# FlexNPU: Transparent NPU Virtualization for Dynamic LLM Prefill-Decode Co-location

arXiv: https://arxiv.org/abs/2606.04415 · HTML: https://mlsanity.com/paper/2606.04415/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Serving systems and request scheduling, one of 12 directions in Efficient LLM inference and serving.
- Builds on 12 of its 28 references that are in the map.
- First posted on arXiv 3 Jun 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, 12 of 12)

- [DistServe: Disaggregating Prefill and Decoding for Goodput-optimized Large Language Model Serving](https://mlsanity.com/paper/2401.09670/) (2024)
- [Splitwise: Efficient generative LLM inference using phase splitting](https://mlsanity.com/paper/2311.18677/) (2023)
- [ServerlessLLM: Low-Latency Serverless Inference for Large Language Models](https://mlsanity.com/paper/2401.14351/) (2024)
- [MuxServe: Flexible Spatial-Temporal Multiplexing for Multiple LLM Serving](https://mlsanity.com/paper/2404.02015/) (2024)
- [Prism: Cost-Efficient Multi-LLM Serving via GPU Memory Ballooning](https://mlsanity.com/paper/2505.04021/) (2025)
- [semi-PD: Towards Efficient LLM Serving via Phase-Wise Disaggregated Computation and Unified Storage](https://mlsanity.com/paper/2504.19867/) (2025)
- [Torpor: GPU-Enabled Serverless Computing for Low-Latency, Resource-Efficient Inference](https://mlsanity.com/paper/2306.03622/) (2023)
- [Nexus:Proactive Intra-GPU Disaggregation of Prefill and Decode in LLM Serving](https://mlsanity.com/paper/2507.06608/) (2025)
- [Singularity: Planet-Scale, Preemptive and Elastic Scheduling of AI Workloads](https://mlsanity.com/paper/2202.07848/) (2022)
- [FaST-GShare: Enabling Efficient Spatio-Temporal GPU Sharing in Serverless Computing for Deep Learning Inference](https://mlsanity.com/paper/2309.00558/) (2023)
- [DuetServe: Harmonizing Prefill and Decode for LLM Serving via Adaptive GPU Multiplexing](https://mlsanity.com/paper/2511.04791/) (2025)
- [Nixie: Efficient, Transparent Temporal Multiplexing for Consumer GPUs](https://mlsanity.com/paper/2601.11743/) (2026)

## Abstract (arXiv)

Modern AI serving increasingly relies on NPUs for conventional inference and large language model serving. However, current NPU deployments commonly expose physical devices directly to applications, which limits runtime control over scheduling and makes it difficult to adapt execution to phase-level workload behavior. This limitation is particularly evident in LLM serving, where the prefill phase is compute-intensive while the decode phase is often constrained by memory bandwidth and KV-cache accesses. Static prefill-decode (PD) disaggregation reduces phase interference, but can introduce resource imbalance and unnecessary data movement. We present FlexNPU, a transparent user-space virtualization layer for Ascend NPUs. FlexNPU interposes on AscendCL APIs and routes NPU operations through per-device daemons, decoupling unmodified from physical NPU devices without modifying model code, AI frameworks, or NPU drivers. This runtime boundary allows FlexNPU to virtualize NPU objects, control operator dispatch, and support phase-aware scheduling for LLM serving. In particular, FlexNPU enables dynamic PD co-location, which adapts scheduling between prefill and decode according to their complementary resource characteristics. We implement FlexNPU on Huawei Ascend NPUs and evaluate it with typical LLM workloads. Compared with direct NPU passthrough, FlexNPU introduces no measurable inference overhead and slightly improves throughput in some scenarios. On a 384-card Ascend 910C deployment of DeepSeek-R1, FlexNPU improves throughput over static PD disaggregation by 5.15% and 26.33%. On Qwen2.5-7B, compared with static PD co-location, FlexNPU maintains comparable throughput while reducing TTFT by over 92% across tested workloads with nearly unchanged TPOT. These results show that transparent NPU virtualization is a practical substrate for efficient and responsive LLM serving.