t] [-c] [-g] [-f] [-p] [-P] [-o] [-l] [-s] [-a] [-r]

                [--setsclk LEVEL [LEVEL ...]] [--setmclk LEVEL [LEVEL ...]] [--resetfans] [--setfan LEVEL]

                [--setperflevel LEVEL] [--setoverdrive %] [--setprofile # # # # #] [--resetprofile]

                [--load FILE | --save FILE] [--autorespond RESPONSE]

 

AMD ROCm System Management Interface

 

optional arguments:

  -h, --help                   show this help message and exit

  --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   Execute command on specified device

 

  -i, --showid                 Show GPU ID

  -v, --showvbios              Show VBIOS version

  -hw, --showhw                Show Hardware details

  -t, --showtemp               Show current temperature

  -c, --showclocks             Show current clock frequencies

  -g, --showgpuclocks          Show current GPU clock frequencies

  -f, --showfan                Show current fan speed

  -p, --showperflevel          Show current PowerPlay Performance Level

  -P, --showpower              Show current power consumption

  -o, --showoverdrive          Show current OverDrive level

  -l, --showprofile            Show Compute Profile attributes

  -s, --showclkfrq             Show supported GPU and Memory Clock

  -a, --showallinfo            Show Temperature, Fan and Clock values

 

  -r, --resetclocks            Reset sclk and mclk to default (auto)

  --setsclk LEVEL [LEVEL ...]  Set GPU Clock Frequency Level(s) (manual)

  --setmclk LEVEL [LEVEL ...]  Set GPU Memory Clock Frequency Level(s) (manual)

  --resetfans                  Reset fans to automatic (driver) control

  --setfan LEVEL               Set GPU Fan Speed Level

  --setperflevel LEVEL         Set PowerPlay Performance Level

  --setoverdrive %             Set GPU OverDrive level (manual|high)

  --setprofile # # # # #       Specify Compute Profile attributes (auto)

  --resetprofile               Reset Compute Profile to default values

 

  --autorespond RESPONSE       Response to automatically provide for all prompts (NOT RECOMMENDED)

 

Is there any reason why rocm-smi does not report GPU Utilization?

 

I was hoping to be able to run a simple Linux Script to report the utilization in a shell using command similar to this:

 

clear; for i in {1..1000} ; do /opt/rocm/bin/rocm-smi -P ; sleep 1 ; clear; done

 

Which is set up to report GPU Power 1000 times, with a pause interval of ~ 1.
It produces an output like this:

 

====================ROCm System Management Interface====================

================================================================================

GPU[1]    : Average GPU Power: 10.25 W
GPU[0]    : Average GPU Power: 15.112 W

================================================================================

====================       End of ROCm SMI Log      ====================

 

System Information is below:

 

Motherboard Asus Z97 Deluxe.

CPU i7-4790K.

GPU1: R9 Fury X.

GPU 2. R9 Nano

PSU Corsair AX1200i

SSHD 2TB Seagate.

OS: Ubuntu 16.04 LTS.

 

Thanks.

Outcomes