In Focus: RDMA

⚡ Remote Direct Memory Access (RDMA)
Low-Latency Data Transfer for AI, HPC and Cloud Infrastructure
Remote Direct Memory Access (RDMA) enables one computer to access the memory of another without involving the remote CPU or operating system. By eliminating unnecessary software overhead, RDMA delivers ultra-low latency, high throughput and efficient CPU utilization for AI clusters, cloud infrastructure, storage systems and high-performance computing.
Why It MattersLarge AI training jobs require GPUs to exchange enormous amounts of data with minimal latency. RDMA enables direct memory-to-memory communication across high-speed networks, improving accelerator utilization while reducing CPU overhead and communication bottlenecks.
CategoryAI Fabrics • High-Performance Networking • Storage Networking • Data Center Interconnect
Key CapabilitiesKernel bypass • Zero-copy transfers • Direct memory access • Low latency • High throughput • CPU offload • Scalable GPU communication • Lossless networking support
CoverageAI Infrastructure • RoCE • InfiniBand • Ultra Ethernet • Ethernet Fabrics • GPU Clusters • High-Performance Computing • Distributed Storage
Related TechnologiesRoCEUltra EthernetMulti-path Reliable Connections (MRC)EthernetInfiniBand
Related OrganizationsInfiniBand Trade AssociationUltra Ethernet Consortium (UEC)