# SonicMoE: Accelerating MoE with IO and Tile-aware Optimizations

arXiv: https://arxiv.org/abs/2512.14080 · HTML: https://mlsanity.com/paper/2512.14080/

## Key facts

- Cited by 21 papers in the map (data through 14 Oct 2026).
- Placed in Efficient MoE training and inference systems, one of 12 directions in Mixture-of-experts.
- Builds on 38 of its 64 references that are in the map.
- First posted on arXiv 16 Dec 2025.

## Placed in

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

## Builds on (most cited first, 30 of 38)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [PyTorch: An Imperative Style, High-Performance Deep Learning Library](https://mlsanity.com/paper/1912.01703/) (2019)
- [Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge](https://mlsanity.com/paper/1803.05457/) (2018)
- [HellaSwag: Can a Machine Really Finish Your Sentence?](https://mlsanity.com/paper/1905.07830/) (2019)
- [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)
- [PIQA: Reasoning about Physical Commonsense in Natural Language](https://mlsanity.com/paper/1911.11641/) (2019)
- [BoolQ: Exploring the Surprising Difficulty of Natural Yes/No Questions](https://mlsanity.com/paper/1905.10044/) (2019)
- [Can a Suit of Armor Conduct Electricity? A New Dataset for Open Book Question Answering](https://mlsanity.com/paper/1809.02789/) (2018)
- [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://mlsanity.com/paper/1909.08053/) (2019)
- [CommonsenseQA: A Question Answering Challenge Targeting Commonsense Knowledge](https://mlsanity.com/paper/1811.00937/) (2018)
- [GShard: Scaling Giant Models with Conditional Computation and Automatic Sharding](https://mlsanity.com/paper/2006.16668/) (2020)
- [Crowdsourcing Multiple Choice Science Questions](https://mlsanity.com/paper/1707.06209/) (2017)
- [Mixture-of-Experts with Expert Choice Routing](https://mlsanity.com/paper/2202.09368/) (2022)
- [FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-precision](https://mlsanity.com/paper/2407.08608/) (2024)
- [GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models](https://mlsanity.com/paper/2508.06471/) (2025)
- [OLMoE: Open Mixture-of-Experts Language Models](https://mlsanity.com/paper/2409.02060/) (2024)
- [Mixture-of-Depths: Dynamically allocating compute in transformer-based language models](https://mlsanity.com/paper/2404.02258/) (2024)
- [Unified Scaling Laws for Routed Language Models](https://mlsanity.com/paper/2202.01169/) (2022)
- [MegaBlocks: Efficient Sparse Training with Mixture-of-Experts](https://mlsanity.com/paper/2211.15841/) (2022)
- [Auxiliary-Loss-Free Load Balancing Strategy for Mixture-of-Experts](https://mlsanity.com/paper/2408.15664/) (2024)
- [Kimi Linear: An Expressive, Efficient Attention Architecture](https://mlsanity.com/paper/2510.26692/) (2025)
- [Scaling Laws for Fine-Grained Mixture of Experts](https://mlsanity.com/paper/2402.07871/) (2024)
- [Mixture of A Million Experts](https://mlsanity.com/paper/2407.04153/) (2024)
- [Memory Layers at Scale](https://mlsanity.com/paper/2412.09764/) (2024)
- [Kernel Thinning](https://mlsanity.com/paper/2105.05842/) (2021)
- [Towards Greater Leverage: Scaling Laws for Efficient Mixture-of-Experts Language Models](https://mlsanity.com/paper/2507.17702/) (2025)
- [AdaMoE: Token-Adaptive Routing with Null Experts for Mixture-of-Experts Language Models](https://mlsanity.com/paper/2406.13233/) (2024)
- [TileLang: A Composable Tiled Programming Model for AI Systems](https://mlsanity.com/paper/2504.17577/) (2025)
- [Insights into DeepSeek-V3: Scaling Challenges and Reflections on Hardware for AI Architectures](https://mlsanity.com/paper/2505.09343/) (2025)

## Built on by (most cited first, 21 of 21)

- [Kimi K3: Open Frontier Intelligence](https://mlsanity.com/paper/2607.24653/) (2026)
- [M2RNN: Non-Linear RNNs with Matrix-Valued States for Scalable Language Modeling](https://mlsanity.com/paper/2603.14360/) (2026)
- [CAKE: Compiler-Agent Co-Design for Frontier Kernel Evolution](https://mlsanity.com/paper/2608.12629/) (2026)
- [PROBE: Co-Balancing Computation and Communication in MoE Inference via Real-Time Predictive Prefetching](https://mlsanity.com/paper/2602.00509/) (2026)
- [Revealing the Challenges of Attention-FFN Disaggregation for Modern MoE Models and Hardware Systems](https://mlsanity.com/paper/2602.09721/) (2026)
- [Dion3: Full-Stack Orthogonal Updates](https://mlsanity.com/paper/2608.11612/) (2026)
- [MoRE: Mixture of Reused Experts](https://mlsanity.com/paper/2609.18176/) (2026)
- [L3: Large Lookup Layers](https://mlsanity.com/paper/2601.21461/) (2026)
- [Complete-muE: Optimal Hyperparameter Transfer and Scaling for MoE Models](https://mlsanity.com/paper/2605.23893/) (2026)
- [OmniMoE: An Efficient MoE by Orchestrating Atomic Experts at Scale](https://mlsanity.com/paper/2602.05711/) (2026)
- [RaMP: Runtime-Aware Megakernel Polymorphism for Mixture-of-Experts](https://mlsanity.com/paper/2604.26039/) (2026)
- [LaCache: Exact Caching and Precision-Adaptive Inference for Diffusion Large Language Models](https://mlsanity.com/paper/2607.16339/) (2026)
- [Decoding the Skew: Distribution-Aware MoE Inference with Adaptive Kernel Dispatch](https://mlsanity.com/paper/2607.23099/) (2026)
- [Training Communication-Efficient Mixture-of-Experts Language Models with Layer Re-Configuration](https://mlsanity.com/paper/2608.28511/) (2026)
- [MonoMoE: An Efficient Fused Mega-kernel for Quantized MoE Decoding](https://mlsanity.com/paper/2609.04244/) (2026)
- [MoEMB: Scaling Universal Multimodal Embeddings with Efficient Mixture-of-Experts Models](https://mlsanity.com/paper/2609.08663/) (2026)
- [TopoEP: Topology-Aware Load Balancing for Expert-Parallel MoE Training](https://mlsanity.com/paper/2609.35481/) (2026)
- [Mixture-of-Kittens: MoE Megakernel for NVL72s](https://mlsanity.com/paper/2609.36070/) (2026)
- [MoRE: Scaling mixture of experts with hardware-aware low-rank routing](https://mlsanity.com/paper/2609.36301/) (2026)
- [Efficient Expert-Parallel Communication on PCIe-Connected Consumer GPUs](https://mlsanity.com/paper/2609.40093/) (2026)
- [Memory-Efficient Expert Routing for Distributed MoE Training](https://mlsanity.com/paper/2610.07333/) (2026)

## Abstract (arXiv)

Mixture of Experts (MoE) models have emerged as the de facto architecture for scaling up language models without significantly increasing the computational cost. Recent MoE models demonstrate a clear trend towards high expert granularity (smaller expert intermediate dimension) and higher sparsity (constant number of activated experts with a higher number of total experts), which improve model quality per FLOP. However, fine-grained MoEs suffer from increased activation memory footprint and reduced hardware efficiency due to higher IO costs, while sparser MoEs suffer from wasted computations due to padding in Grouped GEMM kernels. In response, we propose a memory-efficient algorithm to compute the forward and backward passes of MoEs with minimal activation caching for the backward pass. We also design GPU kernels that overlap memory IO with computation, benefiting all MoE architectures. Finally, we propose a novel "token rounding" method that minimizes the wasted compute due to padding in Grouped GEMM kernels. As a result, our method SonicMoE reduces activation memory by 45% and achieves a 1.86x compute throughput improvement on Hopper GPUs compared to ScatterMoE's BF16 MoE kernel for a fine-grained 7B MoE. Concretely, SonicMoE on 64 H100s achieves a training throughput of 213 billion tokens per day, comparable to ScatterMoE's 225 billion tokens per day on 96 H100s for a 7B MoE model training with FSDP-2 using the lm-engine codebase. On Blackwell GPUs, SonicMoE also achieves a 25% and 15% relative speedup on the forward and backward pass respectively compared to a highly optimized DeepGEMM baseline on OLMoE-sized 7B MoE models. Under high MoE sparsity settings, our tile-aware token rounding algorithm yields an additional 1.16x speedup on kernel execution time compared to vanilla top-K routing while maintaining similar downstream performance on Hopper GPUs. We open-source all our kernels.