Bitfusion FlexDirect can connect any virtual machine (VM), running any AI framework, to any GPU or partial GPUs, hosted by ESXi or bare-metal. These VMs can schedule AI/ML workloads and launch them on any compute server, regardless of whether there are local GPUs attached to the server. FlexDirect supports every CUDA enabled GPU as well as all GPU servers, with any density of GPUs.
We are now able to support a mixed environment of ESXi compute servers, connected to GPU farms which are bare-metal or part of vSphere. This provides the unparalleled flexibility to pull all of an organization’s GPU assets onto vSphere, whether they are installed with ESXi or not.
Together with our partners at VMworld, we will demonstrate a range of use cases. Some examples include:
- VM running an AI workload, connected via a remote VM on a GPU server; allowing live migration of VMs through vMotion
- VM running an AI workload, connected with a bare-metal GPU server; one or more GPUs can be attached to a single VM
- VM running an AI workload, connected with a remote partial GPU; any size can be requested, e.g. 30% of a physical GPU
- VM running an AI workload, connected with two remote physical GPUs, each physical GPU residing in a different GPU server
- Two VMs, running dissimilar AI frameworks, in two different physical compute servers, sharing one remote physical GPU
We are working with our customers and partners on more use cases, such as GPUaaS and shared GPUs pools for development and test. Make sure to connect and arrange time to see us at VMworld by sending email to .