Score: 0

Design in Tiles: Automating GEMM Deployment on Tile-Based Many-PE Accelerators

Published: December 15, 2025 | arXiv ID: 2512.13638v1

By: Aofeng Shen , Chi Zhang , Yakup Budanaz and more

Potential Business Impact:

Makes super-fast computer chips easier to program.

Business Areas:
Field-Programmable Gate Array (FPGA) Hardware

Tile-based many-Processing Element (PE) accelerators can achieve competitive performance on General Matrix Multiplication (GEMM), but they are extremely hard to program, as their optimal software mapping is deeply coupled with hardware design which is unwieldy to manual deployment. We propose "Design in Tiles (DiT)", an automated framework connecting a deployment toolchain with a configurable executable model for these accelerators. For evaluation, we apply our framework to GEMM targeting a large acceleration configuration (e.g., 32x32 tiles, 1979 TFLOPS@FP8, 4 TB/s Bandwidth) comparable to an NVIDIA GH200. We achieve higher PE utilization than GH200 with its expert-tuned GEMM libraries, achieving 1.2-2.0x speedup across diverse matrix shapes.

Country of Origin
🇨🇭 Switzerland

Page Count
7 pages

Category
Computer Science:
Distributed, Parallel, and Cluster Computing