CXL and the AI Memory Hierarchy

Several sources attending the Global OCP Summit told IT Brand Pulse that Marvell intends to continue offering its current CXL switch chip but does not plan to offer future generations, instead directing resources toward UALink and/or NVLink switching products. If accurate, the reported Marvell roadmap change raises an important question for the CXL ecosystem: who will supply the next generation of switch silicon needed to move beyond today’s production platforms toward larger, more flexible memory fabrics?

Handy’s market update identified two immediate reasons to pay attention. AI systems need more DRAM, while DRAM has become more expensive and harder to obtain. Those conditions make it worthwhile to revisit memory already installed in data centers and to reduce capacity stranded inside individual servers.

The most useful contribution of Penguin’s session was its system-level framing. CXL does not eliminate the need for fast local memory, and additional capacity has value only when software and workloads can use it effectively. Its promise lies in giving system designers another placement and capacity option.

Anil Godbole traces CXL’s progress from its first specification in 2019 to deployments by major cloud operators. He also explains why memory pooling could become one of CXL’s most consequential uses for AI.

Siamak Tavallaei — Big Data Feeds Big Compute through Memory Register to attend the presentation on September 30 For most of the AI boom, the infrastructure conversation has revolved around compute: more GPUs, larger clusters, faster accelerators and ever-larger scale-up systems. Siamak Tavallaei believes the industry is reaching the point where that conversation has to change. […]

The first event in the Endless Memory with CXL® series explores one of the most practical opportunities in enterprise AI infrastructure: extending the value of existing DDR4 memory.

The keynote presentation from Penguin Solutions, Turn Yesterday’s DDR4 Memory into Tomorrow’s AI Capacity, demonstrates how organizations can repurpose DDR4 server memory into CXL memory expansion solutions for AI and data-intensive workloads.

Additional presentations from leading CXL vendors and end users will cover real-world use cases, the latest CXL technologies, key hardware and software components, and production-ready systems that are accelerating the adoption of composable memory architectures.

Based on votes from enterprise IT pros, architects, infrastructure engineers, and AI practitioners, the independent, non-sponsored research identifies Market Leaders and Innovation Leaders across 28 enterprise AI infrastructure product categories spanning AI compute, memory, storage, networking, AI operations, observability, and Compute Express Link® (CXL) infrastructure.