GPU Performance Changes the Game for Tigar Hare
Los Angeles-based Tigar Hare creates cinematics and trailers for some of the hottest video games around. These kings of the console have been delivering the goods for Electronic Arts, Warner Bros., Sony Online, Activision, THQ and other companies since 1997, keeping their operation small while increasing their productivity by revving their pipeline to handle increasingly complex graphics. Some of the titles they’ve worked on include the blockbuster “Call of Duty: Black Ops,” “Apache: Air Assault,” “Project Gotham Racing,” “Red Faction: Guerilla” and many more.
“We need solutions for speed,” said Tigar Hare co-founder Dave Hare. To keep things humming and handle a regular flow of compute-intensive tasks like particle simulation, fluid dynamics, 3D motion blur and depth-of-field, the company updated its pipeline with a fiber optic network, and relies on graphics processing units (GPUs) from NVIDIA. The NVIDIA® Quadro® professional graphics solutions drive key renderers and modeling software at Tigar Hare, delivering a 40X speed improvement in some processes for the studio. “We’ve had great GPUs for years,” said Hare, “But until recently, no one had really figured out how to take full advantage of them.”
A long-time user of Chaos Group’s V-Ray rendering packages, Tigar Hare recently upgraded to the new V-Ray RT technology, which is greatly accelerated by Quadro GPUs based on the new NVIDIA Fermi architecture. The studio also exploits the GPU when using the mental image’s iray renderer within Autodesk 3Ds Max modeling software, “We’re finally at a point where software has caught up to hardware,” said Hare. “GPU rendering is a game changer — for us and for the whole industry.”
One of the first projects Tigar Hare put through their GPU pipeline was Activision’s “Apache Air Strike,” creating cinematics that feature beauty shots of the Apache helicopter. “We used a lot of depth-of-field to get a photorealistic look,” said Hare. “On the CPU, depth of field can be really time intensive because you’re constantly dialing in and out of focus to give it the right level of detail. It would have taken us hours or even days to render a whole sequence to see if we’d gotten it right. On the GPU we were able to change the frames, determine the focus and see the results in real time. It really opens up your creative options because you have time to keep trying things instead of living with what you have time for.”
They experienced a similar performance boost on title sequence work created for the upcoming Nicholas Cage movie, “Drive Angry 3D.” Hare explained, “We were using background plates to generate a reflection map for the main title and wanted to dial in the reflection through a move. With V-Ray on Quadro we could see the reflection in real time and dial it out — 600 frames of animation — to get it exactly how we wanted it.”
He added, “With a Quadro, I am comfortable doing this kind of work in client sessions with someone over my shoulder. That would have been unheard of before.”
Hare also spoke to the benefits that GPU-enabled V-Ray RT within 3ds Max brings. “Viewport operations will become more like a render. Before, what we saw in the viewport was flat shaded or wireframe. With the GPU, that’s changing to be more of a photorealistic environment. You can create an object, put on a texture and begin to see reflections, refractions, ambient occlusion, shadows and more, early on.”
In addition to the speed benefits artists can see on their workstations, Tigar Hare also leverages Quadro GPUs behind the scenes, distributing rendering across workstations to boost throughput and productivity even more.
Since they began leveraging the new NVIDIA Quadro 5000 and Quadro 6000 professional graphics solutions, Tigar Hare has seen an incredible 40X improvement when comparing traditional CPU rendering with the new real-time GPU-based rendering.
To compare GPU-based ray tracing, Hare benchmarked a wide range of hardware configurations, rendering the same scene using V-Ray RT 2.0 on each system.
The scene was set up to use a 3ds Max render time of two minutes. In that timeframe, a 12-core CPU processed just 100 samples — while a combination of three NVIDIA Quadro and Tesla GPUs handled a whopping 1,056 samples – a more that 10X improvement. This same GPU combination in one machine was also 3.8X faster than distributing the job across 80 CPU cores. The NVIDIA Fermi architecture incorporated into the company’s new GPUs also demonstrated a dramatic improvement over earlier GPUs, with the Quadro 5000 yielding 5.5X the speed of the older Quadro FX 5600. It also showed great multi-GPU scaling when combining three new GPU's: nearly tripling the performance of a single Quadro GPU.
“It’s an exciting time, seeing how we can leverage the massively parallel power of the GPU in our everyday workflow,” said Hare. “Kudos to Autodesk for integrating iray into 3ds Max, to Chaos Group for their GPU accelerated V-Ray RT, and to NVIDIA for opening up this pipeline.”