SpacemiT K3 vs Nvidia Jetson: 60-TOPS RISC-V AI Boards Released
SpacemiT Inc. launched its K3 RISC-V AI CPU series, delivering 60 TOPS performance and full RVA23 compliance to support 30-billion-parameter LLMs, targeting AI consumer hardware like smart home devices and local AI workloads. The K3 chip features 8 X100 cores and 8 A100 AI cores, supports up to 32GB DDR5 RAM, and ensures compatibility with mainstream Linux distributions and AI frameworks like ONNX Runtime.
SpacemiT Inc., a leading Chinese semiconductor company specializing in RISC-V AI CPUs, has released its latest K3 chip series designed for high-performance AI applications. The K3 integrates 8 high-performance X100 cores and 8 ultra-wide parallel AI A100 cores, delivering 130 KDMIPS of general computing power and 60 TOPS of AI performance. This enables the chip to run large-scale models, including 30-billion-parameter LLMs, while supporting up to 32GB of DDR5 RAM—double the capacity of its predecessor, the K1. The K3 is fully compliant with the RVA23 profile, a standardized feature set for 64-bit RISC-V processors that ensures seamless operation with major operating systems like Linux and Android. This compliance includes support for vector instructions (1024-bit), hypervisor extensions, and advanced AI acceleration, making it developer-friendly without requiring additional software tweaks. SpacemiT’s K3 also includes optional AIA and IOMMU features, further enhancing its versatility for AI workloads. The chip targets AI consumer hardware, including smart home gadgets, AI-powered office tools, content creation devices, and local LLM applications. Its architecture balances affordability and power efficiency, positioning it as a competitive alternative to Nvidia’s Jetson Orin Nano in the embedded AI market. The K3’s design prioritizes efficient parallel compute, high memory bandwidth, and optimized AI acceleration, addressing key challenges in running large-scale models on edge devices. Developers benefit from RVA23 compliance, which standardizes essential CPU features like floating-point operations, multithreading, and vector processing. This eliminates compatibility issues and ensures predictable performance across different hardware setups. SpacemiT’s K3 also supports quantization and model compression, further optimizing its efficiency for AI workloads. The K3 series marks a significant advancement in RISC-V AI processors, offering a scalable and cost-effective solution for deploying LLMs locally. Its release underscores China’s growing influence in the semiconductor industry, particularly in open-source and AI-focused hardware development.
This content was automatically generated and/or translated by AI. It may contain inaccuracies. Please refer to the original sources for verification.