SpacemiT K3 (RVA23) · 8× X100 CPU + 8× A100 AI Cores · Up to 32GB LPDDR5 · 256GB UFS · Pico-ITX · Wi-Fi 6
SpacemiT K3 (RVA23)
8× X100 CPU @ 2.4GHz
8× A100 AI cores
2× RT24 real-time cores
Up to 60 TOPS
1024-bit RVV 1.0 Vector
INT4/INT8/FP8/INT16/FP16/BF16
RISC-V RV64GCVB
RVA23 compliant
Vector 1.0 extension
IMG BXM-4-64-MC1
OpenGL ES 3.2, Vulkan 1.3
OpenCL 3.0
LPDDR5 up to 32GB
6400 MT/s, 64-bit dual-channel
UFS 2.2 up to 256GB onboard
M.2 2280 NVMe (PCIe Gen3 x4)
1× 10GbE SFP+
1× GbE RJ45
Wi-Fi 6 + BT 5.2
USB-C DP Alt (4K@60Hz)
eDP (2.5K@90Hz)
Decode: H.265/H.264/VP9 4K@120fps
Encode: H.265/H.264 4K@60fps
USB 3.0 Type-A × 2
USB 2.0 Type-A × 2
USB-C (OTG, USB 2.0)
40-pin GPIO header
Pico-ITX
Aluminum case included
Runs 30B-class LLMs
at >10 tok/s
Ubuntu 24.04 LTS
Fedora, Debian
8GB LPDDR5 · $300
16GB LPDDR5 · $380
32GB LPDDR5 · $575
Available
Quote
The Milk-V Jupiter 2 is the most capable RISC-V single-board computer for edge AI, built around the SpacemiT Key Stone K3 SoC. With eight high-performance X100 RISC-V CPU cores, an eight-core A100 AI subsystem delivering up to 60 TOPS, and a compact Pico-ITX form factor with aluminum case, it brings production-grade RISC-V compute to AI developers, researchers, and edge deployment teams.
The K3's heterogeneous architecture combines 8× X100 application cores (up to 2.4 GHz), 8× A100 AI cores with dedicated TCM and DMA acceleration, and 2× RT24 real-time cores — enabling simultaneous AI inference, real-time control, and Linux application workloads on a single chip. The 1024-bit RVV 1.0 vector engine supports INT4 through BF16 precision, making it capable of running 30-billion-parameter LLMs at over 10 tokens per second entirely on-device.
With 10GbE SFP+ networking, dual M.2 slots (NVMe + wireless), up to 32GB LPDDR5 at 6400 MT/s, and 256GB onboard UFS storage, the Jupiter 2 has the I/O bandwidth and memory capacity to serve as a development platform, edge inference node, or network-attached AI accelerator. It's the first RISC-V board on QS Compute — opening the door to open-ISA AI acceleration for cost-sensitive and sovereignty-conscious edge deployments.
On-device LLM inference (30B-class models), RISC-V AI software development and benchmarking, edge AI gateway with 10GbE uplink, robotics perception and control, privacy-preserving local AI processing, academic and research RISC-V computing.
Need Milk-V Jupiter 2?
The first RISC-V AI development platform on QS Compute. Available in 8GB, 16GB, and 32GB configurations. Run 30B LLMs at the edge — open ISA, no vendor lock-in.
Request Quote