[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"topic-zh-ampere-cpasync-hbm-":3},{"cluster":4,"timeline":16},{"id":5,"slug":6,"title":7,"pinned":8,"status":9,"summary":10,"category":11,"language":12,"created_at":13,"merged_into":14,"article_count":15,"first_seen_at":13,"last_updated_at":13},"2abca7c0-acbe-4035-a0b5-b16f9a5f4cdb","ampere-cpasync-hbm-","Ampere 的 cp.async 怎麼藏 HBM 延遲",false,"active","A100 上一次 HBM2e 載入約要 450 到 600 cycles。Ampere 的 cp.async 讓資料直進 shared memory，搭配 pipeline 把等待時間藏進計算裡。","research","zh","2026-07-05T09:02:07.499362+00:00",null,1,[17],{"id":18,"slug":19,"title":7,"summary":10,"category":11,"image_url":20,"cover_image":20,"published_at":21,"is_canonical_seed":22},"d458f7db-1e28-4cf1-9bd8-ad9c95dee997","cuda-cp-async-ampere-hbm-latency-zh","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775167621432-n9fo.png","2026-04-02T22:06:35.906+00:00",true]