Overview
- SiFive announced it will add Nvidia’s NVLink Fusion to its high-performance data-center compute platforms to enable coherent CPU–GPU connectivity for AI workloads.
- Reuters reports SiFive is the first RISC‑V design provider to incorporate Nvidia’s interconnect into its chip IP.
- CEO Patrick Little said customer designs using the technology are unlikely to reach market until 2027 or later, and the companies did not disclose financial terms.
- Future SiFive-based silicon with NVLink is reported to gain access to Nvidia’s CUDA‑X software stack, including NCCL and major AI frameworks such as PyTorch and TensorFlow.
- NVLink Fusion has been available for third‑party licensing since 2025 within Nvidia’s OCP MGX rack‑scale ecosystem, which Jensen Huang framed as extending the fabric to RISC‑V.