MeshKV: A Network-on-Chip KV Cache Fabric for Scalable Transformer Decoding Accelerators

Dong Liu; Yanxuan Yu

Home/Archive/arXiv

arXiv · 2609.19207

MeshKV: A Network-on-Chip KV Cache Fabric for Scalable Transformer Decoding Accelerators

Dong Liu·Yanxuan Yu

View original source ↗Read source document ↗Export metadata ↓

Abstract

Autoregressive transformer decoding is constrained by irregular key-value (KV) cache movement on tiled accelerators. Prior compression and DRAM-placement systems still concentrate traffic on centralized memory paths that bottleneck long-context serving. We present MeshKV, a KV cache fabric that moves blocks as packetized flows over a lightweight NoC. It co-designs (i) TaKV affine striping to spread homes and cut hotspot load, (ii) Mare multicast with verified duplicate suppression, and (iii) Pad, which overlaps prefetch, tile multiply, and streaming softmax behind credit-aligned FIFOs. Together they convert bisection back-pressure into useful KV transfer. On our 8x8 FPGA implementation with LLaMA-2-7B and Mistral-7B at 8K-32K, MeshKV reduces interconnect traffic by up to 58%, improves KV bandwidth utilization by 2.1x, and delivers up to 1.9x multi-stream throughput.

Explore related subjects

cs.AR cs.AI

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Dong Liu, Yanxuan Yu. 2026-09-16. MeshKV: A Network-on-Chip KV Cache Fabric for Scalable Transformer Decoding Accelerators. https://arxiv.org/abs/2609.19207

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Bi-SamplerZ: A Rejection-Aware Cooperative Gaussian Sampling Framework for Falcon Signature Hardware

We present Bi-SamplerZ, a rejection-aware cooperative sampling framework that converts this idle capacity into useful computation. After an asymmetric accept/reject outcome, Bi- SamplerZ latches the completed logical result and dynamically reassigns the released physical datapath to the unfinished target. The two paths then evaluate fresh independent candidates for the same remaining distribution. We show that this post-rejection cooperation increases the assisted-round completion probability without modifying the underlying candidate distribution or Bernoulli acceptance rule, and we state the randomness-allocation conditions required to preserve the joint output distribution of the original pair of logical sampler calls

cs.AR↗

arXiv2026

A Multi-Engine Dataflow for MoE Decoding on Scratchpad-Based Tensor Accelerators

Mixture-of-Experts (MoE) decoding on scratchpad-based tensor accelerators (STA) is dominated by moving expert weights while the compute engines sit idle. This traffic is hard to hide, because the experts are known only after routing, and hard to shrink without losing quality or adding critical-path work. We present CARDAN, which represents each expert-weight matrix as a vector-quantized component plus a shared-basis low-rank component and co-designs this representation with a multi-engine decoding dataflow. The representation separates expert-common from expert-private work, so the dataflow overlaps DMA with computation on several engines. Across five MoE families on AWS Trainium3, CARDAN matches or improves BF16-teacher perplexity across all five models and speeds up batch-one decoding by 1.15-1.31x over AWS dense MoE megakernels, rising to 1.7x at batch size 16.

cs.AR↗

arXiv2026

Dissecting How Die Scaling Breaks GPU Fine-grained Scheduling

Modern GPUs are no longer physically symmetric. Die scaling leads to both manufacturing-driven floorsweeping and cache and memory partitioning. The former creates chip-specific compute topologies, while the latter causes non-uniform memory access. These asymmetries are substantial. Topology-oblivious compute unit allocation can lead to up to 1.33x performance variation, while remote accesses increase HBM latency by up to 67% and nearly double L2 latency. However, these asymmetries are hidden behind the GPU's logical resource abstractions and can vary across chips. We develop lightweight characterization methods to uncover per-chip compute topology and memory affinity. We then use the discovered information to make existing fine-grained scheduling asymmetry-aware, considering not only how many resources are allocated but also which physical resources are assigned. Across full-GPU kernel execution, intra-application multiplexing, and inter-application co-location, asymmetry-aware scheduling improves mainstream kernels by up to 1.22x, multiplexed LLM inference by up to 14.3%, and avoids up to 1.33x performance variation.

cs.AR↗