cadalyst
Graphics Cards and CPUs

Rigorous GPU Update Regimen Enables VFX Wizards to Produce Eye-Popping Visuals from CAD Files

27 Apr, 2016 By: Cyrena Respini-Irwin

At the W.I.P visual effects collective, graphics are always a work in progress, and continual updates are necessary to ensure the best performance possible.



Creating Compelling Visuals

An example of a CAD-based workflow for W.I.P is a car manufacturer that seeks to demonstrate how a particular shock absorption system will help — or hurt — performance during a car accident. First, the team creates a “footprint” for the project, determining what exactly they’ll do and deciding on an approach that will successfully convey the desired information. That transitions into an “envisioning session,” where a wish list of desired elements is compiled.

The W.I.P team then turns to the CAD models they’ve received from the client. They prefer to work from AutoCAD or Creo files whenever possible — converting file formats where necessary — to simplify workflows and reduce errors. “It’s amazing how similar [various CAD applications] are in what they do, but how resistant they are in talking to each other,” Bolden said.

Next, the W.I.P team begins building a virtual world for the car. This is where the heavy lifting takes place, as the team adds conditions to that world — rain and wind, for example — and obtains data from various scenarios. The simulations are run in MATLAB, FreeMat, and ASCEND software.

The last step is to create a presentation of that data that’s engaging as well as informative, said Bolden: “How do we show that in an effective way, so people aren’t snoring when we’re talking?” Computer graphics and VR help the team do just that — “everyone connects to the information better,” Bolden noted. The team uses Bunkspeed (now known as SolidWorks Visualize); Autodesk Maya and 3ds Max; and Chaos Software V-Ray. The choice of presentation platform dictates how many GPUs the team will need to put into use, and where they’ll be located; W.I.P uses NVIDIA GRID virtual GPU technology to power some of their graphics experiences

Bolden estimates that about 70% of these projects are initially intended for the client’s internal use, such as demonstrating a particular aspect of a product’s design or performance to C-level executives within the company. Once clients see how creatively their data can be conveyed, however, they often extend their use to other audiences, said Bolden. “We always strive to make them interesting and relatable to a broad audience from the beginning,” said Bolden. “Because no matter who you are, you don’t want to be bored when you’re looking at something.”




Two digitally created views of the Lincoln MKZ. Images courtesy of Lincoln and Hudson Rouge. 


The Costs of a Cutting-Edge Infrastructure

Conducting an upgrade every three to six months is no small task; W.I.P has to replace the GPUs in their expansion chassis, then install the new drivers and test them by GRID cluster with old projects. It’s also, unsurprisingly, not cheap. “The expense is a very real deal,” Bolden affirmed, “but we have hardware vendors that put us on a license cycle, so we can cycle parts in and out as needed. Cloud computing helps as well.”

The most challenging aspect of this rigorous update schedule, however, is keeping up with new drivers, especially in an industry that makes no allowances for downtime. “Most of the time, we run right up to crunch time.” In the past, Bolden noted, he might delay the addition of a new piece of hardware, for fear that something would go wrong and a project would be thrown off schedule. Now, if any problems arise, W.I.P contacts the manufacturer and receives prompt support. “NVIDIA has been really good in that department,” he said. The key is to anticipate that driver-related issues may arise, and be prepared to deal with them.

That’s also an essential approach to take with caching problems that occur during simulations, Bolden noted. “If that goes wrong, it’ll go really, really wrong. The math being crunched can blow up at any time, we can lose the cached information at any time, so we have to make it redundant.”

Bolden also noted that one downside of upgrading so frequently is that the collective’s capabilities sometimes dramatically outstrip those of its clients. “We still have to plan for the lowest common denominator. In the past, we may have made something that worked fine in our ecosystem — our workflow process to the final viewing platform — but it wouldn’t function on our client’s ecosystem. Now, we try to test everything we create on the customer’s ecosystem, or create one for them ourselves.”

An Eye on the Future

In some ways, the future for W.I.P will be more of the same — only faster. “The new GPUs are consistently more efficient,” said Bolden. Currently, the team gains a 15% speed increase when they update, “and it’s growing every upgrade,” he said. That shaves hours off of rendering jobs, for example, accelerating the team’s turnaround times.

But some tasks will be handled in new ways, Bolden believes: “It’s an assumption and a pretty solid conclusion that we’re going to have to rely on the cloud more heavily in the future. Owning all the hardware is expensive, and it’s like owning a yacht — the maintenance is more that the actual cost. Cloud computing definitely solves that problem.”

Another change is that product visualization and creation are increasingly taking place in virtual environments. For example, W.I.P is currently working on a product via HoloLens, said Bolden. “It started in a simple CAD file, but now it’s all happening remotely. That’s where design is now going, especially at that early level.”

1 2 


Add comment