NVIDIA reveals GPUs for blade servers, Linux desktop support

VDI-focussed for now, but the street finds its own use for things

VMworld 2015 NVIDIA has announced the second version of its Grid desktop virtualisation software, complete with a pair of GPUs for blade servers.

NVIDIA is pitching GRID as a hardware offering tuned to the needs of graphically-demanding desktop virtualisation (VDI) workloads. If that sounds a bit exotic, consider environments like the resources industry, where on-site engineers need CAD and modelling tools, but miners are loathe to deploy desktops in the remote sites where stuff gets dug out of the ground. VDI works a treat in such spots.

Hitherto, NVIDIA's “Kepler” GPUs for VDI have been designed for use with rack-mounted servers.

The new Tesla models, described in the table below, are now ready for use in blade servers from a range of vendors. NVIDIA worked with HP in the past for this trick, but has now brought Cisco, Dell and Lenovo into the fold.

Naturally, Citrix and VMware are also all over this release, so that Horizon, vSphere, XenApp, XenDesktop and XenServer all play nice.

Another addition is support for Linux desktops. That gesture, coming as it does on top of Citrix and VMware each announcing their support for penguin-powered virtual PCs, means this really does look to be the year of Linux on the (virtual) desktop.

	Tesla M60	Tesla M6
GPU	Dual High-end Maxwell	Single High-end Maxwell
CUDA Cores	4096	1536
Memory Size	16 GB GDDR5	8 GB GDDR5
H.264 1080p30 streams	36	18
GRID vGPU CCU	2 / 4 / 8 / 16 / 32	1 / 2 / 4 / 8 / 16
Form Factor	PCIe 3.0 Dual Slot	MXM
Power	240W / 300W (225W opt)	100W (75W opt)
Thermal	active / passive	bare board

Don't get all excited about GPUs in blades representing the possibility of a new age of super-dense GPU density: blade servers keep their I/O options lean so you can't get the high CPU:GPU ratios you'd find in a high-performance computing scenario. The company does, however, feel it won't be long before folks investigate the new Tesla GPUs for jobs other than VDI.

For those who stick to the VDI script, NVIDIA is talking up doubled density - up to 128 users per server – and doubled speed for virtual desktops with the new release. ®