NVIDIA GRID has moved to a software model which allows the rapid deployment of new features and vGPU profiles to be rapidly released enabling new business models and needs. The vGPU profiles available are detailed in the product documentation within the “GRID VIRTUAL GPU – User Guide” that is in the product download.
GPU for office and business users
Historically it was 3D/CAD users who had previously been unable to virtualize such heavyweight applications who led the uptake of technologies such as vGPU. However an increasing number of traditional VDI users are now finding that user and application demands are driving a need for GPU acceleration. Whilst these users traditionally did not use GPU-intensive applications they are finding GPU’s increasingly useful:
- Users are using graphically richer apps more frequently – watching YouTube and training videos rather than read manuals, increasing their use of video conferencing software
- The virtualization vendors are adding GPU acceleration to the transmission protocols that encode sessions, meaning servers can reduce CPU load and increase scalability by providing GPU resource
- VMware have recently released Blast Extreme capable of leveraging NVIDIA hardware as an alternative to the PCoIP protocol they support
- Citrix have added Hardware encode to their Linux VDI offering
- AWS’ owned NICE DVC offers NVIDIA hardware support in their encoding protocols
- Multi-monitors and increasingly high-resolution end-points are becoming common. With users of office and 2D apps using 3 or 4 monitors CPU driven remoting often becomes sluggish and struggles. Again rather than the applications requiring acceleration. It is the demands of the protocols and vast arrays of pixels needed to fill those screens.
Dial “B” for business
The observant will have noticed recent additions to the vGPUs available, including profiles such as the M60-0B, M60-1B, and M6-0B and M6-1B.
These profiles fall under the NVIDIA GRID Virtual PC offering, reflecting their intended usage for office desktops rather than heavyweight CAD applications. They offer multi-monitor support on smaller framebuffer profiles. For example the M60-1B offers up to 4 monitors at 2560×1600. The expected use case is in a scenario where you won’t be using CAD rendering applications and so the GPU resource is available to drive the monitors. If you were to add heavier weight applications to this profile you would expect to see a performance degradation and a more appropriate product would be from the virtual workstation portfolio such as an M60-2Q.
NVIDIA GRID has three different editions that let you assign the right level of resources to each of your users; Virtual Applications, Virtual PC, and Virtual Workstation. – See more at: http://www.nvidia.com/object/grid-technology.html#sthash.WgucqXLL.dpuf
I found it quite hard to work out the connection between the editions and available specifications of the vGPU profiles. I hope to update this blog with a link to some better documentation soon. Feedback as to whether this is just me – welcome! [Update – 4th May 2016 – I’ve found the edition documentation online – details here: https://virtuallyvisual.wordpress.com/2016/05/04/dial-b-for-business-new-vgpu-profiles-info-online/]
Other references:
- Richard Hoffman recently did a really good community blog overviewing the GRID offering: https://vgpu.wordpress.com/2016/04/12/nvidia-grid-2-0-deep-dive-configuration/
- There is an interesting blog with some fascinating user comments on the Citrix blogs about how they are seeing office users increasingly adopt GPU: https://www.citrix.com/blogs/2015/12/03/gpu-for-the-masses-with-xenappxendesktop/
- vExpert Sean Massey’s take on how protocols leveraging GPU-acceleration could drive GPU for the VDI masses: It’s Time To Reconsider My Thoughts on GPUs in VDI…
- 4K support – we often get asked for a vGPU profile with a 1GB framebuffer that can support 4K monitors. A 4K monitor is a lot of pixels, and a 1GB simply isn’t sufficient. There’s a super Citrix CTX article, CTX201696 – Citrix XenDesktop and XenApp – Support for Monitors Including 4K Resolution and Multi-monitors. In which they do some maths on the sheer number of pixels involved with remoting protocols – see the section “Verifying Memory Used by the Workstation OS VDA”.
- Virtualisation pundits such as Brian Madden commenting on GPUs becoming mass market: http://www.brianmadden.com/opinion/GPUs-in-VDI-need-to-stop-being-optional
Virtually Visual 
Good article, Rachel. Thanks! You mentioned NVIDIA’s move to “software” to allow for rapid deployment of new features. I’ve heard a lot about that lately and I wonder from a technical standpoint, how that is achieved? Do the new GPU cards have a programmable module that allows for easier upgrades of features? Of course all of the previous cards had firmware but it sounds like there is something significantly different here.
Thanks again!
Richard
LikeLike