The Minisforum MS-A2 continues to be a popular and versatile platform for running VMware Cloud Foundation (VCF) 9.1 in a lab environment. While the MS-A2 may be small in size, it is surprisingly capable and can even be configured to support an NVIDIA RTX 4000 Ada (20GB VRAM) GPU, which I had used to deploy VMware Private AI Services (PAIS).
Although there are several half-height GPUs that can fit in an MS-A2, the overall selection remains relatively limited. For users looking for additional GPU flexibility, the MS-A2 supports an OCuLink half-height PCIe card, enabling the use of a much broader range of discrete GPUs.
Thanks to a fellow colleague who recently shared their success with me using this setup on VMware Cloud Foundation (VCF) 9.1, including GPU passthrough to vSphere Kubernetes Service (VKS) cluster for AI experimentation, below are the specific BOM details for those interested.
Requirements:
- PCIe 4.0 x4 OCuLink add-on card (includes OCuLink cable)
- Minisforum DEG1 eGPU to Oculink Dock
- ATX/SFX Power Supply is required
The ESX host should automatically recognize the eGPU once everything is powered up and simply toggle DirectPath Mode for enabling PCIe passthru for the GPU.
For those planning to use eGPU passthrough with VKS, you will need to create a custom VM Class that includes the eGPU passthrough device and override the VKS deployment specification to only apply the VM Class to the VKS Worker Node, which is what will consume the eGPU. If you forget to apply this override, the VKS deployment will fail as it will attempt to look for a GPU device for both the VKS Control Plane and Worker Node.
Thanks for the comment!