Milk-V Shrinks the RISC-V Jupiter Into the System-on-Module Jupiter NX, Takes Aim at NVIDIA's Jetson

Eight-core SpacemiT chip gets the SODIMM treatment, as Milk-V aims to corner compact low-power edge AI computing.

RISC-V specialist Milk-V has announced yet another device based on the free and open source instruction set architecture (ISA), and this time it's targeting embedded artificial intelligence and looking to take on the mighty NVIDIA Jetson system-on-module range: the Milk-V Jupiter NX.

"[The] Milk-V Jupiter NX is an octa-core AI system-on-module featuring [the] SpacemiT M1/K1 SoC [System-on-Chip], onboard Wi-Fi/BT [Bluetooth], and eMMC [storage]," the company writes of its latest hardware design. "It complies with [RISC-V] RVA22 standards, supports RVV1.0 [the RISC-V Vector Extensions], and is compatible with [NVIDIA Jetson] Nano/Xavier NX baseboards, making it the best replacement for [the Jetson] Nano."

Milk-V has launched another RISC-V device, this time a system-on-module based on the SpacemiT K1/M1 chip: the Jupiter NX. (📷: Milk-V)

The SODIMM-form-factor system-on-module, brought to our attention by Linux Gizmos, is designed to provide competition to the entry-level devices in NVIDIA's Jetson family — to the point of being pin-compatible, meaning the hardware is a drop-in upgrade though without support for NVIDIA's proprietary CUDA general-purpose graphics processor offload language the software is another matter. It's a device the company first unveiled back in August last year, along with a range of other machines and a carrier board designed to accept up to eight Jupiter NX boards for cluster computing projects.

As the name suggests, the Jupiter NX is effectively a system-on-module variant of Milk-V's Jupiter single-board computer. It features the same SpacemiT K1 or M1 64-bit eight-core processor with an Imagination BXE-2-32 graphics processor and an accelerator for artificial intelligence and machine learning (AI and ML) workloads delivering a claimed two tera-operations per second (TOPS) of compute at minimum precision.

There's the choice of 2GB, 4GB, 8GB, or 16GB of LPDDR4x memory, optional eMMC storage, and an edge connector that offers connectivity include two gigabit Ethernet PHYs, one two-lane and two one-lay PCI Express Gen. 2 ports, a single USB 3.0 port shared with one PCIe lane, one USB 2.0 On-The-Go (OTG) and one Host port, HDMI and MIPI Display Serial Interface (DSI) four-lane video outputs, one four-lane and two two-lane Camera Serial Interface (CSI) inputs, and up to 55 general-purpose input/output (GPIO) pins including up to three UART, two I2C, one SPI, and two I2S buses, and up to three pulse-width modulation (PWM) channels.

The company is positioning the module as direct competition to NVIDIA's entry-level Jetson devices. (📷: Milk-V)

Exactly what's available to the end user, of course, depends on the carrier board to which the module is connected. In addition to being a drop-in replacement for carriers designed to fit NVIDIA's Jetson pinout, Milk-V has its own carrier design. Dubbed the Cluster 08, this accepts up to eight modules, offers Non-Volatile Memory Express (NVMe) storage for each, and provides a claimed cluster-wide bandwidth of up to 32Gb/s with each node able to handle up to 16Gb/s.

More information on the Jupiter NX is available on the Milk-V website, along with a purchase link to partner Arace Tech where the device starts at $49.90 for a module with 4GB of RAM and 16GB of eMMC storage.

ghalfacree

Freelance journalist, technical author, hacker, tinkerer, erstwhile sysadmin. For hire: freelance@halfacree.co.uk.

Latest Articles