Harshad Agashe

The talk will delve into the intricacies of GPU fabrics specifically designed for GenAI workloads. It will explore the optimal network topologies that can effectively handle the unique traffic patterns generated during GenAI training. Additionally, the discussion will cover various end-to-end congestion control mechanisms and their profound impact on the performance and efficiency of network hardware. By understanding these key aspects, attendees will gain valuable insights into building robust and high-performance GPU fabrics that can accelerate the development and deployment of cutting-edge GenAI applications.

Bio
I embarked on my VLSI journey at CDAC, where I was instrumental in developing a homegrown RDMA solution leveraging FPGA technology. Following this, I transitioned to Nevis Networks, where I delved into the design of ASIC-based pattern matching solutions. For the past sixteen years, I have been a Principal Engineer at Juniper Networks India Private Limited. At Juniper, I have played a pivotal role in shaping the evolution of multiple generations of cell-based fabric ASICs, including the underlying fabric protocol. Furthermore, I have successfully architected and designed large-scale, hierarchical Quality of Service schedulers for these ASICs.

Harshad Agashe
Juniper Networks, India