LAS VEGAS -- When GPU virtualization comes up in the VDI conversation, most people think of supporting graphics-intensive applications such as CAD and Photoshop. Those resource-hungry applications need vGPU, but they're not the only apps that can benefit.
Applications such as Microsoft Office and web browsers as well as modern operating systems like Windows 10 all reap rewards from vGPU technology, said Citrix CTO Gunnar Berger in a session at Synergy 2016.
When IT runs PowerPoint without a GPU, for example, it uses 10% of the CPU and even spikes to 30% occasionally, according to a demo from Berger and David Cottingham, director of XenServer product management and partner engineering. With a GPU, the average is 5%, with occasional spikes to only 10%. Using a GPU results in more density on the server and a better user experience on the endpoint. This is a win-win situation for IT, which Berger pointed out doesn't come along all that often.
There are three main use cases for vGPU, according to the session: supporting published apps, making virtual desktops run as well as physical ones, and moving high-end workstations that support power users or resource-intensive applications to the data center.
Some of the most common verticals that use applications that require vGPU are education -- especially where students learn CAD -- and automotive, where engineers don't want to ship physical models all over the world. Other major markets for GPU and GPU virtualization are healthcare settings where 3D MRIs and high-tech X-rays are becoming more prevalent.
Even knowledge workers can benefit. Anyone who uses browsers, Skype for Business, GoTo Meeting or Windows 10 -- which uses DirectX and wants a GPU to run smoothly -- can increase server density and improve user experiences.
Choosing a vGPU option
The major players in the GPU market include NVIDIA, Intel and AMD. Each vendor handles GPU virtualization a little differently.
NVIDIA GRID uses software to carve a physical GPU into pieces. There's isolation between the vGPUs, so they don't bleed into one another or use up each other's resources.
Intel's GVT-g cards also virtualize using software, but the cards are GPU built into a CPU. They split into seven vGPUs, and there's no isolation between the parts.
AMD's approach to vGPU still carves up a physical GPU, but the virtualization is hardware based. The hardware breaks itself up, and the vGPUs are passed to the VMs; when they get to the hypervisor, it thinks it's seeing multiple AMD cards instead of several virtualized GPUs.
Which vGPU companies should pick is a personal decision. They should consider budget, the type of server they will snap GPUs into, and whether they want to use a software- or hardware-based approach to GPU virtualization. But businesses don't need to be deterred from using vGPU if they don't support CAD or similar applications. Knowledge workers using the most common applications and operating systems will notice the improved performance vGPU can bring.
Compare VMware and Citrix vGPU support
AMD gives NVIDIA GRID a run for its money
When to support virtual GPU technology
Before migrating to a XenDesktop farm, five things you should know