-
Notifications
You must be signed in to change notification settings - Fork 35
Expand file tree
/
Copy pathCITATION.cff
More file actions
27 lines (27 loc) · 1.23 KB
/
CITATION.cff
File metadata and controls
27 lines (27 loc) · 1.23 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
cff-version: 1.2.0
message: "If you use Iris or reference it in your research, please cite our work."
authors:
- family-names: "Awad"
given-names: "Muhammad"
- family-names: "Osama"
given-names: "Muhammad"
- family-names: "Potter"
given-names: "Brandon"
title: "Iris: First-Class Multi-GPU Programming Experience in Triton"
doi: "10.5281/zenodo.17382307"
date-released: 2025-10-18
url: "https://github.com/ROCm/iris"
repository-code: "https://github.com/ROCm/iris"
license: MIT
keywords:
- "multi-GPU"
- "distributed programming"
- "RMA"
- "Triton"
- "ROCm"
- "AMD"
- "SHMEM"
- "RDMA"
- "GPU communication"
- "performance portability"
abstract: "Iris is a Triton-based framework providing SHMEM-like Remote Memory Access (RMA) primitives for multi-GPU programming. Developed by AMD Research and Advanced Development, it enables intuitive inter-GPU communication through a simple and expressive API that makes distributed GPU programming as straightforward as single-GPU development. The framework supports both standard Triton and experimental Gluon backends, offering fine-grained control over memory access patterns and communication–computation overlap, with support for AMD MI300X, MI350X, and MI355X accelerators."