SiFive, Inc., has introduced the SiFive Intelligence XM Series for AI workload acceleration. This new product line features an AI matrix engine designed for system-on-chip solutions in Edge IoT, Consumer devices, electric/autonomous vehicles, and data centers. SiFive also plans to release an open-source reference implementation of its SiFive Kernel Library (SKL).
The XM Series incorporates a scalable AI compute engine with integrated scalar, vector, and matrix engines. Each cluster contains four X-Cores, delivering 16 TOPS (INT8) or 8 TFLOPS (BF16) per GHz. Memory bandwidth reaches 1TB/s sustained per cluster, accessible via a high bandwidth port or CHI port for coherent access. The architecture supports systems with no host CPU or those based on RISC-V, x86, or Arm.
This new offering expands SiFive’s Intelligence Product family, focusing on high performance per watt for compute-intensive applications. SiFive reports providing RISC-V solutions to five major technology companies and see opportunities in “software first” design strategies for AI solutions across automotive, data center, intelligent edge, and IoT sectors.
Leave a Reply