资源与支持

SiFive 博客

来自 RISC-V 专家的最新洞察与深度技术解析

September 24, 2024

AI Workload Acceleration with SiFive’s New Intelligence XM Series

At our recent SiFive Day event, SiFive’s CEO Patrick Little explained that “AI and RISC-V were made for each other.” The open RISC-V standard frees developers from the restrictions of proprietary compute, giving them more flexibility to innovate. RISC-V efficiently supports specialized computing engines including mixed-precision operations, and developers can further customize solutions with vector instructions and specialized AI extensions. Additionally, RISC-V has a widely supported software ecosystem.

XM Series

Last week SiFive rolled out our Intelligence XM Series to help companies accelerate high-performance AI workloads for edge IoT, consumer devices, next generation electric and/or autonomous vehicles, data centers, and beyond. The XM Series integrates scalar, vector, and matrix engines to enable extremely memory bandwidth-efficient compute. Featuring four X-Cores per cluster, a cluster can deliver 16 TOPS (INT8) or 8 TFLOPS (BF16) per GHz. With its scalable design, customers can build industry leading levels of performance while still optimizing efficiency since SiFive offers extremely high performance per watt.

Watch Krste Asanovic, SiFive Founder and Chief Architect, discuss "why RISC-V for AI" and give an overview of the new XM Series in the video below or keep reading to learn more.

Watch Krste Asanovic talk about SiFive Solutions for AI

As Asanovic shared, “a flexible and scalable hardware solution is needed to maximize AI software investment.” He noted that there is a “tremendous rate of evolution of AI models, the structure of the models, and the data types they use” so companies need the flexibility to tailor their solutions for market-specific constraints. There is 1TB/s of sustained memory bandwidth per XM Series cluster. Clusters can access memory via a high bandwidth port or via a CHI port for coherent memory access. Additionally, the XM Series supports systems incorporating no host CPU or ones based on RISC-V, x86, or Arm.

The XM Series is the first IP from SiFive to include a highly scalable AI matrix engine, helping customers speed up time to market. The XM Series complements SiFive’s existing X series portfolio in the SiFive Intelligence family, which is designed to connect efficiently to customers’ AI engines. SiFive will continue to invest in advancing both the XM and X product serie. SiFive sees AI as a capability that will be added to virtually all systems, from sensors to supercomputers, so we’re offering our customers more choices so they can choose the IP best suited to their roadmap and market constraints.

To learn more about the XM Series or other AI solutions from SiFive, contact sales.

David Miller
David Miller
Head of Corporate Communications, SiFive

David Miller is Head of Corporate Communications at SiFive, where he leads global corporate communications, executive visibility, media relations, and thought leadership as the company expands adoption of RISC-V across emerging computing markets. He is responsible for shaping the company’s narrative around major business milestones, technology innovation, and industry trends, including the growing impact of AI on semiconductor architectures.

Prior to joining SiFive, David held senior marketing and communications leadership roles at Qualcomm, where he led product marketing and communications for key growth businesses, and previously at CSR and LSI Corporation, where he managed corporate communications through major acquisitions and global brand transitions. He is known for translating complex technologies into compelling stories and building high-performing communications programs across media, analyst, and industry audiences.

Read more Insights from the RISC-V Experts

超越 CPU 与 GPU:SiFive 在数据中心基础设施中的应用
最新文章
超越 CPU 与 GPU:SiFive 在数据中心基础设施中的应用
嵌入 NIC、重定时器、光模块、交换机与安全芯片的 RISC-V 处理器,如何助力现代数据中心突破扩展瓶颈
数据中心 AI 的下一次飞跃:借助 SiFive 智能控制单元 (ACU) 编排定制加速器
最新文章
数据中心 AI 的下一次飞跃:借助 SiFive 智能控制单元 (ACU) 编排定制加速器
随着智能体 AI、多模态模型和实时推理在数据中心及边缘系统中的应用不断拓展,工作负载正变得日益复杂、多样,对计算能力的要求也越来越高。为满足这些需求,计算架构设计人员正转向专用定制加速器,以针对矩阵乘法等特定运算提供更高的性能和能效。然而,这些加速器仍需依靠数据通路之外的软件进行配置、供数、同步和监控,并获得相应的软件支持。
SiFive 视角:RISC-V SPMP 规范正式批准,助力嵌入式安全飞跃发展
最新文章
SiFive 视角:RISC-V SPMP 规范正式批准,助力嵌入式安全飞跃发展
这项最新批准的 SPMP 功能专为增强运行实时操作系统(RTOS)的嵌入式系统中的内存保护而设计。SiFive 非常高兴能在这一进程中贡献力量,并将继续与整个生态系统保持合作,以完善 RISC-V 标准。我们坚信,SPMP 将在整个行业内获得快速普及。