SiFive Unveils Next-Gen RISC-V Cores to Boost AI Chip Performance

SiFive's RISC-V Revolution in AI Chip Design

SiFive, a prominent chip designer, has unveiled its latest generation of RISC-V-based processor core designs, specifically tailored for AI applications. This move signals a significant push to challenge Nvidia's dominance in the AI chip market by offering alternatives to custom ASIC designs 1

1

New Core Designs for Edge AI

The company has introduced two new core designs, the X160 and X180, aimed at low-power applications such as IoT devices, drones, and robotics. These cores feature:

• Support for 128-bit wide vector registers
• 64-bit wide data path
• Ability to be arranged in clusters of up to four cores
• X180 is a 64-bit core, while X160 is based on the 32-bit RV32I instruction set architecture 1

  1

The X160 and X180 are designed to accelerate convolutions for computer vision models and data filtering tasks. They can run real-time operating systems, making them suitable for industrial equipment and consumer devices like fitness trackers 2

2

Upgraded High-Performance Cores

SiFive has also updated its existing X280 and X390 core designs:

• Both feature eight-stage dual-issue in-order execution pipelines
• Support for 512-bit (X280) and 1024-bit (X390) wide vector registers
• Upgraded to the RVA23 instruction set with hardware support for BF16 and OCP's MXFP8 and MXFP4 micro-scaling data types
• Optimized cache hierarchy for improved utilization and die area savings 1

  1

The X390 Gen 2 now boasts 4x the compute and 32x the data throughput of the original X280, allowing for up to 1TB/s of data movement in a four-core cluster configuration 1

1

Related Stories

SiFive Unveils RISC-V Chip Design for High-Performance AI Workloads

SiFive raises $400 million to bring open RISC-V chip designs to AI data centers

SiFive becomes first RISC-V designer to integrate Nvidia NVLink for AI data centers

RISC-V AI Revolution: DeepComputing and Fractile Partner with Andes Technology for Next-Gen AI Hardware

Enhanced Integration and Performance

SiFive has introduced new features to improve integration and performance:

• Scalar Coprocessor Interface (SSCI) for direct accelerator access to CPU registers
• Vector Coprocessor Interface Extension (VCIX) for high-bandwidth access to vector registers
• Upgraded memory subsystem in X280 and X390 to reduce latency 2

  2

Market Impact and Industry Adoption

SiFive's RISC-V-based core designs have gained traction in the industry:

• Used in Google's Tensor Processing Units (TPUs) and Tenstorrent's Blackhole accelerators
• Powering chips from five of the "Magnificent 7" companies (Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, and Tesla)
• Valued at $2.5 billion following its most recent funding round in 2022 1

  1

  2

  2

As the AI chip market continues to evolve, SiFive's latest offerings present a compelling alternative to custom ASIC designs, potentially challenging Nvidia's position in the AI acceleration space.