Département Artemis

Massive multimedia GPU-Based Processing

Through the evolution of computing hardware (especially GPUs – graphical processing units), processing heavy multimedia content has never been so fast and so cheap. The complexity of those operations is still increasing while computing devices are getting smaller, and require less energy. The market tends to offer everyday consumers the real‐time or near real‐time capability to acquire, handle and transform images, videos, 3D scenes of today’s unrivaled qualities.

Those objectives cannot be met just by hardware speed‐up. They rather require more complex and more specialized hardware architectures. As a result, considerable efforts will have to be engaged in order adapt exiting software components in order to leverage the newly available additional processing power.

The MediaGPU project will develop a software architecture and will review and adapt a number of classical multimedia algorithms, considering the latest advances offered by the new hardware architectures, such as:

• Hybrid CPU+GPU
• GPGPU “General Purpose computation on Graphical Processing Units”

Initial key target applications are very large still images processing, high definition video encoding, video post‐production, real‐time geometry 3D synthesis. The lower layers of the architecture will also facilitate the development of future applications.

The MEDIA GPU Project is part of a major initiative led by Institut TELECOM to team‐up efforts of the Multimedia group to solve current issues met by Multimedia Industry in the domain of hybrid computing. This effort will be support by hybrid computing specialists both industrial and academic.This partnership will allow novel algorithmic approach to the 3D and video domains capitalizing on recent advances in Hybrid computing

 Partners: HPC Project, ATEME, LABRI

 Type: National

 Keywords: GPU, Multimedia content

 Contact: Marius Preda