Hi,
I'm in the middle of a deployment of Horizon View 6 with Nvidia GRID K1. The end-users are going to use the Virtual Desktops with Maptek Vulcan software to 3D modeling.
The ESXi host is a Dell R720 with 2 Nvidia GRID K1, so we can have 8 Virtual Desktop with vDGA.
Every Virtual Desktop have this virtual hardware:
- 8 vCPU
- 8 GB of RAM
- 2 Virtual Disk (one of them with RDM).
- GPU Passthrough (vDGA)
- Windows 7 64 bits.
I've followed the "Graphics Acceleration in View Virtual Desktops" guide, tuning PCoIP settings in order to get the best performance available.
- All the communications is internal (LAN), without remote connections (WAN).
- The end-users are using Dell notebooks with Windows 7 and Horizon View client (the latest version).
We are running some performance tests, using Maptek Vulcan software emulating the production environment. The most of the tests were successful, however, when we try with heavier 3D graphics, the graphic performance of the virtual desktop is really poor.
- The CPU usage average is about 20%. CPU Ready is less than 5%, so no Ready time issues.
- The memory usage average is just of 4 out 8GB
- There is not high disk latency
- No network packet dropped
We are using nvidia-smi command to measure the GPU performance. Through all the tests, the GPU usage goes between 20% and 60%, however the graphic performance is not good enough when the user is working with heavy 3D graphics. I think an 20-60% GPU usage is OK for the most of the tests, but if we are trying to work with heavy 3D graphic I think the GPU usage should be higher, but isn't, and the graphic performance is really poor. The overall performance (CPU, memory, disk and network) looks fine.
How can I get the max performance available on the GPU? How can I take advantage of all the GPU capacity?