跳到正文
arXiv cs.AI· R. Jay Martin II·· 2 小时前AI 评分28

OPAL 论文提出 Compute ICE-AGE 架构:推理空闲时持久化应用记忆

The Compute ICE-AGE: OPAL ONE, Continuity That Persists While Reasoning Is Idle

AI 导读

arXiv 论文提出 OPAL,一个驻留 CPU 的 C++17 图引擎,可在推理会话之间保留记录身份、关系和已接受的数值状态,实现应用记忆与概率推理分离。在 Apple Silicon M2 Pro 上,常驻更新 P50 延迟为 0.083 微秒,可扩展至 500 万个 Atomic Memory Vertices(AMVs)。

正文 · 原文

View PDF

Abstract:The Compute ICE-AGE separates persistent application memory from probabilistic reasoning. We present OPAL, a CPU-resident C++17 graph engine that preserves record identities, relationships, and accepted numerical state between reasoning sessions. Applications evolve that state through defined updates, save canonical checkpoints, and recover continuity after restart. Bounded Local Generator Classes (BLGC) provide the mathematical foundation for this bounded local evolution.
On Apple Silicon M2 Pro hardware, the current-engine resident-update P50 measured 0.083 microseconds at 100,000, 1,000,000, and 5,000,000 retained Atomic Memory Vertices (AMVs). A same-session 1M-5M-1M sequence reproduced that median using one frozen binary and 1,024 prepared targets, with every process passing its correctness checks. These measurements cover complete resident-update calls, including validation, indexed lookup, and numerical state mutation. Initialization and storage-growth costs are reported separately.
Reliability tests verified event replay, conflict rejection, concurrent calls, allocation-failure handling, and fresh-process recovery with continued operation. Engine-version comparisons produced byte-identical checkpoints and continuation files. The installed PYAMV Python package passed 17 SDK contract checks and 20 workload checks. KINDRED's Model Context Protocol interface passed 22 distinct focused checks covering memory operations, authorization, and transport.
Together, these results demonstrate persistent continuity with low recurring local-update costs through five million retained AMVs, providing working interfaces for applications and developers. The architecture keeps accepted memory available while reasoning is idle, allowing applications to invoke models only when tasks require interpretation, planning, or generation.
Comments: 37 pages. v4 shifts focus from thermodynamic scaling analysis to concrete latency measurements, deterministic state recovery, and the PYAMV/KINDRED MCP server engineering interfaces. Historical 25M node and paging observations retained in Appendix A
Subjects: Operating Systems (cs.OS); Artificial Intelligence (cs.AI)
Cite as: arXiv:2602.16736 [cs.OS]
  (or arXiv:2602.16736v4 [cs.OS] for this version)
  https://doi.org/10.48550/arXiv.2602.16736

arXiv-issued DOI via DataCite

Submission history

From: Raymond Martin Jr [view email]
[v1] Tue, 17 Feb 2026 20:57:34 UTC (2,290 KB)
[v2] Thu, 5 Mar 2026 21:20:28 UTC (2,335 KB)
[v3] Mon, 11 May 2026 19:47:18 UTC (3,764 KB)
[v4] Mon, 16 Nov 2026 22:07:12 UTC (3,653 KB)

来源:arXiv cs.AI · arxiv.org