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
Nebulous Hot Chip updates include apparent acceleration for north-south traffic and a multiplane configuration ‘spotlight’
Ciena among backers of DWDM link player using gallium arsenide over supply-chain-constrained indium phosphide
Trio weather industry flux while DDN drops off
South Korean giant backs light to unlock next-gen AI memory performance
Successful tests in Europe will see the pair deploy Omni-Path and eventual Ethernet-based platforms
CS-4 promises 30-times faster AI workloads compared to GPU platforms with latency as low as two milliseconds
Lightweight Serial Interconnect effort looks to create open options to smash the AI memory wall
Lightmatter and OCP work to standardize silicon photonic interconnects for future AI cluster backbones
AI infrastructure push pays off with triple-digit hyperscale order growth
Qualcomm-backed Lumilens debuts at a $5.5 billion valuation, with a mystery hyperscale customer already on board
SCADA and cuFile API push AI storage past the processor straight onto the GPU
Thiel Foundation fellow-led startup entices Nicira, Barefoot Networks co-founder to lend networking expertise
NAND flash alternative to HBM redefine low latency for AI inference infrastructure
CEO Wendell Weeks sees exponential adoption for co-packaged optics in AI infrastructure across scale-up and scale-out networking
Network giant touts potential of OpenAI’s distributed AI training transport
Sovereignty-centric team up to challenge non-European AI networking dominance
Scale-up interconnects emerge as a multihundred-billion-dollar market category
Literal chip giant to integrate Wafer-Scale Engine in Helios racks for enhanced AI inferencing
Featuring programmable 400G DPUs and a single-hop fabric, Helios targets a streamlined, open-standards approach to rival NVL72
Gartner server CPU gong goes to AMD while Marvell outclasses on optical connectivity
Hyperscale wants photonics boffins to wire up its AI infrastructure with light
HPC and AI veteran joins to scale global quantum business
First-ever cloud AI infrastructure report pits hyperscale giants against agile, purpose-built neocloud alternatives