Nvidia VGX may redefine the datacenter VDI

When you mention Nvidia to the average computer user, they think graphics cards. Gaming oriented folks will comment on their battle to the death with AMD as the two companies that dominate the graphics business fight back and forth for dominance in the desktop gaming market. Mobile device users may be aware of their Tegra line of mobile CPUs that have been pushing new products forward with two and four core designs. And a very small subset of the computing community will be aware of their presence in the high-performance computing world, where their Tesla line of GPUs, which enable hybrid supercomputing that is energy efficient and high performance.

But now Nvidia is taking their latest GPU architecture, AKA Kepler, solidly into the mainstream datacenter world by offering GPU virtualization for the virtual desktop. Their VGX platform can allow datacenters to deliver workstation class graphics to a VDI, a place where GPU accelerated graphics, needed for most design and simulation tools, simply wasn't available.  And this technology isn't just for that limited market; by virtualizing the GPU, Nvidia is able to deploy high-performance graphics to any VDI user, enabling foster graphics performance for common knowledge worker applications, as well a GPU intensive tasks.

The technology is comprised of three major components. The Nvida VGX board, which is the underlying hardware that utilizes the Kepler-based GPUs to deliver the graphics processing power where IT decides that it needs to go by making use of the low-latency remote display capability that is part of the Kepler architecture.

Second is the VGX Hypervisor. This allows a VM client with the Nvida driver to talk directly to the hardware, allowing each VM access to the GPU capabilities of the Kepler architecture and offloading video tasks from the VM CPU to the VM GOPU, improving overall performance of each system and increasing capacity of the VDI provider by freeing up basic CPU cycles that would otherwise be handling the display for each VDI client.

VGX Architecture

Last is the User Selectable Machines control for IT. The administrator is able to define the video profile of each VDI client, allowing them to be configured as a standard high-performance video client, a GPU accelerated desktop,  a full implementation of the Nvidia Quadro architecture, which supports professional graphics APIs that are used by the majority of workstation-specific applications and even  the NVS implementation used by those who need multi-monitor support with full multimedia functionality. This level of control on a per-user basis allows for the greatest efficiency in deploying a true virtual PC VDI experience to all users

If the Nvida VGX model is able to live up to all of these claims in extended real world use, the only limitation on VDI clients will become the bandwidth available between the datacenter and the desktop.