One Size Does Not Fit All: vGPU Profiles

By CMD Technology Group

3 min read

In our previous HDX blog post, we explored the intersection of Citrix Virtual Desktops and HDX 3D Pro with NVIDIA Virtual GPU (vGPU). 

Where HDX 3D Pro affords 3D graphics users the power of a GPU in their virtual desktop, NVIDIA vGPU allows multiple virtual machines simultaneous access to a single physical GPU. 

Can Performance Sacrifice Be Avoided?

By sharing a GPU among multiple workloads, questions arise around scalability and performance. How do we achieve cost-effective scalability without sacrificing the graphics performance of our VDI sessions?

The Role of Profiles

With advancements in virtual desktop hosting, administrators have been able to scale performance while customizing user experience. NVIDIA vGPU profiles provide an opportunity to address both. Profiles serve as a framework for allocating fixed share resources to each virtual machine (VM) from the total pool of memory available – all based on user needs.  

Profiles Dependency On GPU and License Edition

As GPUs have different features and resources, they also support different profiles types. The chosen GPU dictates the profiles available. The selected NVIDIA software license edition also unlocks specific profiles.

Determining Profile

Type

When determining a vGPU profile type, there are a few key questions to ask of your use case:

How much frame buffer (memory) does each desktop require?
What are the maximum resolution limits per display head?
How graphic intensive is their work?
How many display heads (monitors) does each user need?
What kind of work are users doing?

Profile Naming Conventions

NVIDIA vGPU profile names indicate answers to the above questions. Let’s take a closer look at the Tesla series M6 GPU accelerator. Examining the M61B profile, we know:

  • This profile is available to an M6 graphics card
  • This profile offers 1 GB per vGPUThe intended workload class are Power (B) users and requires a vPC edition license
  • Q-series: Designers and power users; vDWS
  • B-series: Power users; vPC
  • A-series: Virtual applications users; vApps


Profile Sizing

Profiles provide structure for how a GPU is shared among multiple VMs. Profiles directly impact the number of machines each host can scale to. Sizing exercises aid in planning your deployment.

Let’s take the example using the M6 further.

Sizing Example:

M6An M6 graphics card has a total of 8 GB of frame buffer and one physical GPU per board.

What are some ways we can allocate 8 GB across machines? 

  • 1 VM with 8 GB
  • 2 VMS with 4 GB
  • 4 VMS with 2 GB
  • 8 VMS with 1 GB
  • 16 VMs with 512 MB 

Each combination correlates to a vGPU profile(s)*

  • 8Q: 8 GB per vGPU
  • 4Q: 4 GB per VGPU
  • 2B: 2 GB per VGPU
  • 1B: 1 GB per VGPU
  • 0B: 512 MB per VGPU

*profiles are further classified by the number of supported monitors and maximum resolution

Sizing

Example:

M10

The M10 has a total of 32 GB of frame buffer and 4 physical GPUs per board.

Each of the four GPUs has 8 GB of frame buffer available to it.


If we apply an M10-2Q profile, we allocate 2 GB of memory per vGPU.


So, if we have 8 GB available to each GPU and we allocate 2 GB to each VM, each GPU can power 4 VMs.


We can, therefore, accommodate 4 VMs per GPU


Furthermore, if we have 4 VMs per GPU and 4 GPUs per M10 card, we have can fit 16 VMs per board.


We are therefore restricted to 16 VMs per host.

Mixing Profiles

As mentioned, the M10 graphics card has a total of 32 GB of frame buffer. It contains 4 GPUs with 8 GB of frame buffer available to each GPU. 

Profile types can vary per board but must stay the same per GPU. For example, one GPU may apply a 1Q profile, while another a 4Q profile. In this scenario, GPU (0) fits 8 VMs with 1 vGPU each. GPU (1) accommodates 2 VMs with 4 vGPU each. A different profile could be applied to GPU (3) and GPU (4).

When sharing 32 GB of frame buffer among GPUs on a single physical M10 card, frame buffer must be allocated in divisible factors (i.e., 1, 2, 4, 8) across GPUs with one profile per GPU. Mixing profiles allows administrators the flexibility to serve multiple use cases with a single graphics card. 

Scalable Performance

When sharing a single GPU among multiple machines, vGPU profiles are critical. The number of machines each host can accommodate is directly affected by the vGPU profile, and in turn, the scale of desktop deployment. 

With so many advantages today, graphics-rich VDI should be commonplace. NVIDIA vGPU and HDX 3D Pro offer versatility and performance, without compromising user density. 


If you have a question about any of our solutions or any feedback you’d like to share, contact us. We would love to hear from you!

afernandez@cmdtg.com | (407) 442-0265

Get in touch

You can email us at afernandez@cmdtg.com

Give us a call at 1-800-806-4173

Or contact us using the form below