NVIDIA GRID, Tesla M10 and VMware ESXi - Sizing Tips

I am planning a M10 PoC with our VMware ESXi stack. In my Googling, it’s mentioned that only 4 VMs are supported per M10 card. Is that with pass-through only or with sharing the vGPU with GRID among many RDSH servers?

We have multiple RDSH servers in my environment (~50) and we’re potentially looking to share the vGPU cycles among the VMs. Before we purchase the M10 cards, I need to capacity plan, and thus looking for sizing guide or tips.

Any help is appreciated.

You should be planning to use vGPU, not Passthrough. vGPU is a lot easier to manage and it’s where all the features are added.

Each M10 has 4x physical GPUs on it. With RDS it’s best practice to assign a full 8GB vGPU to each VM (in your case, you’ll be using vApps licensing so you’ll assign an 8A Profile to each RDS VM). So that’s 4 RDS VMs per M10 and if you have a pair of M10s installed that’s 8x VMs per Host. Multiply that by however many users you typically get per RDS VM (typically 20 - 30 depending on Apps and usage), and there’s your capacity planning. If you aim at 160 users per physical server (assuming 20 per RDS VM), anything more than that is a bonus. Obviously CPU, RAM and Storage all make a difference to density as well …

Don’t forget to set the “Use the Hardware default graphics adapter for all Remote Desktop Services sessions” by using a GPO located at: Local Computer Policy > Computer Configuration > Administrative Templates > Windows Components > Remote Desktop Services > Remote Desktop Session Host > Remote Session Environment.

Hope that helps!


Thank you for the awesome reply Ben, your information certainly helps me with my planning!