One of the key themes as SC09 this year is the rise of the 3D Internet (the mass adoption of 3D imagery and virtual spaces on websites). In fact, no less a figure than Intel CTO Justin Rattner predicted in an interview with insideHPC that this could be how HPC “goes consumer,” since all that 3D-ness will need some hefty computation and rendering to get out to users at 60 frames per second.
But other than AMD’s Fusion Cloud PR-ware that wasn’t, there hasn’t been a lot of major product announcements yet. NVIDIA is hoping to change that with yesterday’s announcement of its RealityServer solution.
The solution is a combination of NVIDIA hardware and mental images’ (yes, it’s supposed to be lowercase) software. The hardware portion brands the previous preconfigured workstation/personal supercomputer spec into what is being called the Tesla RS server. The software includes version 3 of mental images’ RealityServer software and a new kick-ass CUDA-based ray tracer called iray (no caps at mental images).
“This is one giant leap closer to the goal of real-time photorealistic visual computing for the masses,” said Dan Vivoli, senior vice president, NVIDIA. “mental images fully embraced the concept of GPU co-processing to enable Interactive photorealism anywhere, any time – something that was science fiction just yesterday.”
The NVIDIA Teslas do the computation, iray makes all the pretty images, and RealityServer sits in the middle and connects users — accessing the service from iPhones, netbooks, laptops, and the like — to the images (jpegs) generated by the ray tracer. In a conversation about the announcement, Sumit Gupta made it clear that this solution is about the graphics. NVIDIA’s other ray tracer, Optix, is more of a general purpose ray tracing engine, and is being used in a variety of scientific and military applications (acoustic design, radar signatures, and so on).
NVIDIA sees themselves as a provider of infrastructure that other companies (Facebook, cloud services providers, and so on) will deploy to offer added value for their customers. There isn’t a huge HPC angle here, although the technology could certainly make it easier for data visualization centers to get highly detailed images out to their users. I had a personal demo of the software running in realtime and the image quality was really remarkable, and the response was near real time even in the remote application sharing environment I was being briefed through.
VizWorld takes an more in depth look at the announcement here. Availability? End of November.