SiFive announced plans to integrate NVIDIA NVLink Fusion into its high-performance, data-center-class RISC-V compute platforms, enabling coherent, high-bandwidth connectivity between SiFive CPUs and NVIDIA GPUs and accelerators. The move targets next-generation AI data centers where training and inference workloads increasingly stress power budgets and data movement efficiency.
AI system architects are rethinking CPU–GPU coupling as workloads scale faster than available power and cooling envelopes. SiFive positions its customizable RISC-V cores, coherent fabrics, and memory hierarchies as a way for cloud providers and system vendors to tailor CPUs to specific AI workloads while retaining software portability across the RISC-V ecosystem.
By adopting NVLink Fusion, SiFive platforms gain direct, coherent links to NVIDIA accelerators, reducing latency and improving utilization compared with traditional CPU-to-accelerator interconnects. SiFive said aligning its roadmap with NVLink Fusion supports heterogeneous, co-designed AI systems that combine open CPU architectures with high-bandwidth interconnects.
- Integration of NVIDIA NVLink Fusion with SiFive data center RISC-V platforms
- Coherent, high-bandwidth CPU–GPU connectivity for AI training and inference
- Focus on performance per watt and efficient data movement at scale
- Support for heterogeneous systems combining CPUs, GPUs, and custom accelerators
- Target customers include hyperscalers, cloud providers, and system vendors
“By integrating NVLink Fusion with SiFive’s high-performance compute subsystems, we’re enabling customers with an open and customizable CPU platform that pairs seamlessly with NVIDIA’s AI infrastructure to deliver efficiency at data center scale,” said Patrick Little, President and CEO of SiFive.
🌐 Analysis
The announcement reflects a broader industry push toward tighter CPU–GPU coupling as AI infrastructure shifts to co-designed architectures optimized for power efficiency and scale. It also places RISC-V more directly into competition with Arm and x86 platforms that already support advanced coherent interconnects for AI data centers.
Founded in 2015 by Dr. Krste Asanović and Dr. Yunsup Lee—co-creators of RISC-V at UC Berkeley—SiFive’s mission is to enable customizable, power-efficient compute platforms spanning embedded, edge, automotive, and data center workloads. The company’s core technology includes configurable RISC-V CPU IP, SoC platforms, and design services, with flagship offerings such as the Intelligence and Performance core families targeting AI acceleration and high-end compute. SiFive has raised over $350 million in venture funding from investors including Sutter Hill Ventures, Spark Capital, and Qualcomm Ventures, and works with a broad ecosystem of silicon partners, foundries, and system vendors. Key milestones include the commercialization of RISC-V processor IP at scale, expansion into data center-class designs, and strategic collaborations aimed at positioning RISC-V as a viable alternative in AI and high-performance computing infrastructure.
🌐 We’re tracking the latest developments in networking silicon. Follow our ongoing coverage at: https://convergedigest.com/category/semiconductors/
