Banana Pi Launches World's First RVA23-Certified RISC-V AI Platform with 60 TOPS Performance
Key Takeaways
- ▸First commercially available RVA23-certified RISC-V AI platform with 60 TOPS performance and on-device 30B LLM inference
- ▸Ultra-low power consumption (18W–35W) enables deployment in edge computing, IoT, and battery-constrained robotics applications
- ▸Two form factors—developer kit and compact industrial SBC—provide flexibility for research, commercial, and embedded use cases
Summary
Banana Pi has announced two new AI computing platforms based on SpacemiT's K3 RISC-V processor: the BPI-SM10 Developer Kit and K3 Pico-ITX 2.5-inch industrial SBC. Both products are built around the world's first RVA23-certified RISC-V AI chip, delivering 60 TOPS of AI compute and the ability to run 30-billion-parameter language models locally at speeds exceeding 10 tokens per second.
The SpacemiT K3 platform features eight high-performance X100 RISC-V general-purpose cores and eight dedicated A100 AI compute cores, with up to 130 KDMIPS for general computing. The devices support 4K video encoding at 60fps, operate at just 18W–35W power consumption, and come in memory configurations up to 32GB LPDDR5. This combination positions them as turnkey solutions for edge AI deployment, robotics, industrial automation, autonomous agents, and smart vehicle systems.
The announcement marks a significant step in advancing RISC-V as an open-source alternative to proprietary architectures for AI workloads. With support for local inference of large language models and rich expansion capabilities via NVMe and UFS storage, these products target industries ranging from autonomous systems and industrial control to scientific research and RISC-V ecosystem development.
- Open-source RISC-V architecture strengthens ecosystem alternatives and supports growing demand for privacy-preserving, offline-capable AI
Editorial Opinion
Banana Pi's launch of RVA23-certified RISC-V AI platforms represents a meaningful expansion of the open-source ISA ecosystem into practical edge AI hardware. The ability to run 30B-parameter models locally at viable inference speeds opens new possibilities for privacy-preserving, offline AI applications in robotics, industrial automation, and autonomous systems. However, the real-world impact depends on software ecosystem maturity—developers must adopt RISC-V tooling and compilers at scale for these platforms to gain traction. This is a strong technical offering, but RISC-V adoption still faces the broader market challenge of competing against established x86 and ARM ecosystems.
