Meta’s custom transport protocol embraces packet chaos to boost AI throughput
MetaRoCE reimagines Ethernet data transfers to fix network latency at AI scale
News - RDMA
The latest news on networking, cloud, and AI infrastructure
-
Vast Data expands Nvidia love-in for AI inference era
Firms prepare for agentic avalanche
-
Meta open-sources transport stack to scale AI training to over 100K GPUs
CTran and NCCLX releases look to challenge Nvidia's NCCL dominance
-
Alibaba Cloud claims it can reduce GPU use by 82% with pooling system
Enables a 9x output increase, says company
-
Broadcom unveils 800G Thor Ultra NIC to bust AI cluster scaling bottlenecks
New network interface card doubles Thor performance and adds congestion control for clusters scaling beyond 100,000 XPUs
-
Industry group to create open standards to improve AI data services
AMD, Cisco, and Pure Storage join Storage.AI initiative