The VDI Solutions team used NVIDIA nVector to enable all SPECviewperf viewsets to automate across multiple VDI desktop VMs. The VMs accessed individual GPUs simultaneously with NVIDA virtual GPU (vGPU) software.
As shown in the following table, each VM configuration was identical on both systems. The increased GPU memory capacity on the T560 allowed the test to run concurrently across 30 VMs, which is a 50 percent increase from 20 VMs on the T550.
Component | PowerEdge T550 | PowerEdge T560 |
vCPU | 4 | 4 |
Memory | 8 GB | 8 GB |
vGPU Profile | 4 Q | 4 Q |
VDI Storage | 120 GB | 120 GB |
User Storage | 100 GB | 100 GB |
VMs per Host | 20 | 30 |
The derivation for the VMs per host calculation is as follows:
- Total Frame Buffer = Total GPU Memory X Number of GPUs
For the T560 with 24 GB memory per L4 GPU -> 24 X 5 = 120 GB
For the T550 with16 GB memory per A2 GPU -> 16 X 5 = 80 GB
- Number of VMs = Total Frame Buffer / Number of GB per VM
For the T560 -> 120 / 4 GB per VM = 30 VMs
For the T560 -> 80 / 4 GB per VM = 20 VMs