Containerizing Hardware Accelerated Applications Using GPUs and FPGAs

Explore containerization techniques for hardware-accelerated applications in this Docker conference talk. Dive into the benefits and challenges of containerizing resource-intensive applications that utilize GPUs and FPGAs for acceleration. Learn how to leverage containers to reduce setup time, minimize dependency conflicts, and simplify updates for applications with complex stacks spanning kernel and user space. Examine a real-world case study of a media processing stack using GPU acceleration within containers, including insights on kernel and user space interactions. Discover the minimal performance overhead of containerization compared to native execution and the advantages of quick deployment across machines. Discuss limitations in portability due to custom kernel requirements and potential areas for innovation, such as Docker plugins for compatibility checks between container software and host kernels. Gain valuable insights into containerizing hardware-accelerated applications through this comprehensive presentation, complete with technical details, observations, and a Q&A session.