当前位置:网站首页>VMware esxi 6.5 installation NVIDIA K2 GPU graphics card hardware driver and configuration

VMware esxi 6.5 installation NVIDIA K2 GPU graphics card hardware driver and configuration

2020-12-06 10:28:27 osc_ kf7nv2km

Please refer to  https://blogs.vmware.com/apps/2018/09/using-gpus-with-virtual-machines-on-vsphere-part-3-installing-the-nvidia-grid-technology.html

GPU Compatibility list query :https://www.nvidia.com/en-us/data-center/tesla/tesla-qualified-servers-catalog/


 

stay vSphere Superior emissary GPU And virtual machine installation NVIDIA GRID technology

Host on system maintenance

esxcli system maintenanceMode set --enable true

( You can operate on the web by yourself )

image.png


take GPU The device is set to vGPU Pattern

esxcli graphics host set --default-type SharedPassthru

image.png

return Default Graphics Type: SharedPassthru

esxcli graphics host get

image.png 

install VIB

Install the video card driver
esxcli software vib install -v /NVIDIA/NVIDIA-VMware_ESXi_6.5_Host_Driver_390.42-1OEM.670.0.0.7535516.vib

The document path here is in the root directory .

image.png
Update video card driver
esxcli software vib update -v /tmp/NVIDIA-VMware_ESXi_6.5_Host_Driver_367.128-1OEM.650.0.0.4598673.vib

 


ESXi The host server exits maintenance mode

esxcli system maintenanceMode set --enable false

Installed in the ESXi Host computer VIB, And make sure that NVIDIA VIB

esxcli software vib list | grep -i NVIDIA


 


Drive check

gpuvm  vGPU

image.png 

Drive check 2

lspci -n | grep 10de

image.png


among 10de Represents that this device comes from NVIDIA company

0300: Graphics — Represents the graphic mode

0302: Compute — Then it shows the calculation mode , Need to pass through gpumodeswitch Change to graphic mode

GPU Card and ESXi Whether it can work normally

nvidia-smi

image.png 

Check if there is vgpu resources

nvidia-smi vgpu –s

image.png

 

Partial error reporting failed to initialize NVML:Unknown Error

Problem analysis

  1. DELL The server BIOS Set up I/O Parameters
  2. ECC Set up
  3. The driver version of NVIDIA graphics card is wrong

 

terms of settlement

  1. Set up BIOS Memory mapping I/O The library is set to 512GB( By default 56TB)
  2. close ECC
  3. Change the driver version of the graphics card


版权声明
本文为[osc_ kf7nv2km]所创,转载请带上原文链接,感谢
https://chowdera.com/2020/12/202012061028036193.html