Supported GPUs for GPU Partitioning in Windows Server 2025
Virtualization has transformed IT, enabling us to run multiple VM’s and OS’s on a single server. But for resource-intensive tasks like AI and machine learning, powerful graphics processing is essential. This is where Windows Server 2025’s GPU partitioning comes into play, allowing multiple virtual machines (VMs) to share a single GPU’s power, optimising usage and enhancing workload capacity.
What is GPU Partitioning?
With GPU partitioning, a single physical GPU can be split into multiple virtual GPUs (vGPUs), each assigned to different VMs. This enables simultaneous execution of resource-heavy tasks, like AI and ML workloads, all on a shared GPU—making it a game-changer for high-demand environments.
Supported GPUs
Currently only a handful of NVIDIA GPUs currently support partitioning with Windows Server 2025. Here’s a list of the compatible graphics cards supported for Windows Server 2025 for GPU Partitioning:
| GPU Model | Rough Cost (USD) | CUDA Cores | TF32 teraFLOPS or Tensor Cores | Memory (GB) | TDP (Watts) |
|---|---|---|---|---|---|
| NVIDIA A2 | £1300-1800 | 1280 | 40-60 | 16 | 40-60 |
| NVIDIA A10 | £2300+ | 8192 | 275-410 | 24 | 150 |
| NVIDIA A16 | £2700+ | 5120 (4x 1280) | 4x 40 Cores | 64 | 250 |
| NVIDIA A40 | £5100+ | 10,752 | 74.8 – 149.6 | 48 | 300 |
| NVIDIA L2 | Not out yet | n/a | 48.3 | 24 | TBD |
| NVIDIA L4 | £2500+ | n/a | 120 | 24 | 72 |
| NVIDIA L40 | £7500+ | 18176 | 568 | Gen 4 Cores | 48 | 300 |
| NVIDIA L40S | £9700+ | 18,176 | 366 | 48 | 350 |
Notes
- My pick would be the NVIDIA A16 currently offering what is basically 4 GPU’s on one card already making it ideal for partitioning.
- Details for some GPUs, especially newer models, are limited and may change as they become more widely available.
- Most of these cards are made for the enterprise market, so don’t go thinking you will suddenly be able to set up 4 gaming PC’s on one server and get good graphic results! Whilst it may be possible, these are designed more around tensor cores, useful for AI and deep learning than Cuda cores, which are more multipurpose.
Windows Server 2025’s GPU partitioning unlocks powerful capabilities for optimising hardware and running demanding workloads. While limited to specific NVIDIA GPUs, it’s a step forward for those looking to enhance their system’s efficiency and boost VM computational power. Understanding which GPUs work best for what workloads will become the next big learning curve!