GTC 2013: 8 to 16 Virtual Machines, Hosted in NVIDIA Grid VCA
Mountain climbers know Piz Daint, measuring 9,700 feet, as part of Switzerland’s snow-dusted Ortler Alps. Researchers and supercomputer nerds, however, know another Piz Daint, installed inside the Swiss National Supercomputing Center (abbreviated as CSCS in Swiss). The center is a unit of the Swiss Federal Institute of Technology in Zurich, where Albert Einstein once studied. Since supercomputers are used for, among other things, accurate weather prediction, the micro-climates of the Piz Daint in the Alps could very well be computed on the Piz Daint at the CSCS.
The supercomputer is a Cray XC30 system. Its current performance is listed as 216 TFlops, according to Top 500 Supercomputers. It’s the largest supercomputing giant Cray has assembled and delivered to date. But it’s about to get faster. When it’s retrofitted with Kepler GPUs, its speed will go up to 1 PFlops (1,000 trillion floating point operations per sec), announced NVIDIA. By early 2004, the Piz Daint will become “the fastest GPU accelerator-based scientific supercomputer in Europe,” NVIDIA noted.
NVIDIA Grid VCA: 8 to 16 Virtual Machines in a Box
But for those who need of GPU-acceleration on a more modest scale, NVIDIA CEO Jen-Hsun Huang unveiled a new product, dubbed NVIDIA Grid VCA (visual computing appliance) today during his keynote address to the audience at the annual GPU Technology Conference (GTC). In NVIDIA’s words, the Grid VCA is “a powerful GPU-based system which runs complex applications such as those from Adobe, Autodesk and Dassault Systemes, and sends their graphics output over the network to be displayed on a client computer.”
Priced beginning at $24,900, the appliance — literally a box housing a set of GPUs — can support anywhere from 8 to 16 users. (The hardware can be configured with 8 or 16 GPUs, enabling you to assign one GPU to each virtual machine.) It could become the solution for businesses looking to provide virtual machines with powerful graphics, accessible remotely.
Cloud-Hosted CAD, Accelerated by GPU
For a presentation, Gian Paolo Bassi, SolidWorks‘ VP of R&D, joined NVIDIA’s Huang on stage. The joined demonstration showed SolidWorks mechanical modeling software running on a remote virtual machine, hosted on NVIDIA Grid VCA.
With such a solution, Windows-based SolidWorks could be remotely delivered to engineers and designers on Mac OS, for example. (SolidWorks on Mac is on the wish-list of many. For more, read about the uproar resulting from a rumored Mac version in the past.) Similarly, workstation-level performance could be delivered to lightweight mobile devices, provided the bandwidth is wide enough to accommodate the data traffic.
SolidWorks’ parent company Dassault Systemes has also done additional work to add GPU-acceleration to its flagship high-end modeler, CATIA. The latest version, CATIA V6, includes CATIA Live Rendering, an interactive display mode that allows you to work with large assemblies in fully-rendered, ray-traced state. CATIA Live Rendering is powered by iray rendering technology from NVIDIA. In Maximus-class workstations, CATIA’s rendering workload could be distributed on as many as three GPUs.
CaaS (CAD as a Service)
A new breed of companies, like MAINFRAME2 (displaying its products at GTC), are emerging with an intriguing proposition. Suppose your company develops and markets 3D modeling software, traditionally sold as a program to install and run on a workstation. MAINFRAME2 can step in and provide you with the necessary hosting, IT setup, billing, analytics, and load balancing capacity to offer your software as an on-demand cloud-hosted solution. According to the company, “MAINFRAME2-powered apps work on all PCs, Macs, and even tablets. These apps are delivered as interactive streaming video and require minimal local resources. ” Nikola Bozinovic, founder of Mainframe2, is not prepared to name names, but said he’s in discussion with several major CAD developers to explore SaaS offerings (to be more precise, CaaS, standing for CAD as a Service).
High performance computing, previously the exclusive domain of universities and academic institutions with large budgets, is getting more affordable, driven by the parallel processing power inherent in multicore CPUs and GPUs. But the hardware required, clusters equipped with hundreds or thousands of CPUs and GPUs, is not a justifiable investment for small and midsized businesses that only need access to HPC occasionally. By the same token, a business may not be able to justify equipping every engineer or designer in its workforce with a high-powered workstation. In that case, delivering such a machine on-demand to those who need it is a better option. The virtualization technology in NVIDIA’s Kepler-brand GPUs and its new product Grid VCA point to a future where more users will log on to a virtual machine instead of a real machine. New companies like MAINFRAME2 are poised to harvest the trend, as the middlemen and middleware providers to turn popular software titles into SaaS offerings.
(Note: DE is a GTC media partner.)