Skip to content

SiFive Intelligence XM Series

An AI compute engine combining scalar, vector and matrix units

Visit official site

What SiFive Intelligence XM Series is

The engine is described as an AI compute design that combines scalar, vector and matrix units for compute-heavy applications with strong energy efficiency.

What you can do with it

  • Run compute-heavy AI training and inference
  • Feed activation layers through the dual vector units
  • Store model data in SRAM through a high-bandwidth port

Getting started

  • Download the SiFive Kernel Library from the site
  • Pick an XM Series configuration for the project
  • Integrate the series into an existing hardware or software setup
  • Use the open-source library during development

Pros & cons

✓ What we like

  • Each cluster offers 1TB/s of sustained bandwidth
  • An open-source kernel library shortens development time

! What to watch out for

  • Integration requires choosing an XM Series configuration
  • New matrix instructions draw source data from vector registers

FAQ

How much bandwidth is listed?

The material states 1TB/s of sustained bandwidth per XM Series cluster.

What memory links are described?

The material names a CHI port for coherent access and a high-bandwidth port to SRAM.

Last reviewed: 2026-11-15

More AI coding tools tools

View all →

How we review