VMware Horizon Community
serinfbco
Enthusiast
Enthusiast

Nvidia GRID K1 - nvidia-smi gpu usage - horizon view 7.4

Hello,

We noticed a decreased performance with our Nvidia K1 Lab.

Just showing the windows desktop, with PCOIP or BLAST,  it takes around of 15-20% gpu (taskmgr) and it looks like the same on a core in nvidia-smi command

This computer lab is used for Autodesk 2018 (Autocad, Revit 3D, Civil 3D,etc.) so the gpu goes up to 100% full load and users complaints about lags

I found out a KB on Nvidia website for this issue, but it should be fixed for BLAST from Horizon View 7.0.1

nvidia-smi shows high GPU utilization for vGPU VMs with active Horizon sessions

- on VM, nvidia grid k1 120q profile

- on ESXi, nvidia vgpu manager 367.124 (vib)

- on OS, nvidia grid drivers 370.21

VMware ESXi, 6.5.0, 7388607

Tested Windows 10 1703, 1709, 1803 x64 education and Windows 7 x64

Vmware View Agent 7.0.1, 7.4.0, 7.5.0 (direct-connect)

Thank you for your help !

Reply
0 Kudos
2 Replies
techguy129
Expert
Expert

I would recommend setting your grid profile to at least 140q and see if that solves your issues. There were some issues with 120q and memory on the VM. 

----https://docs.nvidia.com/grid/4.2/grid-vgpu-release-notes-vmware-vsphere/index.html#nvenc-needs-at-le...

GRID Virtual GPU for VMware vSphere Release Notes :: GRID Software Documentation

NVENC requires at least 1 Gbyte of frame buffer

Description

Using the frame buffer for the NVIDIA hardware-based H.264/HEVC video encoder (NVENC) may cause memory exhaustion with vGPU profiles that have 512 Mbytes or less of frame buffer. To reduce the possibility of memory exhaustion, NVENC is disabled on profiles that have 512 Mbytes or less of frame buffer. Application GPU acceleration remains fully supported and available for all profiles, including profiles with 512 MBytes or less of frame buffer. NVENC support from both Citrix and VMware is a recent feature and, if you are using an older version, you should experience no change in functionality.

The following vGPU profiles have 512 Mbytes or less of frame buffer:

  • Tesla M6-0B, M6-0Q
  • Tesla M10-0B, M10-0Q
  • Tesla M60-0B, M60-0Q
  • GRID K100, K120Q
  • GRID K200, K220Q

Workaround

If you require NVENC to be enabled, use a profile that has at least 1 Gbyte of frame buffer.

Reply
0 Kudos
serinfbco
Enthusiast
Enthusiast

Hello,

I have tried out the profile with 1GB / 140q and the issue still exists on both protocols, PCOIP and Blast.

I looked if the gpu were in compute or graphics mode, they have the graphics class:

0000:06:00.0 Class 0300: 10de:13bd [vmgfx0]

0000:07:00.0 Class 0300: 10de:13bd [vmgfx1]

0000:08:00.0 Class 0300: 10de:13bd [vmgfx2]

0000:09:00.0 Class 0300: 10de:13bd [vmgfx3]

The gpu usage equals on how many pixels you display, (1920x1080 vs 1024x768)

1920x1080 = 15%

1366x768 =  8%

1024x768 = 6%

If we decrease the resolutions, we might gain 5~10% on each computer and it will lower the utilization until a patch is released.

Reply
0 Kudos