# mKernel: Fast Multi-GPU, Multi-Node Fused Kernels

arXiv: https://arxiv.org/abs/2609.13585 · HTML: https://mlsanity.com/paper/2609.13585/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 22 of its 36 references that are in the map.
- First posted on arXiv 11 Sep 2026.

## Builds on (most cited first, 22 of 22)

- [FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning](https://mlsanity.com/paper/2307.08691/) (2023)
- [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://mlsanity.com/paper/1909.08053/) (2019)
- [GShard: Scaling Giant Models with Conditional Computation and Automatic Sharding](https://mlsanity.com/paper/2006.16668/) (2020)
- [Efficient Large-Scale Language Model Training on GPU Clusters Using Megatron-LM](https://mlsanity.com/paper/2104.04473/) (2021)
- [Ring Attention with Blockwise Transformers for Near-Infinite Context](https://mlsanity.com/paper/2310.01889/) (2023)
- [NanoFlow: Towards Optimal Large Language Model Serving Throughput](https://mlsanity.com/paper/2408.12757/) (2024)
- [FLUX: Fast Software-based Communication Overlap On GPUs Through Kernel Fusion](https://mlsanity.com/paper/2406.06858/) (2024)
- [Comet: Fine-grained Computation-communication Overlapping for Mixture-of-Experts](https://mlsanity.com/paper/2502.19811/) (2025)
- [Breaking the Computation and Communication Abstraction Barrier in Distributed Machine Learning Workloads](https://mlsanity.com/paper/2105.05720/) (2021)
- [TACCL: Guiding Collective Algorithm Synthesis using Communication Sketches](https://mlsanity.com/paper/2111.04867/) (2021)
- [MegaScale-MoE: Large-Scale Communication-Efficient Training of Mixture-of-Experts Models in Production](https://mlsanity.com/paper/2505.11432/) (2025)
- [TileLink: Generating Efficient Compute-Communication Overlapping Kernels using Tile-Centric Primitives](https://mlsanity.com/paper/2503.20313/) (2025)
- [Triton-distributed: Programming Overlapping Kernels on Distributed AI Systems with the Triton Compiler](https://mlsanity.com/paper/2504.19442/) (2025)
- [T3: Transparent Tracking & Triggering for Fine-grained Overlap of Compute & Collectives](https://mlsanity.com/paper/2401.16677/) (2024)
- [MPK: A Compiler and Runtime for Mega-Kernelizing Tensor Programs](https://mlsanity.com/paper/2512.22219/) (2025)
- [Rail-only: A Low-Cost High-Performance Network for Training LLMs with Trillion Parameters](https://mlsanity.com/paper/2307.12169/) (2023)
- [ParallelKittens: Systematic and Practical Simplification of Multi-GPU AI Kernels](https://mlsanity.com/paper/2511.13940/) (2025)
- [UCCL-EP: Portable Expert-Parallel Communication](https://mlsanity.com/paper/2512.19849/) (2025)
- [MSCCL++: Rethinking GPU Communication Abstractions for AI Inference](https://mlsanity.com/paper/2504.09014/) (2025)
- [An Extensible Software Transport Layer for GPU Networking](https://mlsanity.com/paper/2504.17307/) (2025)
- [GPU-Initiated Networking for NCCL](https://mlsanity.com/paper/2511.15076/) (2025)
- [The Multipath Reliable Connection (MRC) Transport](https://mlsanity.com/paper/2606.18170/) (2026)

## Abstract (arXiv)

Communication has become a bottleneck in distributed training and inference of large models. Overlapping communication with computation at the granularity of kernels, on separate streams, reduces only part of this communication cost. Fused kernels often have better performance by transmitting each output tile as soon as it is produced, but existing fused kernels are largely confined to a single NVLink domain. We present mKernel, a library of multi-GPU, multi-node fused kernels that overlap computation, intra-node NVLink communication, and inter-node RDMA at tile granularity. mKernel partitions the streaming multiprocessors (SMs) of a persistent kernel into compute and communication roles, and an on-GPU controller tunes the SM partition adaptively at run time, since the best SM partition varies with the kernel and the input shape. It structures data movement hierarchically so that data traversing the inter-node network is minimized. Finally, it drives the network from the GPU through a lightweight command queue and host proxy implemented directly on RDMA verbs, which allows the same kernels to run on any network backend (e.g. InfiniBand and on AWS EFA); we observe, surprisingly, that GPUDirect Async (IBGDA) yields little additional benefit over host-assisted GPU-initiated communication. We implement five kernels spanning tensor, sequence, and expert parallelism. On two 16-GPU H200 clusters, mKernel achieves speedups of up to 1.72x on GEMM+AllReduce and 1.88× on Ring Attention.