NVIDIA Launches Tesla M10 Enterprise GPU For VDI & Updates To GRID Virtualization

Enterprises that deploy large-scale virtual desktops and environments that require some graphics compute on the side will soon find a new compute card available: NVIDIA’s Tesla M10. While server-side workstation graphics deployments to virtual environments is covered by the Tesla M6 and M60, the new M10 is about user density, rather than raw compute. That doesn’t mean the card is underpowered, though.

In virtualized systems, be they apps deployed through XenApps, or Virtual Desktop Infrastructure (VDI), there is a growing demand for graphics capabilities by users. This demand for graphics compute is not just about professional tools such as Photoshop and various design tools, but it now entails industry-standard applications which make use of GPU acceleration, such as web browsers, office documentation, and humble video players – all making use of DirectX and OpenGL. Even modern desktop operating systems use GPUs for the graphical interfaces (as hard as it seems to believe, we didn’t always use the GPU for desktop environments).

There is now an expectation from software to have access to GPUs, which proves problematic when running dozens of virtual machines and user sessions on a server. According to SysTrack Community data provided by Lakeside Software, the use of GPU acceleration from apps has doubled over the last 5 years, which will continue to grow at a more rapid pace as virtual environments become more common.

NVIDIA has offered up a solution to this GPU acceleration demand in the virtual space with something called GRID, which allows for certain compute cards to be used concurrently by users across multiple virtual environments. While the Tesla M6 and M60 were meant to offer high performance to a small group of users, the new Tesla M10 is meant to provide ‘basic’ graphics compute to dozens of environments simultaneously.

Each Tesla M10 can provide support for up to 64 concurrent users, or 128 users per server with a dual GPU configuration. The way this is achieved is through a rather intriguing setup of a quad-GPU card (640 CUDA cores per GPU for a total of 2560), each of which has 8GB of GDDR5 available, for a total of 32GB. This allows each user access to 512MB of GPU RAM, which for the vast majority of cases, is plenty for office/desktop and some light design productivity. If a particular user’s needs are greater, then NVIDIA supports profiles within GRID to give that environment a greater share of the GPU. For graphics and CAD design tasks, NVIDIA still recommends the Tesla M6 and M60, as these are more geared to workstation needs.

GRID, the software and management stack that sits on top of the Tesla hardware, also got some improvements this April, as well as some tweaks to the licensing options. First, it allows for dual M10 GPUs to support 128 users, over a broader choice of software. NVIDIA Virtual Applications (vApps) and NVIDIA Virtual PC (vPC) can provide native Windows desktop environments and application support, with updated pricing of $2 per user, per month over a 3 year contract for vApps, and $6 per user for vPC for the same time frame.

NVIDIA’s new Tesla M10 will become available this August, with final pricing to be determined at that time.