Overview
Learn how GPUDirect-TCPX, a custom open-source networking stack, revolutionizes network performance for AI/ML and HPC workloads in this technical session presented by industry veterans Hassan Tasneem and Hugo Huang. Discover the optimization techniques for RDMA (Remote Direct Memory Access) and GPU-Direct communication that address the limitations of traditional TCP in meeting high throughput and low latency requirements. Gain deep insights from Hassan Tasneem, a former AWS and Google Cloud leader with 15 years of experience in pioneering cloud services, and Hugo Huang, a digital transformation expert with 18 years of experience in Cloud Computing, 5G, AI, and Cyber Security. Master the implementation of TCPX to unlock maximum performance potential for data-intensive applications, enabling faster and more efficient network architectures that scale with modern computing demands.
Syllabus
How TCPX optimizes RDMA, and GPU Direct performance for AIML, and HPC workloads
Taught by
Canonical Ubuntu