nvidia-smiを使ってGPU使用率と温度と消費電力を取得してくるPowerShellスクリプト mackerel-agent出力準拠

Gpu.bat

@echo off
cd %~dp0

powershell -ExecutionPolicy RemoteSigned -File .\Gpu.ps1

Gpu.ps1

#define
$path = "C:\Program Files\NVIDIA Corporation\NVSMI\nvidia-smi.exe"

Function GetNvidiaSMIList($p)
{
$ninfo = New-Object System.Diagnostics.ProcessStartInfo
$ninfo.filename = $p
$ninfo.RedirectStandardError = $true
$ninfo.RedirectStandardOutput = $true
$ninfo.UseShellExecute = $false
$ninfo.Arguments = "-L"
$n = New-Object System.Diagnostics.Process
$n.StartInfo = $ninfo
$n.Start() | Out-Null
$n.WaitForExit()
$stdout = $n.StandardOutput.ReadToEnd() -split "`r`n"
return $stdout
}

Function GetNvidiaSMIStatus($p,$cmd,$i)
{
echo $args 
$ninfo = New-Object System.Diagnostics.ProcessStartInfo
$ninfo.filename = $p
$ninfo.RedirectStandardError = $true
$ninfo.RedirectStandardOutput = $true
$ninfo.UseShellExecute = $false
$ninfo.Arguments = "-q -i $i -d $cmd"
$n = New-Object System.Diagnostics.Process
$n.StartInfo = $ninfo
$n.Start() | Out-Null
$n.WaitForExit()
return $n.StandardOutput.ReadToEnd() | tee -Variable text
}

Function Epoch_utc
{
$date1 = Get-Date -Date "01/01/1970"
$date2 = Get-Date
$date = (New-TimeSpan -Start $date1 -End $date2).TotalSeconds
$epoch_jst = [math]::Truncate($date)
$epoch_utc = $epoch_jst - 32400
return [string]$epoch_utc | tee -Variable text
}

#GPUList
$gpulist = @()
foreach($g in GetNvidiaSMIList $path ){
  if ($g -ne ""){
    $garray = @()
    $g -match "GPU (?<Gnum>.*?): (?<Gname>.*?) \(UUID: (?<Guuid>.*?)\)" | Out-Null
    $garray = @($Matches.Gnum,$Matches.Gname.Replace(' ', '_'),$Matches.Guuid)
    $gpulist += ,@($garray)
  }
}

$time = Epoch_utc

#GPUUtilization
foreach($gg in $gpulist){
  $cmd = "UTILIZATION"
  $out = GetNvidiaSMIStatus $path $cmd $gg[0]
    if ($out -notmatch "Gpu +: (?<Gpuutil>.*?)\%\r\n.*?Memory +: (?<Gmemutil>.*?)\%\r\n.*?Encoder +: (?<Gencutil>.*?)\%\r\n.*?Decoder +: (?<Gdecutil>.*?)\%")
    {
      $stat = @(0,0,0,0)
    }
    else
    {
      $stat = @($Matches.Gpuutil,$Matches.Gmemutil,$Matches.Gencutil,$Matches.Gdecutil)
  }
  Write-output "GPU.Utilization.$($gpulist[$gg[0]][1]).GPU`t$($stat[0])`t$time"
  Write-output "GPU.Utilization.$($gpulist[$gg[0]][1]).Memory`t$($stat[1])`t$time"
  Write-output "GPU.Utilization.$($gpulist[$gg[0]][1]).Encoder`t$($stat[2])`t$time"
  Write-output "GPU.Utilization.$($gpulist[$gg[0]][1]).Decoder`t$($stat[3])`t$time"
}

#GPUTemp
foreach($gg in $gpulist){
  $cmd = "TEMPERATURE"
  $out = GetNvidiaSMIStatus $path $cmd $gg[0]
  $out -match "GPU Current Temp +: (?<Gtemp>.*?) C" | Out-Null
  $stat = @($Matches.Gtemp)
  Write-output "GPU.Temp.$($gpulist[$gg[0]][1]).Current`t$stat`t$time"
}

#GPUWatt
foreach($gg in $gpulist)
{
  $cmd = "POWER"
  $out = GetNvidiaSMIStatus $path $cmd $gg[0]
  if ($out -notmatch "Power Management +: N/A")
  {
    $out -match "Power Draw +: (?<Gwatt>.*?) W" | Out-Null
    $stat = @($Matches.Gwatt)
    }
    else
    {
      $stat = "0"
    }
  Write-output "GPU.Power.$($gpulist[$gg[0]][1]).Current`t$stat`t$time"
}