Home > Workload Solutions > Virtual Desktop Infrastructure > White Papers > Technical White Paper—VDI Graphics Acceleration on a Dell MX7000 Modular Chassis > NVIDIA nVector, SPECviewperf 13, Maya–Single VM Test
This section describes a single workstation that ran the NVIDIA nVector automated workload. The test was automated and data was captured by NVIDIA nVector. The single workstation VM was able to consume more than its default allocated resources due to the “Best Effort” vGPU scheduler being used. Because there was only a single VM per GPU, there was no contention for GPU resources. For additional information, see the NVIDIA vGPU documentation.
The following table shows the single Virtual Workstation VM configuration:
Configuration | Value |
vCPU | 6 |
vMemory | 32768 |
HardDisk | 120 |
GPU | grid_t4-8q |
GPUDriverVersion | 442.06 |
FRL | Disabled |
vSYNC | Default |
vDAVersion | 7.10.1 |
DirectConnectVersion | 7.10.1 |
CPUAffinity | Unset |
Screen Resolution | 1920 x 1080 |
QuantityMonitors | 1 |
The following table shows the average host utilization metrics for this test:
Workload | Density per host | Average CPU | Average GPU | Average free memory | Average net Mbps per user |
SPECviewperf 13 Maya | 1 | 5% | 34% | 334 GB | 10.08 Mbps |
The following figure shows the CPU core utilization for the system under test. CPU core utilization remained relatively low, with the highest point reaching 7.24 percent.
The following figure shows the GPU utilization for the system under test. GPU metrics were gathered using NVIDIA-SMI on the VMware ESXi host. GPU utilization for one of the GPUs was quite high for this test, while the other remained unused. This result is expected for this workload with a single VM.
The following figure shows the free memory in the system during the test. There were no variations in memory usage throughout the test because all vGPU-enabled workstation memory was reserved. There was no memory ballooning or swapping on either host.
The following figure shows the network utilization of the system during the test. More than enough bandwidth was available for the workload and the display protocol to accommodate the workload. Each server had a single 25 GbE network card (25,000 MBits) connected to it.