Hi All,
Please find my talk proposal for the kernel meetup Blr, December 2025.
Talk Title:
Scale-Up & Scale-Out Semantics for GPU Connectivity Across UALink, UEC & ESUN — The Emerging Alternative to NVLink
Abstract:
High-performance AI and HPC workloads are rapidly outgrowing traditional GPU interconnects. While NVIDIA’s NVLink has dominated tightly coupled GPU communication, new open standards—PCIe Gen6/7, CXL 3.0+, and the emerging UALink GPU fabric—are now redefining how accelerators scale in both Scale-Up (intra-node) and Scale-Out (multi-node) environments.
This talk presents a structured model using UALink (UAL), UEC, and ESUN to explain scaling semantics for GPU fabrics: how the system decides when to scale vertically (bandwidth, shared memory domains, cache-coherent access) versus horizontally (multi-GPU pooling, fabric switching, distributed inference).
Attendees will learn how open, vendor-neutral interconnects can offer a portable and scalable alternative to NVLink for future GPU clusters.
Preferred Format: 25+5 mins (but open for Lightning talk as well for selection)
--
You received this message because you are subscribed to the Google Groups "Kernel Meetup Bangalore" group.
To unsubscribe from this group and stop receiving emails from it, send an email to kernel-meetup-ban...@googlegroups.com.
To view this discussion, visit https://groups.google.com/d/msgid/kernel-meetup-bangalore/CAMGjyMt5YqK4UuL7S305PTmz1iafgRqfWYcu8o1ea_pBLe7YFw%40mail.gmail.com.
--
You received this message because you are subscribed to the Google Groups "Kernel Meetup Bangalore" group.
To unsubscribe from this group and stop receiving emails from it, send an email to kernel-meetup-ban...@googlegroups.com.
To view this discussion, visit https://groups.google.com/d/msgid/kernel-meetup-bangalore/CAMGjyMt5YqK4UuL7S305PTmz1iafgRqfWYcu8o1ea_pBLe7YFw%40mail.gmail.com.
--