# Not All Thoughts Need HBM: Semantics-Aware Memory Hierarchy for LLM Reasoning

arXiv: https://arxiv.org/abs/2605.09490 · HTML: https://mlsanity.com/paper/2605.09490/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning.
- Builds on 9 of its 22 references that are in the map.
- First posted on arXiv 10 May 2026.

## Placed in

- LLM reasoning / [Test-time compute and efficient thinking](https://mlsanity.com/lr/test-time-compute-and-efficient-thinking/)
- Efficient LLM inference and serving / [Efficient reasoning and agent inference](https://mlsanity.com/inference/efficient-reasoning-and-agent-inference/)

## Builds on (most cited first, 9 of 9)

- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://mlsanity.com/paper/2501.12948/) (2025)
- [Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge](https://mlsanity.com/paper/1803.05457/) (2018)
- [PyramidKV: Dynamic KV Cache Compression based on Pyramidal Information Funneling](https://mlsanity.com/paper/2406.02069/) (2024)
- [Rethinking Memory in LLM based Agents: Representations, Operations, and Emerging Topics](https://mlsanity.com/paper/2505.00675/) (2025)
- [TriAttention: Efficient Long Reasoning with Trigonometric KV Compression](https://mlsanity.com/paper/2604.04921/) (2026)
- [HeadInfer: Memory-Efficient LLM Inference by Head-wise Offloading](https://mlsanity.com/paper/2502.12574/) (2025)
- [Hold Onto That Thought: Assessing KV Cache Compression On Reasoning](https://mlsanity.com/paper/2512.12008/) (2025)
- [ELMo-Tune-V2: LLM-Assisted Full-Cycle Auto-Tuning to Optimize LSM-Based Key-Value Stores](https://mlsanity.com/paper/2502.17606/) (2025)

## Built on by (most cited first, 1 of 1)

- [Random Attention: Rethinking KV Cache Eviction for Efficient Reasoning](https://mlsanity.com/paper/2609.03430/) (2026)

## Abstract (arXiv)

Reasoning LLMs produce thousands of chain-of-thought tokens whose KV cache must reside in scarce GPU HBM. The dominant response -- permanently evicting low-importance tokens -- is catastrophic for reasoning: accuracy collapses to 0-2.5% when half the cache is removed. We ask a different question: must every token live in HBM, or can some live elsewhere? We introduce a semantics-aware memory hierarchy that sorts tokens into four tiers -- HBM, DDR, compressed, and evicted -- using cumulative attention scoring. Low-importance tokens are moved to CPU memory rather than destroyed; before each attention step they are prefetched back at full precision, contributing exactly the same terms as if they had never left the GPU. We formalize this as zero-approximation-error offloading and derive our central finding: accuracy depends solely on how many tokens are permanently discarded (the eviction ratio), not on how many remain in HBM. A controlled 3x3 grid over HBM and eviction ratios confirms this across three model scales (7B-32B) and four benchmarks. With only 3% eviction, the hierarchy retains 91% of full-cache accuracy on GSM8K and 71% on MATH-500 (n=200); at 14B scale it matches the uncompressed baseline (90% vs. 86%) while halving HBM occupancy. A head-to-head reproduction of R-KV -- the current SOTA eviction method -- on our setup achieves only 0-32% at comparable budgets. A system prototype with real GPU-CPU data movement shows that the price of this preservation is modest -- 5-7% transfer overhead -- and scaling analysis projects 2-48 GB HBM savings at production batch sizes.