{"type":"post","id":"sglang-ships-unified-memory-a-single-gpu-arena-a96b4f","slug":"sglang-ships-unified-memory-a-single-gpu-arena-a96b4f","title":"SGLang ships Unified Memory, a single GPU arena shared by full-attention KV, sliding-window KV, and recurrent state.","description":"SGLang’s experimental Unified Memory feature lets full-attention KV, sliding-window KV and recurrent state share a GPU memory arena rather than fixed, separately sized pools.","retrieval_nugget":"SGLang’s experimental Unified Memory feature lets full-attention KV, sliding-window KV and recurrent state share a GPU memory arena rather than fixed, separately sized pools. In tests reported by LMSYS, the change increased concurrency and throughput on selected hybrid-model workloads, but it remains limited in model and feature compatibility.","published_at":"2026-11-12","updated_at":"2026-11-12","record_date":"2026-11-12","date_kind":"published_at","topics":["memory"],"entities":["SGLang","Unified Memory","LMSYS","Qwen3.5-4B","GPT-OSS-20B","H200"],"related_patterns":["harness-architecture-outlives-model-choice"],"novelty_label":"new_fact","what_changed":"SGLang adds Unified Memory, letting full-attention KV, sliding-window KV, and recurrent state dynamically share one GPU arena instead of requiring a fixed memory split.","prior_refs":["https://newruntime.com/posts/sglang-v0-5-23-ships-megakernel-low-latency-2254f8/"],"source_url":"https://www.lmsys.org/blog/2026-11-11-unified-memory","source_urls":["https://www.lmsys.org/blog/2026-11-11-unified-memory"],"schema_version":"newruntime-agent-readable-v0.2","stable_id":"post:sglang-ships-unified-memory-a-single-gpu-arena-a96b4f","status":"published","visuals":[],"editor":{"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]},"routes":{"html":"https://newruntime.com/posts/sglang-ships-unified-memory-a-single-gpu-arena-a96b4f/","markdown":"https://newruntime.com/posts/sglang-ships-unified-memory-a-single-gpu-arena-a96b4f.md","json":"https://newruntime.com/posts/sglang-ships-unified-memory-a-single-gpu-arena-a96b4f.json"},"markdown_url":"https://newruntime.com/posts/sglang-ships-unified-memory-a-single-gpu-arena-a96b4f.md","next_reads":[{"type":"topic","path":"/topics/memory/","url":"https://newruntime.com/topics/memory/","title":"Memory","reason":"Explore the memory topic hub.","media_type":"text/html","markdown_url":"https://newruntime.com/topics/memory.md","json_url":"https://newruntime.com/topics/memory.json"},{"type":"collection","path":"/posts/","url":"https://newruntime.com/posts/","title":"Field Notes","reason":"Open the latest editorial analysis.","media_type":"text/html","json_url":"https://newruntime.com/posts.json"},{"type":"collection","path":"/sources/","url":"https://newruntime.com/sources/","title":"Sources","reason":"See the original sources this archive cites.","media_type":"text/html"}]}