Source URL: https://www.theregister.com/2024/10/30/ualink_consortium_incorporated/
Source: The Register
Title: No-Nvidias networking club convenes in search of open GPU interconnect
Feedly Summary: Ultra Accelerator Link consortium promises 200 gigabits per second per lane spec will debut in Q1 2025
The Ultra Accelerator Link Consortium – an alliance of enterprise tech vendors that pointedly excludes Nvidia because it wants a shared standard for accelerator-to-accelerator links – has opened its doors and promised to deliver a spec in the first quarter of 2025.…
AI Summary and Description: Yes
Summary: The emergence of the Ultra Accelerator Link Consortium, which aims to establish an open industry standard for AI accelerator communication, highlights a competitive shift in the technology landscape, specifically targeting Nvidia’s dominance. The consortium’s efforts reflect an increasing demand for interoperability and scalability in AI systems, which has significant implications for data center infrastructure and market dynamics.
Detailed Description:
– The Ultra Accelerator Link Consortium is formed by several major tech vendors including AMD, AWS, Broadcom, Cisco, Google, HPE, Intel, Meta, and Microsoft, purposefully excluding Nvidia.
– The consortium’s goal is to create an open standard to improve the communication between AI accelerators, which could lead to enhanced integration, scalability, and flexibility within AI-connected data centers.
– Nvidia’s substantial market share in networking, driven by its proprietary technologies like InfiniBand and NVLink, raises questions concerning the industry’s reliance on proprietary products versus openness:
– Nvidia’s networking sector boasts an annual revenue surpassing $14 billion.
– Open standards are favored both by vendors for market competitiveness and by buyers for cost efficiency.
– The consortium has announced plans to release a specification capable of supporting connections of up to 200Gbit/sec per lane, significantly surpassing the current capabilities of Nvidia’s offerings.
– Although the specification is set for a general review in Q1 2025, the actual implementation may take additional years.
– Nvidia’s CEO has downplayed the threat posed by the consortium, emphasizing that Nvidia will likely progress its technology further ahead of UALink’s release.
– The consortium’s chair reiterated the critical milestone that the forthcoming UALink 1.0 specification represents, emphasizing its potential to enhance communication efficiency and expand memory access necessary for large AI models.
This development underscores the necessity for security and compliance professionals to stay abreast of industry shifts towards open standards and the implications for infrastructure security, data interoperability, and potential adjustments in market dynamics. Understanding these changes will be vital for effective strategic planning, risk assessment, and the implementation of security protocols in an evolving landscape characterized by rapid technological advancements.