MORI Documentation

MORI (Modular RDMA Interface) is a bottom-up, modular, and composable framework for building high-performance communication applications with a strong focus on RDMA + GPU integration.

New to MORI? Install it, then work through the Quickstart.

MORI architecture

Components

Component

What it is

Start here

MORI-EP

Intra and inter-node dispatch/combine kernels for MoE Expert Parallelism

MORI-EP User Guide (EPv1 on MORI-SHMEM, EPv2 on MORI-CCO, and how to choose), then Choosing an EP API and backend

MORI-FUSED

GEMMs fused with the all-reduce, reduce-scatter, all-to-all or all-gather next to them, overlapping the transfer with the GEMM over the SDMA copy engines; built on MORI-CCO

MORI-FUSED: Fused GEMM + Collective Operators

MORI-SHMEM

OpenSHMEM-style symmetric memory APIs for GPU memory and RDMA

MORI Shmem Guide

MORI-CCO

Next-Generation Collective Communication Object for GPUs: symmetric memory, natively designed for compute-communication fusion (LSA / GDA / SDMA)

MORI CCO Guide

MORI-IO

Point-to-point communication library for KVCache transfer via RDMA/XGMI

MORI-IO Introduction

MORI-UMBP

Unified memory and bandwidth pool with tiered storage and distributed KV access

MORI-UMBP Overview

MORI-VIZ

Warp-level kernel profiler with Perfetto integration

MORI-VIZ (Visualizer)

Supported Hardware

GPUs: MI308X, MI300X, MI325X, MI355X

NICs: AMD Pollara (AINIC), Mellanox ConnectX-7, Broadcom Thor2 (Volcano under development)

Documentation versions

The page title identifies an exact release tag or a development source commit. Development documentation can describe features that are not in the latest stable package. Check the release notes and the backend restrictions before choosing a deployment configuration.

Indices and tables