With an eye on the evolving BYOD and VDI markets, Nvidia today unveiled the Nvidia VGX platform, designed to virtualize server-side GPUs to deliver graphics-intensive applications and multimedia content to any device. The announcement plays into the company's broader vision of a GPU-accelerated cloud, where users can access high-end graphical applications and content anywhere in the cloud from any device.
The ambitious idea behind the VGX platform is to enable power users and designers to access their local desktop and all their heavy-duty apps -- such as apps for 3D graphic design or medical imaging -- from a thin client, a laptop, a tablet, or a smartphone without any performance hiccups.
"The way VDI works today, the CPU is basically doing all the rendering," said Jeff Brown, vice president and general manager for Nvidia's Professional Solutions Group. "Inherently, CPUs aren't intended to do graphic processing. It's a very computationally challenging task."
The VGX platform includes the VGX board, dubbed Kepler, which is built to support hardware virtualization and low-latency remote display. The board includes four GPUs and 16GB of frame buffer. There's also the VGX GPU hypervisor, which sits within a standard hypervisor from the likes of Citrix or VMware and intercepts graphics commands for the VGX board to handle. The board then streamlines the delivery of the graphics-heavy processes directly out of the frame buffer, eliminating intermediate, software-intensive rendering that the CPU would otherwise perform.
"Microsoft Remote Desktop tries to deliver graphics this way," said Brown. "It's a stop-gap solution that gives you partial graphics [support] up to DirectX 9."
According to Brown, the VGX platform is capable of fully supporting applications build on the most up-to-date versions of OpenGL and Direct X 11.
There are obstacles to Nvidia's bid to unite its GPUs with the virtualized world. First, the VGX Platform is best suited for private LAN environments, meaning a user won't enjoy the local-desktop experience on, say, public Wi-Fi.
"Some of the early data points show that at 4Mbps to 6Mbps, you can have a great experience using these virtual desktops," said Brown. "As you get to 4G, it becomes practical [to deliver over mobile wireless]. It's not where we are today, but it's where we want to go."
Second, the success of the platform depends on support from vendors on the server and virtualization side. Here, Nvidia looks to be in good shape. Dell, HP, IBM, and Supermicro all offer high-volume servers enabled for NVidia's GPUs -- including Dell's R720, HP's, DL 380 from HP, and IBM iDataPlex. Those same servers can be used for the VGX line, according to the company. What's more, Cisco today announced that it will enable GPUs in its line of servers. On the virtualization side, Nvidia has support from Citrix, Microsoft, Xen, and VMware.
The company also needs buy-in from developers to whip out GPU-intensive applications. To that end, the company unveiled this week Nsight Eclipse Edition, an IDE for developing GPU-accelerated applications on Linux and Mac OS. The offering includes debugging and profiling tools designed to help programmers build apps optimized for CPUs and GPUs on Compute Unified Device Architecture (CUDA), the company's parallel computing architecture for graphics processing.
Nvidia also announced an updated version of Nsight, Visual Studio Edition, geared toward Windows developers. Enhancements include local single GPU debugging, which enables developers to debug CUDA C and C++ code natively on the hardware with any system equipped with any CUDA 1.1 or higher capable GPU. Other features include performance improvements to the frame profiler and debugger, and support for DirectX 9 frame debugging, frame profiling and analysis.
This story, "Nvidia unveils virtualized GPU with eye on graphics-intensive cloud," was originally published at InfoWorld.com. Get the first word on what the important tech news really means with the InfoWorld Tech Watch blog. For the latest developments in business technology news, follow InfoWorld.com on Twitter.