Fabric Software Breaks New Ground with GPU Compute at Zero Cost

June 26, 2014

Fabric Software launches Fabric Engine 1.12 with automated GPU acceleration

Tuesday, June 26, 2014 9:00 am EST

Montreal, 26 June 2014 – Fabric Software has revealed its pioneering work that enables all visual effects (VFX) studios to harness the power of the graphics card (GPU), rather than just those with dedicated R&D. Due to the power of Fabric’s core execution engine, coupled with the KL scripting language, TDs as well as R&D teams can now write high performance custom tools that run either on the CPU or the GPU, leading to massive performance gains.

“Often high performance code becomes a bottleneck for both R&D and production,” said Paul Doyle, Fabric Software CEO and co-founder. “Now, thanks to our work with NVIDIA and AMD, TDs can be self sufficient for longer, while R&D concentrates on the big production challenges.”

Zero cost experimentation

With Fabric Engine 1.12, studios can develop extremely powerful custom tools without investing in specialized GPU programming efforts in CUDA or OpenCL. “KL code runs without modification on CUDA GPUs,” said Peter Zion, Fabric Software chief architect and co-founder. “You take your code in KL and it’s as easy as flipping a switch – it simply works on the GPU. It’s GPU compute at zero cost.” This means that many more tools can be GPU accelerated, as the cost and risk of doing so is near zero. The tool is simply run on either the CPU or the GPU, depending on which offers the highest performance for a particular tool.

Massive performance gains

Performance testing using simple KL code for a Mandlebrot set, which is easily parallelizable, showed a frame rate increase from 2.1 frames per second (fps) on the CPU to 23 fps on the GPU. A Maya deformation scene, with more complexity and with performance dependent on the Maya renderer, resulted in the same code running at 5.1 fps on the CPU and 24 fps on the GPU.

Damien Fagnou, MPC global head of VFX said: “Being able to visualize the work that you’re producing as you’re producing it is really important. KL was already as fast as C++ on the CPU and now it’s as fast as CUDA on the GPU. That’s really, really exciting for us and why we’re working with Fabric – it’s the future of computing.”

Accessibility with Fabric KL

Fabric KL is as simple to learn as JavaScript or Python, and as powerful as C++. The difference between KL and other scripting languages is that KL uses the Fabric Core Engine which handles multithreading, memory management, and now with Fabric Engine 1.12, GPU computation. It is this difference that enables studios, for the first time, to write custom tools that are extremely high performance without the need for additional investment in highly optimized C++ or CUDA programming.

Breaking new ground

The company was an early adopter of the LLVM compiler, which is rapidly becoming an industry standard. As a result, Fabric Software engineers were quickly able to take advantage of NVIDIA’s new CUDA 6 unified memory feature to adapt KL to run on the GPU. Ujval Kapasi, NVIDIA CUDA product manager said: “We’re impressed with what Fabric Software has achieved with our CUDA 6 Unified Memory feature – they’re one of the first companies in the world, and the first in the visual effects space, to access the power of the GPU in this way.”

Fabric 1.12 supports NVIDIA GPUs with a compute capability of at least 3.0. Subsequent releases will support AMD’s HSA architecture as well. Presentations about Fabric Software’s work with NVIDIA and AMD on the GPU are available at http://fabricengine.com/splice-2/fabric-engine/gpu-compute/


Fabric Software Inc., maker of the Fabric Engine development framework, was founded in 2010 by Paul Doyle, Philip Taylor, and Peter Zion to provide a new approach to content creation. Fabric’s revolutionary technology makes it possible for visual effects and games studios of all sizes to build high performance custom tools. The team is known for its expertise in 3D animation tools, VFX production, game development and high performance computing, having worked together at Autodesk and, prior to that, on the design, development and commercialization of ICE at Softimage. The company markets and licenses its products directly via its sales team and website www.fabricengine.com

All products or brand names mentioned are trademarks or registered trademarks of their respective holders.
© Copyright 2014 Fabric Software.