Everything you need to know about GPU virtualization
A comprehensive collection of articles, videos and more, hand-picked by our editors
IT shops that want to use VDI but have end users with graphic workload requirements can deliver virtual desktops...
with good performance if they add a GPU to the mix.
GPU options for virtual desktops have improved in recent years. Nvidia Corp.'s virtual GPU (vGPU) and other graphics-delivery technologies are used with success in Citrix virtual desktop infrastructures (VDI) today, with platform support expanding to VMware vSphere by 2015.
In addition, Intel plans to deliver the Xeon E3-1200, code-named Crystal Well, with a built-in GPU later this year.
Ruben Spruijt CTO of PQR
But IT administrators must know their workload requirements before they invest in a GPU technology, and whether their VDI platform supports the GPU power they need.
"If you have no clue about your requirements, you may as well just flip a coin," said Ruben Spruijt, CTO of PQR, an IT consultancy based in the Netherlands, during his GPU session at Citrix Synergy earlier this month.
VDI GPU support: who needs what?
Once end users' graphics uses are determined, IT can decide the best approach for delivering workloads.
That can mean dedicated GPU, a vGPU that delivers slices of a graphics processing unit to multiple end users, or GPU pass-through, which allows IT to assign physical GPUs to virtual machines.
In general, a vGPU gives similar performance to integrated video chipsets found in most corporate desktops and laptops, said Matt Kosht, an IT manager with a utility company based in Alaska.
"These are good for most applications a power user would use -- 2-D CAD/GIS, Google Earth/Maps," Kosht said. "A graphic artist or 3-D modeler would need -- [and] prefer -- the performance gained by getting a dedicated GPU for their virtual desktop."
A dedicated GPU provides performance closer to a low-to-mid-range dedicated 3-D graphics card, Kosht said.
A high-density vGPU or apps delivered with Citrix XenApp might be adequate for knowledge workers. Task workers with basic graphic dependencies in Windows 7 or 8 that use typical business apps could also benefit from a high-density vGPU, software rendering GPU or XenApp with GPU sharing, Spruijt said.
GPU support in Citrix, Microsoft, and VMware VDI
Once IT understands their users' needs and the type of graphics support to deliver, they must determine if their VDI platform supports the graphics requirements.
IT needs to know whether the VDI platform supports vGPU, GPU pass-through and software GPU and if there is USB redirection and 3-D mouse support. Graphics remoting API support for applications is also critical, Spruijt said.
Citrix XenDesktop provides the widest range of GPU support, including software 3-D graphics, pass-through GPU and, when paired with XenServer, it supports Nvidia's's vGPU. In addition, Citrix's HDX 3-D Pro works great on LAN, WAN and for mobile devices, according to Spruijt.
Microsoft Remote Desktop Session Host (RDSH) and Remote Desktop Virtualization Host (RDVH) with RemoteFX support are good enough for IT shops that don't want to pay for Citrix's premium products, Spruijt said. RDVH client OS supports soft 3-D graphics and GPU sharing for VDI, but does not support pass-through GPU or vGPU. RDSH supports soft 3-D graphics and GPU sharing for XenApp/RDSH.
However, Microsoft's RDSH user management focus is less than 500 users, and enterprises need scalability, Spruijt said. The lack of vGPU support, pass-through support and limited integration with secure access tools is also a problem.
VMware Horizon View GPU support options include soft 3-D graphics, pass-through GPU, GPU sharing for VDI and future vSphere support for Nvidia's vGPU. VMware's protocol provides CPU offloading with Teradici's PCoIP APEX card, but like RemoteFX, secure access tools for PCoIP are a challenge, Spruijt said.
Until Nvidia support comes, "the video card driver used on the virtual desktop could give a subpar performance versus a vendor-supplied driver that is tuned for their specific hardware and supports [the] latest DirectX and OpenGL APIs," Kosht said.
However, the GPU isn't the only indicator of performance.
"You still need to get those pixels to the client, which means relying on the network," Kosht said. "Most of the modern remoting protocols do a good job at optimizing, but if you have a high latency, high packet loss or low-bandwidth connection to the VDI server, they can only do so much before you get an unacceptable user experience."
IT pros must also check server hardware vendor support for GPUs, and the amount that is supported; a GPU extension kit may be necessary, said Spruijt, who recently co-authored a 3-D graphics for virtual desktop whitepaper.
They'd also be wise to ensure the 3-D applications provider supports delivery via VDI, and ascertain whether client-side devices are compatible. Thin clients may be limited in their decoding capacity, he said.
IT pros who want to try Nvidia's graphics acceleration can do so with the new GRID Test Drive program. Once registered, users have 24 hours to test it with remote desktops and graphics-rich applications in a virtualized environment. The Test Drive is available on Windows clients, and a version for Mac OS and Linux will be available soon.