The XM Series is designed to accelerate SoC development for industries ranging from IoT and autonomous vehicles to data centers. It offers exceptional performance per watt.
SiFive, Inc. has launched its new SiFive Intelligence XM Series, designed to accelerate high-performance AI workloads. This marks the company’s first IP to incorporate a highly scalable AI matrix engine, aimed at helping semiconductor companies fast-track their system-on-chip (SoC) solutions for edge IoT, consumer devices, autonomous vehicles, data centers, and beyond.
SiFive revealed plans to open-source a reference implementation of the SiFive Kernel Library (SKL) to strengthen support for its customers and the broader RISC-V community. This commitment aligns with the company’s strategy to advance RISC-V adoption and foster collaboration across the ecosystem. The company also provided updates on its business roadmap, strategy, and ongoing industry partnerships.
The technical features include:
- Scalable AI Engine:Integrates scalar, vector, and matrix engines.
- Energy Efficient:Delivers high performance per watt for demanding applications.
- Advanced Clusters:Features four X-Cores per cluster for parallel processing.
- High Throughput:Offers 16 TOPS (INT8) or 8 TFLOPS (BF16) per GHz.
- Fast Data Access:Provides 1TB/s sustained memory bandwidth per cluster.
The XM Series introduces a highly scalable AI compute engine by combining scalar, vector, and matrix engines to optimize memory bandwidth. This product builds on the company’s legacy of delivering high performance per watt, addressing the needs of compute-intensive applications across industries. The XM Series offers impressive specifications, featuring four X-Cores per cluster, delivering 16 TOPS (INT8) or 8 TFLOPS (BF16) per GHz, with 1TB/s sustained memory bandwidth per cluster.
“Many companies are embracing open processor standards to stay ahead of the fast-evolving AI landscape,” said Patrick Little, CEO of SiFive. “AI aligns perfectly with SiFive’s strengths, and our ability to offer tailored solutions gives us a unique advantage. We’re proud to be already working with five of the ‘Magnificent 7’ tech giants and leaders in automotive, data centres, and IoT.”
Krste Asanovic, SiFive Founder and Chief Architect, emphasised RISC-V’s suitability for specialised AI applications. “RISC-V was developed to handle complex computing, including mixed-precision operations and AI-specific instructions, which is why many major data centre companies have adopted RISC-V AI accelerators.”
For more information, click here.
