Sponsor Talk
Sharada Yeluri, AVP Engineering, Astera Labs
Title: Why UALink: A Protocol-Level Comparison for Rack-Scale AI
Abstract:
As AI workloads push the limits of system bandwidth and latency, the need for a purpose-built, memory-semantic fabric for scale-up AI connectivity becomes critical.
In this session, the UALink Consortium will explore how UALink is uniquely optimized for scale-up AI architecture, offering lower latency, higher efficiency, and native memory coherency for GPU-GPU communication. Attendees will gain insight into a technical comparison of UALink with Ethernet-based protocols and understand why UALink is emerging as the preferred fabric for next-generation open rack-scale AI infrastructure.
In this session, the UALink Consortium will explore how UALink is uniquely optimized for scale-up AI architecture, offering lower latency, higher efficiency, and native memory coherency for GPU-GPU communication. Attendees will gain insight into a technical comparison of UALink with Ethernet-based protocols and understand why UALink is emerging as the preferred fabric for next-generation open rack-scale AI infrastructure.
Biography:
Sharada Yeluri is AVP of Engineering at Astera Labs where her focus is on building scale-up fabric chips. Prior to Astera Labs, Sharada was Sr. Director of Engineering at Juniper where she headed the team that delivered Express silicon for PTX routers.
Sharada has nearly 30 years' experience building high performance CPUs and networking silicon. She also blogs frequently on LinkedIn on all topics related to networking and fabrics for AI.
Sharada has nearly 30 years' experience building high performance CPUs and networking silicon. She also blogs frequently on LinkedIn on all topics related to networking and fabrics for AI.