# Harvest: Opportunistic Peer-to-Peer GPU Caching for LLM Inference

arXiv: https://arxiv.org/abs/2602.00328 · HTML: https://mlsanity.com/paper/2602.00328/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Placed in Serving systems and request scheduling, one of 12 directions in Efficient LLM inference and serving.
- Builds on 20 of its 28 references that are in the map.
- First posted on arXiv 30 Jan 2026.

## Placed in

- Efficient LLM inference and serving / [Serving systems and request scheduling](https://mlsanity.com/inference/serving-systems-and-request-scheduling/)

## Builds on (most cited first, 20 of 20)

- [Efficient Memory Management for Large Language Model Serving with PagedAttention](https://mlsanity.com/paper/2309.06180/) (2023)
- [Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer](https://mlsanity.com/paper/1701.06538/) (2017)
- [FlexGen: High-Throughput Generative Inference of Large Language Models with a Single GPU](https://mlsanity.com/paper/2303.06865/) (2023)
- [Unified Scaling Laws for Routed Language Models](https://mlsanity.com/paper/2202.01169/) (2022)
- [MT-Bench-101: A Fine-Grained Benchmark for Evaluating Large Language Models in Multi-Turn Dialogues](https://mlsanity.com/paper/2402.14762/) (2024)
- [A Survey on Mixture of Experts in Large Language Models](https://mlsanity.com/paper/2407.06204/) (2024)
- [GEAR: An Efficient KV Cache Compression Recipe for Near-Lossless Generative Inference of LLM](https://mlsanity.com/paper/2403.05527/) (2024)
- [MoE-Lightning: High-Throughput MoE Inference on Memory-constrained GPUs](https://mlsanity.com/paper/2411.11217/) (2024)
- [MoE-Infinity: Efficient MoE Inference on Personal Machines with Sparsity-Aware Expert Cache](https://mlsanity.com/paper/2401.14361/) (2024)
- [Characterization of Large Language Model Development in the Datacenter](https://mlsanity.com/paper/2403.07648/) (2024)
- [Towards Efficient Mixture of Experts: A Holistic Study of Compression Techniques](https://mlsanity.com/paper/2406.02500/) (2024)
- [AQUA: Network-Accelerated Memory Offloading for LLMs in Scale-Up GPU Domains](https://mlsanity.com/paper/2407.21255/) (2024)
- [KVPR: Efficient LLM Inference with I/O-Aware KV Cache Partial Recomputation](https://mlsanity.com/paper/2411.17089/) (2024)
- [Accelerating Mixture-of-Expert Inference with Adaptive Expert Split Mechanism](https://mlsanity.com/paper/2509.08342/) (2025)
- [FlexPipe: Adapting Dynamic LLM Serving Through Inflight Pipeline Refactoring in Fragmented Serverless Clusters](https://mlsanity.com/paper/2510.11938/) (2025)
- [HarMoEny: Efficient Multi-GPU Inference of MoE Models](https://mlsanity.com/paper/2506.12417/) (2025)
- [Accelerating Edge Inference for Distributed MoE Models with Latency-Optimized Expert Placement](https://mlsanity.com/paper/2508.12851/) (2025)
- [Bandwidth-Efficient Adaptive Mixture-of-Experts via Low-Rank Compensation](https://mlsanity.com/paper/2512.17073/) (2025)
- [Accelerating LLM Inference via Dynamic KV Cache Placement in Heterogeneous Memory System](https://mlsanity.com/paper/2508.13231/) (2025)
- [Joint Encoding of KV-Cache Blocks for Scalable LLM Serving](https://mlsanity.com/paper/2601.03067/) (2026)

## Built on by (most cited first, 2 of 2)

- [SMetric: Rethink LLM Scheduling for Serving Agents with Balanced Session-centric Scheduling](https://mlsanity.com/paper/2607.08565/) (2026)
- [EMA: Elastic and Performance Transparent Memory Across GPUs](https://mlsanity.com/paper/2609.27040/) (2026)

## Abstract (arXiv)

Large Language Model (LLM) inference is increasingly constrained by GPU memory capacity rather than compute throughput, driven by growing model sizes and the linear growth of the key-value (KV) cache during autoregressive decoding. Existing approaches mitigate memory pressure by offloading model state and KV tensors to host memory, but incur substantial latency due to limited PCIe bandwidth. We present Harvest, an opportunistic GPU cache management framework that exploits high-bandwidth peer-to-peer GPU interconnects to dynamically place model weights and KV cache in unused GPU memory. Harvest treats peer GPU memory as a transient cache tier, preserving correctness while reducing data movement overhead under dynamic memory availability. We demonstrate significant throughput speedup of more than 2 times by using Harvest to accelerate the retrieval of two widely-used inference components: expert layer weights and KV cache entries.