Hopper Architecture vGPU Types

NVIDIA Hopper GPUs target large-scale AI, HPC, and analytics in virtualized infrastructure. Hopper adds a dedicated Transformer Engine, fourth-generation Tensor Cores, FP8 compute on supported SKUs, NVLink 4.0 where the board provides it, and high-bandwidth memory (HBM3 or HBM3e depending on product).

MIG-backed vGPU types map to fixed GPU instance profiles (slices and framebuffer per profile); time-sliced types share one physical GPU with per-vGPU framebuffer caps. Open a SKU family below for maximum vGPUs per GPU, slice counts, and display limits.

See also

• MIG-Backed vGPU - MIG-backed vGPU concepts and constraints

• Scheduling Policies - vGPU scheduling policies for time-sliced types

• Configuration - Host MIG provisioning and guest reconfiguration steps

Browse Hopper vGPU types by SKU family:

Table 86 Hopper SKU families

Family	Boards
Hopper H800 vGPU Types	H800 NVL, H800 PCIe 80 GB, H800 SXM5 80 GB
Hopper H200 vGPU Types	H200 NVL, H200 SXM5 141 GB
Hopper H100 vGPU Types	H100 NVL, H100 PCIe / SXM5 94 GB / 80 GB / 64 GB
Hopper H20 vGPU Types	H20 SXM5 141 GB, H20 SXM5 96 GB