CS Paper Radar

MoE Expert Execution in Disaggregated LLM Serving with a High-Bandwidth ReRAM Near-Memory Architecture

Kunming Shao, Ming Zeng, Xin Yuan, Binbin Liao, Yangming Zhang, Wei Wang, Tim Kwang-Ting Cheng, Chi-Ying Tsui

cs.AR ·submitted 2026-08-14 05:12 UTC · v1

Abstract

Attention-FFN disaggregation maps LLM modules to specialized pools, creating an opening to keep Mixture-of-Experts (MoE) weights resident in a high-bandwidth FFN pool. Decode SLOs, however, cap the run-batch while sparse routing expands the activated-expert union, so weight traffic amortizes poorly and routing skew idles cold-expert resources. The FFN pool must therefore deliver weight-read bandwidth density under sparse unions and recover occupancy under skew without a global sharing fabric. We present a ReRAM near-memory architecture that keeps expert weights resident behind high-bandwidth local reads. The design factors actual MFU into ideal MFU and occupancy, recovers occupancy with bounded core-local multicast pooling, coactivation-aware placement, and load-aware fetch, and sizes each communication level from induced demand. A measured + modeled study on Qwen3.5-35B-A3B, Qwen3.5-397B-A17B, and GLM-5.2 shows that side-4 pooling raises occupancy from 0.328 to 0.519 and, at iso-peak compute, lowers per-token FFN-pool latency by 9.5x versus H20 with 20x lower weight-movement energy; an H20-attention + ReRAM-FFN system reduces decode TPOT by 1.25-4.0x, 2.4-10.3x, and 2.5-10.4x versus a homogeneous H20 pool.

Closest papers

Nearest neighbours by meaning across the whole index.

Beyond Capacity: Scalable MoE LLM Inference via High-Bandwidth Flash with Direct GPU and HBM Paths

cs.AR ·2026-08-14 ·Seeyeon Kim, Juhyeong Jin, Joo-Young Kim
highly cited similarity 0.71
Abstract

Modern mixture-of-experts (MoE) language models increasingly strain the capacity and cost efficiency of high-bandwidth memory (HBM), as rapidly growing expert weights must be provisioned close to GPUs. High-bandwidth flash (HBF) offers substantially greater capacity, but conventional designs typically deliver HBF-resident expert weights to the GPU through HBM, leaving an additional direct GPU-HBF connection underutilized. We explore an HBF organization that simultaneously exploits two independent expert-delivery routes: a direct path that transfers expert weights from HBF to the GPU and a relay path that transfers them from HBF through the HBM base die to the GPU. Whole experts are assigned to one of the two routes, and transfers over both routes proceed concurrently, increasing aggregate expert-delivery bandwidth without replicating expert weights or introducing a shared relay bottleneck. Early expert determination identifies upcoming experts ahead of their conventional execution point, allowing HBF read latency to overlap with preceding computation, while separate management of immutable expert weights and mutable KV-cache data reduces interference between the two traffic classes. We evaluate the architecture using an event-driven continuous-batching LLM serving simulator with empirically measured GPU compute latencies. Across representative MoE workloads, concurrently utilizing the direct GPU-HBF and HBF-HBM-GPU routes consistently improves expert-delivery efficiency over designs restricted to either route alone. For a representative workload, the proposed architecture can achieve 1.94$\times$ higher throughput and 1.90$\times$ end-to-end speedup over a design that delivers all HBF-resident expert weights to the GPU through the HBM base die.

0.43

MoE-CORE: Coordinated Expert Offloading and Residency for Memory-Constrained MoE Inference

cs.DC ·2026-10-01 ·Ke Yang, Yongji Gao, Xushi Li, Kui Luo, Sicheng Zhang +14 more
similarity 0.70
Abstract

Sparse expert activation reduces MoE models' computation, yet expert weights can exceed limited device memory. Offloading makes inference feasible on a compact AI appliance but exposes host-to-device transfers to the inference path. We present MoE-CORE, a system that coordinates expert offloading and residency for memory-constrained MoE inference. It stages complete expert layers in alternating buffers during prefill. During decode, it combines nonuniform layer-wise cache capacity, domain-informed initialization, routing-history-aware replacement, and cross-layer prefetching. The main configuration executes router-selected experts exactly; an optional score-based substitution path handles eligible low-score misses. The main comparison uses 1K- and 128-token output caps for MoE-CORE and vLLM Prefetch, respectively. Across five workloads per model, MoE-CORE records a mean time per output token (TPOT) of 38.0-44.8 ms versus 1268.9-1269.1 ms for the evaluated vLLM Prefetch configuration on DeepSeek-V4-Flash-W4A8; the corresponding values on GLM-5.2-W4A8C8 are 206.6-220.5 and 5941.5-5941.8 ms. Under an 84-GB NPU-memory cap, the best measured DeepSeek GSM8K configuration achieves a TPOT of 21.5 ms with approximate expert substitution and multi-token prediction (MTP) at depth 2. These results support coordinated expert residency and transfer scheduling under a device-memory constraint. The code is here.

0.39

3D-MoE: Complementarity-Guided Hardware-Software Co-Design for MoE Decode on 3D-Stacked NMP

cs.AR cs.AIcs.DC ·2026-10-19 ·Yudong Pan, Lian Liu, Shixin Zhao, Yintao He, Kaiyan Chang +6 more
similarity 0.70
Abstract

Mixture-of-Experts (MoE) have become a mainstream way to scale Large Language Models (LLMs), because each token activates only a small subset of experts, reducing inference cost. However, MoE decode remains highly memory-bound and exhibits strongly skewed, time-varying expert loads. 3D-stacked DRAM near-memory processing (NMP) provides the bandwidth that decode demands, yet its tiled organization makes load balancing expensive: expert parallelism (EP) preserves locality but suffers from hotspot-induced imbalance, while more distributed strategies weaken locality and increase NoC traffic. To address this locality-balance conflict, we observe that domain-specialized token routing creates offsetting expert-load fluctuations across decode steps, a property we call expert complementarity. Using this insight, we propose 3D-MoE, a hardware--software co-designed system for MoE decode on 3D-stacked NMP. 3D-MoE uses offline, one-time complementarity-guided grouping and locality-aware mapping to establish coarse-grained inter-group balance, handles residual intra-group hotspots through a lightweight Collaborative Memory-Network (CMN) that streams each hot expert's weight tiles once from its host DRAM channel to helper cores over a reconfigured local broadcast path, and uses cost-aware online remapping to remain robust to workload drift. Across MoE models, 3D-MoE achieves an average 4.63$\times$ MoE-layer speedup over EP and improves end-to-end throughput by 1.58$\times$--1.86$\times$ over the strongest NMP baselines, with 2.11\% additional logic-die area.

0.14

AFORE: Attention-FFN Disaggregation with Overlapped Reconfiguration of Experts

cs.DC ·2026-10-02 ·Wenshuang Li, Youhe Jiang, You Peng, Jiawei Jiang, Binhang Yuan
similarity 0.69
Abstract

Efficient serving of Mixture-of-Experts (MoE) models is challenging due to large expert parameters, input-dependent expert activation, and dynamic workloads. Expert parallelism distributes expert computation across GPUs, while attention-FFN disaggregation (AFD) separates attention and feed-forward computation into independent worker pools. However, we observe that a naive AFD implementation could make expert load imbalance more harmful: once FFN computation becomes an independent pipeline stage, overloaded experts directly slow the FFN stage and degrade end-to-end serving performance. To solve this problem, we present AFORE, a timely expert reconfiguration system for AFD-based MoE serving. AFORE exploits two architectural properties of AFD. First, AFD exposes the expert-token distribution of upcoming microbatches before they reach FFN execution, enabling placement decisions based on near-future demand instead of stale historical profiles. Second, AFD creates a pipeline window in which expert migration for a target microbatch can be overlapped with the computation of preceding in-flight microbatches. AFORE formulates expert reconfiguration as a microbatch-aware scheduling problem and uses a migration-aware scheduler to decide when and which experts to migrate. AFORE further implements lightweight demand prefetching and NVLink-based GPU-GPU expert migration to reduce reconfiguration overhead. Evaluation on a 110B-parameter MoE model across four dynamic workloads shows that AFORE improves output throughput by 10.1-17.6% and reduces P95 inter-token latency by 7.1-9.5% compared with the strongest competing baseline. Compared with static placement, AFORE improves throughput by 29.8% on average and reduces P95 inter-token latency by 18.2% on average. Migration profiling further shows that AFD pipeline overlap can fully hide expert-migration latency.

0.15

The Other Half of the Memory Wall: Serving 35B MoEs from SSD with Trained Routing Prediction

cs.AI ·2026-09-16 ·Yu Lin, Yiming Wang, Runyuan Cai, Hanze Liu, Xiaodong Zeng
code/data similarity 0.68
Abstract

Mixture-of-experts (MoE) inference on consumer hardware is bounded by weight memory: a 35B-class model is 19.5GB at 4-bit, and sparsity shrinks the compute per token, not the bytes that must be held. Naive offloading to SSD does not help on its own, because layer N+1's experts must be chosen before layer N's output exists, so the reads cannot start early enough to hide behind compute. We present Edge0, a streaming MoE inference engine that closes the gap with a prerouter: a per-layer head predicts the next layer's routing one token ahead, and the prediction is consumed as the routing itself, so the staged expert set equals the routed set and nothing is dropped. An unmerged recovery LoRA, trained on the student path, pays back the quality lost to int4 quantization and routing replacement. On a single 24GB machine, Edge0 serves a 35B MoE at 20tok/s inside 3GiB of peak active memory, within a few points of its fp16 teacher on average across five public benchmarks. An 8B tier runs on the same framework, and the framework, checkpoints, and adapters are open source.

0.26

FlashOffloading: Low-Latency KV Cache Offloading for Long-Context LLM Serving

cs.AI ·2026-11-05 ·Xuan Wang, Zhuanhao Wu, Tingzhu Fang, Xushi Li, Sicheng Zhang +15 more
code/data similarity 0.67
Abstract · 13 pages including references,13 figures

In large language model (LLM) serving systems, the key-value (KV) cache grows linearly with sequence length and concurrency, so long-context inference easily hits the high-bandwidth memory (HBM) capacity wall and results in low concurrency. Offloading the KV cache to host dynamic random-access memory (DRAM) frees HBM space and raises concurrency, but the additional cache management and host-to-device (H2D) load overhead inflates inference latency, higher concurrency fails to translate into effective throughput. We present FlashOffloading, a host-DRAM-centric inference architecture that fully offloads the KV cache in both the prefill and decode phases. FlashOffloading builds rolling HBM cache for prefill, a hot cache for decode, a reverse sparse index with zero extra miss-lookup overhead, and a compute--load overlapped kernel, optimizing KV offloading latency across the entire pipeline and delivering throughput gains under a time-per-output-token (TPOT) service-level objective (SLO). In experiments deploying GLM-5.2 with multi-token prediction (MTP3) speculative decoding on an Ascend910C prefill--decode (PD)-disaggregated platform, FlashOffloading improves prefill throughput by up to 1.38\(\times\) and decode throughput by up to 1.91\(\times\) under a 30 ms TPOT constraint. FlashOffloading is integrated into an open-source serving framework: https://github.com/vllm-project/vllm-ascend/tree/releases/v0.30.0.

0.31

SPIMOE: Exploiting Hybrid Sparsity for Reasoning MoE Inference on Heterogeneous PIM Architectures

cs.AR ·2026-09-28 ·Rubing Yang, Cenlin Duan, Yingjie Qi, Xiaolin He, Xiao Ma +1 more
published similarity 0.67
Abstract

Long-reasoning Mixture-of-Experts (MoE) models expose two coupled inference bottlenecks: growing KV caches shift the critical path toward attention, while sparse expert activation causes load imbalance and low hardware utilization. Although Processing-in-Memory (PIM) offers a promising way to mitigate data movement overhead, existing PIM-based accelerators typically optimize attention or FFNs in isolation. We propose SPIMOE, the first co-design framework that exploits hybrid sparsity for efficient MoE inference on heterogeneous PIM architectures. SPIMOE combines adaptive expert routing with block-sparse attention and physical KV-cache eviction, and disaggregates attention and FFNs across SRAM-PIM and HBM-PIM. Static expert mapping and dynamic sub-batch scheduling further balance channel loads and overlap the two paths. Evaluations show that SPIMOE achieves up to $8.35\times$ end-to-end speedup over an NVIDIA A100 GPU and $3.33\times$ speedup in MoE FFN execution over PIMoE, while preserving reasoning accuracy comparable to full-attention baselines.

0.24

LRMoE: Demand-Side Load Re-Calibration for Efficient MoE Training on Memory-Constrained Nodes

cs.AI cs.DCcs.LG ·2026-10-27 ·Jiaming Hu, Zhida Jiang, Zhaolong Xing, Yu Li, Zhen Chen +1 more
similarity 0.67
Abstract

Sparse Mixture-of-Experts (MoE) models offer scalable capacity for Web applications, but uneven routing creates computation and communication bottlenecks. Existing placement and replication methods redistribute this workload, yet additional expert copies compete with token-related buffers for limited GPU memory. We present LRMoE, a memory-aware MoE training system that jointly recalibrates backward demand and balances the retained workload. During forward execution, it plans retention quotas and expert placement under resource constraints while preserving all routed computations. During backpropagation, a load recalibration metric estimates each token-expert pair's contribution to the expert update and selects pairs within these quotas. A capacity-aware routing algorithm distributes retained pairs among eligible instances of their original experts. Physical compaction then forms smaller communication payloads and variable-size grouped GEMMs, reducing backward traffic, computation, and execution-buffer demand. Experiments on Qwen3-30B-A3B and GLM4.5-106B-A12B show that LRMoE reduces cumulative MoE-layer training time to fixed-loss targets by 21.2-44.8% relative to SYMI.

0.16