NVIDIA Introduces Virtualization and Remote Computing

Is there a studio executive anywhere who hasn't dreamed about being able to tuck away the workstations behind the scenes, so artists work with a light client? What about the possibility of people being able to look at and collaborate on the same material in different locations?

NVIDIA just opened up the possibility to this scenario and many others with the introduction of its NVIDIA VGX platform, which enables users access to a virtualized desktop with the graphics and GPU computing performance on any connected device. "With this platform, employees can now access a true cloud PC from any device -- thin client, laptop, tablet or smartphone -- regardless of its operating system, with the responsiveness previously only available on the workstation," says NVIDIA General Manager of the Professional Solutions Group Jeff Brown.



Jeff Brown, GM of the Professional Solutions Group at NVIDIA at the press QA on the new release of the VGX


"The trend is that more people want to be able to access all their data and applications anytime, anywhere on any device," he continues. "What's been missing has been the ability to deliver the experience people expect via the cloud. The missing link has been to be able to "virtualize" the GPU, which means to put a lot of users on a single GPU."

To do so, NVIDIA worked on three key technologies to create the VGX platform. First, the company redesigned the GPU; a new MMU (memory management unit) within the GPU effectively manages all the different requests coming in from different users. These NVIDIA VGX Boards are designed for hosting large numbers of users in an energy efficient way. The initial NVIDIA VGX board features four GPUs, each with 192 NVIDIA CUDA architecture cores and 4 GB of frame buffer. The boards are passively cooled and fit within existing server-based platforms. With support for low-latency remote display (which greatly reduces lag) and redesigned shader technology to deliver higher power efficiency, the VGX board is aimed at many users running hosted virtual desktops to share a single GPU.



Kepler GK110 Full chip block diagram


"One barrier to the cloud is the user experience of the perceived latency," says Brown. "We can reduce latency -- which is measured in milliseconds -- by half. An acceptable user experience today is about 50 to 100 milliseconds in lag. The current non-optimized remote approach is more in the 300 to 400 millisecond range. With what we've done, it feels now like a local experience."

The second NVIDIA move to make the virtualized GPU possible is the VGX GPU Hypervisor, a software layer that sits between the user and the hardware. "It sits on top of the GPU like a traffic cop, allowing access to virtualized GPU resources," explains Brown. "This allows multiple users to share common hardware and ensure virtual machines running on a single server have protected access to critical resources. A single server can now economically support more users while providing native graphics and GPU computing performance."

Third is NVIDIA User Selectable Machines, which allow the NVIDIA VGX platform deliver the advanced experience of professional GPUs to those requiring them; in other words, multiple types of users can be supported with a single server.



GTC 2012 Keynote: Jen-Hsun Huang with Grady Cofer of ILM


What does this mean for the media and entertainment world? Industrial Light + Magic Visual Effects Supervisor Grady Cofer [see video below] demonstrated how he can have two or three people look at and collaborate on a scene from Battleship.

"Different companies will have different needs," says NVIDIA's media and entertainment industry executive Greg Estes. "I think for the studios, the remoting part is the most interesting. They'll like the fact that they can back-rack the workstation and run all the applications they want with low latency. "

Still, how visual effects facilities and post houses might use the new NVIDIA cloud-based paradigm remains to be seen.

But we won't have to wait for long. Although no ink has dried on any agreements, NVIDIA's new offering will clearly be of great interest to any software vendor that's got a cloud strategy...as well as any facility that has an outpost in Asia, Vancouver or Australia. "You can imagine the interest that's going on now in beginning to work with us in our very early days," says Estes. "We're showing a vision we have and starting to announce the tangible pieces, but we're freely admitting that it's a long and exciting process. Many people are ready to hold up their hand and say, we're the first on the block to want to try this."

Brown points out that Citrix has already partnered with NVIDIA for both remote and virtualization applications. "We'll go into beta with virtualization in the fall," promises Brown.

As to when we'll see the first media and entertainment company take the plunge? "This is proof of concept phase," says Brown. "Even with the technology available, it'll be months before we start to see real deployments."

Stay tuned. NVIDIA has just introduced a paradigm that makes new ways of working feasible. It's a huge step in a direction that has long been the Holy Grail. This summer will show developments, and Creative COW will stay on top of them as NVIDIA begins to partner with companies in the M&E space.



GTC 2012 Keynote (Part 07): ILM's Grady Cofer on VFX and Kepler

---