Graphics Processing Units (GPUs) represent a key accelerator technology on vSphere for those who are working in the fields of machine learning (ML) and high performance computing (HPC). Traditionally, GPUs have been used by those who want to do high-end graphics, such as product design/CAD in VDI workloads on vSphere. We place our focus here specifically on the ML and HPC areas and bring together all the information on setting up your GPUs on virtual machines on vSphere, along with a summary of the key performance information into one GPUs for Machine Learning on VMware – Learning Guide for your convenience.
The GPUs on VMware vSphere learning guide goes into the details of the architecture of a GPU and then into three different, though complementary usage scenarios. Firstly you may want to dedicate one or more full GPUs to one virtual machine only. You may subsequently be responding to a request from a data scientist or machine learning practitioner for more than one GPU, fully dedicated to a single VM, for their larger-scale training work. On the other hand, you may want to provide a sharing environment for your physical GPUs, such that more than one VM can take a part of a GPU for its purposes. The learning guide goes into details of setting up your GPUs in passthrough mode, (DirectPath I/O), in NVIDIA Grid vGPU style (for sharing of GPUs or multiple GPUs) and the BitFusion FlexDirect method for remote access to GPU-capable VMs over the network from other VMs that may not have a GPU assigned to them. As you can deduce here, you can combine these methods to make the best use of your GPUs in the virtualized environment in very innovative ways. Hope you enjoy using this GPU learning guide for VMware vSphere. You can learn a lot more about machine learning and high performance computing on vSphere here.
