English

Blink: CPU-Free LLM Inference by Delegating the Serving Stack to GPU and SmartNIC

Distributed, Parallel, and Cluster Computing 2026-04-10 v1 Machine Learning Operating Systems Performance Software Engineering

Abstract

Large Language Model (LLM) inference is rapidly becoming a core datacenter service, yet current serving stacks keep the host CPU on the critical path for orchestration and token-level control. This makes LLM performance sensitive to CPU interference, undermining application colocation and forcing operators to reserve CPU headroom, leaving substantial capacity unutilized. We introduce Blink, an end-to-end serving architecture that removes the host CPU from the steady-state inference path by redistributing responsibilities across a SmartNIC and a GPU. Blink offloads request handling to the SmartNIC, which delivers inputs directly into GPU memory via RDMA, and replaces host-driven scheduling with a persistent GPU kernel that performs batching, scheduling, and KV-cache management without CPU involvement. Evaluated against TensorRT-LLM, vLLM, and SGLang, Blink outperforms all baselines even in isolation, reducing pre-saturation P99 TTFT by up to 8.47×\times and P99 TPOT by up to 3.40×\times, improving decode throughput by up to 2.1×\times, and reducing energy per token by up to 48.6%\%. Under CPU interference, Blink maintains stable performance, while existing systems degrade by up to two orders of magnitude.

Keywords

Cite

@article{arxiv.2604.07609,
  title  = {Blink: CPU-Free LLM Inference by Delegating the Serving Stack to GPU and SmartNIC},
  author = {Mohammad Siavashi and Mariano Scazzariello and Gerald Q. Maguire and Dejan Kostić and Marco Chiesa},
  journal= {arXiv preprint arXiv:2604.07609},
  year   = {2026}
}
R2 v1 2026-07-01T12:00:11.100Z