Banana Pi BPI-SM10 Ships May 11 — A 60 TOPS RISC-V SBC That Runs 30B Local LLMs

A Genuinely New RISC-V SBC Reaches the Market

The Banana Pi BPI-SM10 single-board computer begins shipping globally on May 11, 2026, and it is the most serious RISC-V SBC the maker market has seen to date. The board is built around the SpacemiT K3 — the world's first RVA23-certified RISC-V AI chip — and delivers up to 60 TOPS of AI compute alongside enough general-purpose performance to run 30-billion-parameter large language models locally on device. For makers building self-hosted AI projects, edge inference deployments, or curiosity-driven experiments with the RISC-V architecture, this is a meaningful step forward for the SBC category.

The launch lineup is two products: the BPI-SM10 (K3-CoM260) developer kit and a 2.5-inch K3 Pico-ITX industrial SBC, both aimed at slightly different deployment shapes but sharing the same SpacemiT K3 silicon. The 8GB developer kit lands at roughly $381.65 with an active heatsink and semi-open case included.

Why 60 TOPS on a RISC-V SBC Matters

The headline 60 TOPS figure puts the BPI-SM10 squarely in the same AI compute neighborhood as NVIDIA's Jetson Orin Nano Super, which has been the default choice for hobbyist and small-team edge AI deployments for the past two years. The structural difference is the architecture — Jetson Orin Nano runs an Arm CPU paired with a CUDA-capable GPU, while the BPI-SM10 runs a RISC-V CPU paired with the SpacemiT K3's dedicated AI compute pipeline. For developers who want to evaluate or commit to RISC-V in a real workload, this is the first SBC where the answer to "but is it fast enough for AI?" is unambiguously yes.

Running 30B-Parameter Local LLMs Is the Real Tell

The most striking capability claim in the BPI-SM10 launch material is on-device inference for 30-billion parameter models in the 30B-A3B family. That is a meaningful threshold for the local LLM use case. A 30B model is large enough to be genuinely useful for general-purpose conversation, summarization, and tool-use workflows; running one fully on-device, with no cloud round trip, is the kind of capability that defines a self-hosted AI box rather than a thin client. With the right RAM configuration, the BPI-SM10 makes that workload feasible on a single board at a price point that fits a maker budget.

How the BPI-SM10 Compares to Jetson and Raspberry Pi 5

The competitive picture is interesting. Against the NVIDIA Jetson Orin Nano Super, the BPI-SM10 is broadly comparable on AI compute headline numbers but uses a different software ecosystem — RISC-V plus the SpacemiT toolchain rather than CUDA. Against the Raspberry Pi 5, which has been hammered by the AI-driven RAM price spike that pushed the 16GB model to over $300 retail, the BPI-SM10 looks more competitive than it would have a year ago. The general-purpose computing on RISC-V is now mature enough that a SpacemiT K3-based board is a credible Raspberry Pi alternative for many maker workloads, and the AI compute is in a different league than the Pi 5's modest NPU.

What the RVA23 Certification Buys You

The SpacemiT K3 is the first RVA23-certified RISC-V AI chip, and the certification matters more than it sounds. RVA23 is the RISC-V profile that locks in the vector and bitmanip extensions modern compilers and Linux distributions are starting to assume. A board that ships with RVA23 certification is a board that runs the modern open-source software stack out of the box — which historically has been the friction point for RISC-V SBCs. The BPI-SM10 starts on the right side of that compatibility line.

The Bigger RISC-V SBC Story

The BPI-SM10 sits inside a broader 2026 wave of credible RISC-V SBCs reaching the maker market. The Milk-V Jupiter from earlier in the year established the SpacemiT K3 platform as a viable foundation for serious boards. The BPI-SM10 takes that platform and packages it into a developer kit form factor with the AI compute headroom to handle modern local LLM workloads. Together, these releases are the clearest signal yet that RISC-V is no longer a curiosity in the SBC space — it is a real architecture choice for makers who want an alternative to Arm and x86 single board computers.

For makers, robotics builders, self-hosted AI enthusiasts, and the broader Raspberry Pi alternative ecosystem, the Banana Pi BPI-SM10 shipping on May 11 is the right kind of release to evaluate. The 60 TOPS of AI compute, the 30B local LLM headroom, the RVA23 RISC-V foundation, and the maker-friendly developer kit format make it one of the most interesting SBC launches of 2026.

Sources: Banana Pi forum announcement, May 2026; Notebookcheck, May 2026; Jon Peddie Research, May 2026; Android Pimp, May 2026.