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

arXiv: https://arxiv.org/abs/2608.14333 · HTML: https://mlsanity.com/paper/2608.14333/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Efficient MoE training and inference systems, one of 12 directions in Mixture-of-experts.
- Builds on 17 of its 48 references that are in the map.
- First posted on arXiv 14 Aug 2026.

## Placed in

- Mixture-of-experts / [Efficient MoE training and inference systems](https://mlsanity.com/moe/efficient-moe-training-and-inference-systems/)
- Efficient LLM inference and serving / [Diffusion LLM and MoE inference](https://mlsanity.com/inference/diffusion-llm-and-moe-inference/)

## Builds on (most cited first, 17 of 17)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Deep Residual Learning for Image Recognition](https://mlsanity.com/paper/1512.03385/) (2015)
- [Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer](https://mlsanity.com/paper/1701.06538/) (2017)
- [Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity](https://mlsanity.com/paper/2101.03961/) (2021)
- [Mixtral of Experts](https://mlsanity.com/paper/2401.04088/) (2024)
- [Root Mean Square Layer Normalization](https://mlsanity.com/paper/1910.07467/) (2019)
- [GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints](https://mlsanity.com/paper/2305.13245/) (2023)
- [Fast Transformer Decoding: One Write-Head is All You Need](https://mlsanity.com/paper/1911.02150/) (2019)
- [Efficient Large-Scale Language Model Training on GPU Clusters Using Megatron-LM](https://mlsanity.com/paper/2104.04473/) (2021)
- [Kimi K2: Open Agentic Intelligence](https://mlsanity.com/paper/2507.20534/) (2025)
- [Efficiently Scaling Transformer Inference](https://mlsanity.com/paper/2211.05102/) (2022)
- [Taming Throughput-Latency Tradeoff in LLM Inference with Sarathi-Serve](https://mlsanity.com/paper/2403.02310/) (2024)
- [Tutel: Adaptive Mixture-of-Experts at Scale](https://mlsanity.com/paper/2206.03382/) (2022)
- [Fiddler: CPU-GPU Orchestration for Fast Inference of Mixture-of-Experts Models](https://mlsanity.com/paper/2402.07033/) (2024)
- [Pre-Attention Expert Prediction and Prefetching for Mixture-of-Experts Large Language Models](https://mlsanity.com/paper/2511.10676/) (2025)
- [On-Package Memory with Universal Chiplet Interconnect Express (UCIe): A Low Power, High Bandwidth, Low Latency and Low Cost Approach](https://mlsanity.com/paper/2510.06513/) (2025)
- [Speculating Experts Accelerates Inference for Mixture-of-Experts](https://mlsanity.com/paper/2603.19289/) (2026)

## Abstract (arXiv)

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× higher throughput and 1.90× end-to-end speedup over a design that delivers all HBF-resident expert weights to the GPU through the HBM base die.