Hi everyone! I’m Derrick Isoka, a program manager on the RemoteFX team responsible for the virtual GPU component. We’ve received a lot of feedback and questions regarding the cards we recommend for the RemoteFX virtual graphics processing unit (vGPU). In this blog post we’ll share our recommendations to help you understand the options available to you, and most importantly to help you make a decision on the cards that you can consider as you deploy a VDI solution with RemoteFX vGPU.
RemoteFX GPU Requirements
Use of RemoteFX with GPU acceleration on Windows Server 2012 R2 requires a compatible graphics card. Most likely, the servers hosting the RemoteFX workloads will be located in a datacenter and as such, we recommend using passively cooled, server class graphics cards. However, it’s also acceptable to use a workstation card for testing on small deployments depending on your needs. At a minimum, the requirements for graphics cards to be used with RemoteFX are: • DirectX 11.0 or later • WDDM 1.2 driver or later Support in Windows Server 2012 R2 is provided for DX 11.0, DirectCompute, and C++ AMP. Most of the latest graphics cards will support OpenGL 4.0 and OpenCL 1.1 or later, but these APIs are currently unsupported by RemoteFX in Windows Server 2012 R2.RemoteFX-Compatible GPUs
The following list is not meant to be exhaustive but representative of the mainstream cards from NVIDIA and AMD.
Rank |
NVIDIA |
AMD |
Best |
NVIDIA Grid
|
|
Better |
|
|
Good |
|
Note: 1. Best: These are server class cards, designed and certified for VDI workloads by hardware vendors like NVIDIA and AMD. They target the best application performance, experience, and virtual machine densities. Some of the cards are particularly recommended for designer and engineering workloads (such as Autodesk Inventor or AutoCad). 2. Better: These are workstation class cards that provide acceptable performance and densities. They are especially capable cards for knowledge worker workloads (such as Microsoft Office or Internet Explorer). 3. Good: These are lower-end cards that provide acceptable densities knowledge worker workloads.
Notes on Performance and Scale
In addition to a GPU’s total memory and power consumption, the performance and scale of a VDI system is determined by a variety of additional factors such as storage speed, system memory speed, amount of system memory, number of CPU cores, NUMA implementation, and CPU clock frequency. We're completing some tests for a select set of the cards mentioned in this post, and will be sharing those results in a separate blog post.In conclusion, deploying a VDI solution requires assembling a lot of components. We hope this blog post helps in reducing the complexity and provides guidance in selecting the right GPU component to address the appropriate experience for your end users.