3天前Nvidia opens NVLink Fusion to third-party accelerators in bid to lock in the AI data centerArs Technica · T1基础设施与芯片#gpu#interconnect#cuda◆AI 生成88◀
3天前NVLink Fusion is Nvidia's play to become the connective tissue of every AI clusterTechCrunch · T2基础设施与芯片#gpu#interconnect#cuda◆AI 生成74◀
3天前Why interconnects, not GPUs, may decide who wins the next decade of AI hardwareMIT Technology Review · T2基础设施与芯片#gpu#interconnect#cuda◆AI 生成71◀
3天前OpenAI cuts API inference prices again as serving costs fall below training for the first timeOpenAI News · T1模型与训练#llm#efficiency#inference◆AI 生成79◀
4天前Google details how it squeezed 40% more tokens per dollar out of serving fleetsGoogle Research Blog · T1模型与训练#llm#efficiency#inference◆AI 生成66◀
4天前Understanding KV cache reuse across conversational workloadsOpenAI News · T1人工智能#llm#efficiency◆AI 生成58◀
4天前Rust 1.84 stabilizes async closures, clearing the last major ergonomic gap in async RustArs Technica · T1开发者工具#rust#async#open-source◆AI 生成68◀
4天前What async closures mean for the 2.4 million lines of Rust we maintainGoogle Research Blog · T1开发者工具#rust#async#open-source◆AI 生成61◀