Virtualization can provide ease of management, redundancy, reliability, mobility, and multiple security advantages. It can also, however, add some computing overhead. Virtualizing GPUs can make GPUs sharable, which is especially useful for compute-intensive workloads like artificial intelligence or machine learning, but the high costs of GPUs makes the overhead a potential concern.

We tackled this problem by comparing the image classification throughput of a virtual environment, a Dell PowerEdge R7525 server powered by AMD EPYC processors with VMware vSphere 7.0 Update, to that of a bare-metal environment, a Dell PowerEdge R7525 with two AMD EPYC 7543 processors. Both environments had an NVIDIA A100 Tensor Core GPU.

Good news: The configuration with the virtualized GPU delivered 97.5 percent of bare-metal performance while also providing the sharing benefit of virtualization.

Dig in to learn more about what virtualized GPUs on Dell PowerEdge R7525 servers can do.