Damien Fagnou

MPC in London Will Use Nvidia's Iray VCA for VFX Pre-Vis

Damien Fagnou, a stereographer and digital supervisor at London's MPC who has worked on films like the Harry Potter franchise and Prometheus, was recently made the company's global head of VFX operations after a short tenure as global head of software. One of his first tasks has been to oversee the continued development of an impressive pre-visualization pipeline powered by GPUs. Working closely with Nvidia, Fagnou and his team have used the Fabric Engine development platform and an OptiX Ray Tracing Engine-based renderer to bring complex scenes to life during pre-vis.

Toward the end of a talk he gave about this pipeline-in-progress at Nvidia's GPU Technology Conference last week, Fagnou mentioned that MPC was also looking at adding a bunch of Nvidia's just announced $50,000 GPU-based rendering appliances, the Iray VCA, to the mix. I was intrigued, especially since Iray VCA is primarily aimed at CAD designers in the automotive industry. (Honda in Japan is an early adopter.)

Iray is a so-called unbiased renderer found in software like 3ds Max and CAD software like Catia. Unbiased renderers are exceptionally accurate when interpreting every reflection, shadow and angle seen in the physical world, making them ideal for rendering globally illuminated, shiny, complex mechanical products like a car—but not so ideal for rendering scene upon scene of imagined worlds in a film. Iray also does not yet render hair and fur, another reason visual effects artists regularly choose other, biased renderers like Iray's sibling mental ray, which is adept at correctly rendering physical lighting but is also much more flexible (read programmable) and more efficient when it comes to rendering effects that defy the physics of the natural world.

As I learned from Fagnou after his talk, Iray's unbiased nature didn't deter him and his team from immediately considering Iray VCA's potential. Rendering speed, performance and real-time interaction with scenes are critical components of previs and virtual production, and with the appliance's 10 Gig-E and InfiniBand connections, a VFX studio could easily cluster several Iray VCAs together for even more power. Built-in cluster-management software tells each Iray VCA where to push the GPU horsepower as needed. Pair that with Nvidia's GRID VCA, an appliance for connecting renderers to the GPU cloud that the company rolled out last year, and you're now talking about a very powerful way to render versions of scenes in a short amount of time.

"We're not talking about final, fully textured renders here but about highly interactive, complex previews," Fagnou further explained to me. "But that's incredibly important. Say the director wants to drop by to reference several scenes and we've got to ramp up really quickly, on a large scale, to give him a full, interactive look at the scene in progress, one he can move around in virtually in real time. A few of these appliances strung together to create a larger virtual machine could let us do that, pushing graphics processing to where we needed it, really really fast."

Fagnou said MPC artists, who just finished work on the upcoming Godzilla, are already making intensive GPU use of one or more Quadro K6000s or Tesla K40s on their workstations. But it is the distributed and clustered power of the GPU render appliances, he says, that could open up many more cost-effective possibilities during regular workflows. "We are looking at all of it right now, starting with previsualization. Who knows what we'll discover we can do with it as we work closely with Nvidia? It's very exciting."

VP of Marketing Greg Estes, the general manager of Nvidia's media and entertainment (M&E) business, admits it is only a matter of time before Nvidia figures out how to expand the Iray appliance model to other renderers across a VFX pipeline. M&E is now, after all, Nvidia's second-largest market. As luck would have it, I sat next to mental ray's software product manager, Steffen Roemer, on the plane from San Jose to Los Angeles following the conference. "Was mental ray the obvious next step in Nvidia's GPU appliance rollout?" I asked him. Roemer is based in Berlin, where Iray and mental ray are developed (Nvidia acquired the renderers' parent company, Mental Images, in 2007, renaming it Nvidia Advanced Rendering Center, or Nvidia ARC). He confirmed that his team is investigating how to port the GPU appliance concept to mental ray, though he couldn't say exactly how long the roadmap will be, or how difficult the task.

Nvidia rolled out a great number of GPU-based technology solutions at GTC that have enormous potential for visual-effects studios. Nvidia's cloud-based GRID and the The VMware Horizon DaaS (desktop-as-a-service) platform, a new partnership also announced last week, will bring complete virtual workspaces, including desktops and any number of 3D software applications like Maya and Max, to users via the public cloud on a subsription basis. No more pimping out entire workstations every time a freelance or full-time artist is hired, or even worries about security issues. On these virtual desktops, users never have downloadable access to what they are working on. The DaaS Platform is only on Windows at the moment, but once virtual desktops can handle Linux, there's no telling how VFX studios will build, manage and maintain their geographically distributed artist teams in the future, perhaps even investing more in the artists themselves.