SiFive - 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.
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, Head Corporate Communications