Redshift Ramps up Render Speed & Efficiency at Outpost VFX

At the time Outpost set up the studio in 2013, one of the first tasks was evaluating the major renderers for potential adoption into its pipeline. During his 10 years of visual effects supervision experience, Duncan has experimented with many different systems and a huge variety of pipelines. Once he began running his own studio, however, buying into the right renderer became critical from both an artistic and a practical point of view.

Artist Time

“As a combined hardware and software cost, the Redshift biased rendering system turned out to be the most affordable option for us, not just in monetary terms but also in terms of our biggest cost, artist hours,” Duncan said. “Artist time is always a huge factor when working on render intensive projects and in this regard, Redshift made a significant difference.”

Biased renderers allow users to specify the quality – that is, number of samples - of effects such as glossy reflections, brute-force GI, AO and so on. This level of control is both a strength and weakness of biased rendering. The user’s eye is naturally skilled at determining if a particular object, material or light is more visually important and should be afforded more computational effort, that is, more samples, to produce a clean result. Meanwhile, less important or demanding elements receive less of the budget.

In this way, a chance to appropriately adjust the various sampling parameters can produce cleaner results across a scene in less time than on renderers that don’t allow the same kind of control, even taking the time spent on adjustments or trial-and-error into consideration.

Jaguar Jam

As an example, Duncan chose a recent commercial project they completed for the Jaguar FPace SUV. In it, the audience sees vehicle parts fly in from across the neon-bright nighttime cityscape to elegantly assemble the car piece by piece, straight onto the motorway as it speeds its way through the night. 

“Because this was Jaguar's first SUV, the stakes were higher than usual,” he said. “Creating dynamic movement through a nocturnal, hyperreal city was the broad remit of the project, but a key part of it was to show off the precision design and engineering of the car before revealing its final form at the end.”  Watch the finished spot here.

Outpost built, textured and lit the CG components of the car, and created the dramatic animation as the vehicle constructs itself. “As with any shoot, on the day, conditions can dictate the feel of the piece and demand some extra flexibility. In this case, the heavens opened shortly before the shoot, adding a layer of unanticipated complexity into the VFX work because now we had to build reflections into almost every shot,” Duncan said.

“The car model also contained millions of polygons and we were a lot of 8K textures, which increased the density. Redshift is very stable when handling high poly counts owing to the way it manages memory. At times the scene load time was longer than the actual render. “Throughout production we were test rendering at 5 seconds a frame. When it came to final production renders, our maximum time was 10 minutes per frame and our overall average 6.5 minutes per frame – this included global illumination, in-camera depth of focus and in-camera motion blur. That’s very fast.”

Balancing the Pipeline

According to Duncan, a competitive VFX facility works toward three considerations – quality, time and price. Quality is the key factor, and speed of render is directly linked to this – if their artists can output 20 iterations of their lighting set-up an hour on the GPU as opposed to five or six on the CPU, then they can trial more ideas in less time and reach a more creative result.

From a financial viewpoint, he noted that Redshift licenses and hardware have beencheaper per frame createdthan any CPU system they have used, meaning again that if they have a fast renderer, they are also running a more efficient system.

As well as speed, Redshift’s power has made further positive changes at Outpost VFX. Duncan said, “One example is the on-going issue we have withdepth of field and motion blur as a 2D process. For years we had to compromise between depth passes and motion vectors, and the results were never truly satisfying. Consequently, for the hero shots we would have to create all such effects in-camera and then wait an extremely long time for the render. But on the Jaguar project, we could set up a motion blur shot, balance the lighting and do test renders at 6 seconds per frame - and then apply motion blur and DOF for the final, before hitting render.”

To simulate camera motion blur, Redshift tracks the trajectories of cameras, objects/lights and vertices respectively, and represents curves using a series of up to 31 linear segments or steps. The more steps, the more memory is required but also the more accurate the trajectory simulation will be.

“This kind of capability is significant when it comes to making a production pipeline work. The speed of Redshift, along with power and reliability, meant that we could create all of the challenging shots for Jaguar with a compact team in under four weeks, from start to finish. In the end, the project stayed true to the brief, delivered ahead of schedule, and earned an excellent reception from the Jaguar team.”  www.redshift3d.com