Search form

Elastic Cloud Computing at DreamWorks with Cerelink

Read how a New Mexico tech company is conquering the holy grail of CG rendering.

Check out Megamind trailers and clips at AWNtv!

Having the ability to quickly scale compute needs to meet high-quality production demands is one of the holy grails in CG rendering. Images courtesy of DreamWorks Animation.

It's no secret that cloud computing was discussed at SIGGRAPH as the next holy grail. The real question has been getting it ready for prime time and convincing studios that it's secure. Well, earlier this year one of the initial breakthroughs came as a result of the five-year remote rendering agreement between New Mexico-based Cerelink and DreamWorks Animation.

That's because during the past year DreamWorks rendered portions of Shrek Forever After, How to Train Your Dragon and Megamind on Cerelink servers in New Mexico. In fact, using a private computer cluster, or "cloud," DreamWorks Animation's campus in Glendale connected to a Cerelink operated data center in Rio Rancho to perform nearly 2,000,000 render hours for the Oscar-contending How to Train Your Dragon.

This not only achieved cheaper power and access to competitively priced ultra fast broadband networks, but also a cost in savings as well as a 25% cash-back rebate because of the New Mexico tax breaks.

"The experiment with our remote rendering system has been successful and it's exciting for us to be part of the DreamWorks production system," suggests Dr. Rod Sanchez, Cerelink's president and chief strategy officer. "We want to replicate this service with every other animation or CG production studio out there. We've talked to half a dozen other major studios and have technical proof of concept projects with three of them. But our goal is to be the dominant provider of render services in the U.S."

Nearly 2,000,000 render hours of cloud computing were performed securely on How to Train Your Dragon.

In a prepared statement, DreamWorks Animation CTO Ed Leonard adds, "Our first collaboration with Cerelink on How to Train Your Dragon was a huge success, as it is working proof that high-performance computing clouds are not only possible but practical. Having the ability to quickly scale compute needs to meet DreamWorks Animation's high-quality production demands is one of the holy grails in CG rendering. There is no doubt in my mind that this is the future of rendering."

The genesis started with the following proposition: How does Cerelink use technology for economic development in New Mexico. Since the founders were all Intel vets, they decided to put their technical skills to work and decided to focus on the rendering of CG content for studios. They deployed $15 million in computer equipment for the DreamWorks experiment and are now ready to expand their capabilities.

"Right now we have two operating platforms just under 5,000 cores, the majority of which are Westmere core processors on HP Blade," explains Cerelink CEO James Ellington. "So we're in the early stages of providing diversity -- physical, network, security protection, which all of the studios want."

"Every studio has a different production protocol and so our proof of concept intent with three other studios is to identify what the elements of their production pipeline are and how we can refine those elements to take advantage of remote rendering," Sanchez adds.

Cerelink hopes to become the industry standard for cloud computing.

The results of their findings with DreamWorks and the other studios will hopefully lead to a standardized protocol that all of them can use in the future. "We were a small but integral part of the production," Ellington reports. "After it was validated last March, there was a ramp up to have a more significant role in the production of DreamWorks movies. We've increased our capacity over the last two months to about three-fold. Today we have just under 5,000 cores.

"What we were able to learn is that the data links we've experienced between New Mexico and Los Angeles is really small. Data is transmitted with an IP sensitivity so that the IP is secure. It's about 17 milliseconds and what that means is the time it takes a data package to travel from point A to point B is that it's so small that it's considered by the animator to be real time. That's significant because if the data latency is too long, the production pipeline or the methodology for which they produce the animation begins to breakdown. And one important factor in the test was the demonstration that the data latency was under the tolerance of about 30 milliseconds."

Cerelink, which intends on working with vfx studios as well, considers its remote render capacity on a par with DreamWorks' onsite render capacity. "They have a scheduler that transmits our render jobs from one location to another, depending on capacity load in each of those locations; it's not outsourcing the rendering: it's an extension of their capacity on campus and treated exactly the same as their computing assets would be treated," Ellington continues.

"What it means is that we were able to demonstrate that remote rendering is not only possible from a technical standpoint but also efficient. Moreover, while this makes business sense, the lower cost of power, the film incentives in New Mexico and the overall cost of doing business in New Mexico vs. California are critical to the bottom line of the production.

"What sets us apart is that, as far as we know, we're the only company that has done this a high-profile client and two high-profile feature films."

Bill Desowitz is senior editor of AWN & VFXWorld.

randomness