Meta’s custom transport protocol embraces packet chaos to boost AI throughput
MetaRoCE reimagines Ethernet data transfers to fix network latency at AI scale
MetaRoCE reimagines Ethernet data transfers to fix network latency at AI scale
Network giant touts potential of OpenAI’s distributed AI training transport
The Open Compute Project launches specifications for post-InfiniBand networking
The vendor is a big Ethernet backer
Also, Oracle first to deploy AMD's MI355X GPUs in OCI Supercluster
Tech giant reveals networking backbone for Prometheus, its multi-gigawatt AI clusters
We're better together, non-profits say
ESUN working group brings together AMD, Nvidia, Cisco, and others to develop open standards for connecting GPUs within AI clusters
Working groups will examine power, cooling, and physical design specifications to enable interoperable AI facilities and avoid costly retrofits
New standard aims to break vendor lock-in and lay groundwork for next-gen System-in-Package architectures
Google and Microsoft are backing the nonprofit’s project promoting interoperability for photonic switching
Startup members can lead projects
SSIA plans to get operators and vendors to talk about cooling, circulatory and other pressing issues
Open source specification for mass-produced Edge boxes
Making building chiplets easier
Both new groups formed just days before the granddaddy of open source hardware groups, Open Compute Project (OCP), holds its annual summit. All three focus on “fundamentally different areas”