CORE-K3JD4

System on Modules (SoM) & Computer on Modules (CoM)

Manufacturer part number
CORE-K3JD4
Manufacturer
Firefly

Firefly CORE-K3JD4, listed under “System on Modules (SoM) & Computer on Modules (CoM)”.

Product overview

CORE-K3JD4

8G+128G

( 1 )

Skip to product information

Close

CORE-K3JD4

CORE-K3JD4

Shipping calculated at checkout.

Spec: 8G+128G

8G+128G

16G+128G

32G+128G

60 TOPS RISC-V AI performance, with SpacemiT K3, PCIe 3.0, MIPI camera interfaces and CAN FD for embedded devices.

Decrease quantity

Increase quantity

Manual Wiki Download

CORE-K3JD4 RISC-V Edge SoM

CORE-K3JD4 is a high-performance edge intelligent computing core module built on SpacemiT's latest-generation RISC-V processor K3. Featuring a CPU+AI core fusion architecture, it delivers both powerful general-purpose computing capability and edge-side AI acceleration. It provides 130K DMIPS of general-purpose compute power and 60 TOPS of general-purpose AI inference power, capable of running large language models with up to 30 billion parameters.

RISC-V Processor K3

Integrates 8 high-performance 64-bit computing big cores X100, with a maximum frequency of 2.4GHz and computing power of 130K DMIPS. It is the world's first processor to fully support the RVA23 Profile, delivering strong performance to meet the demands of high-performance computing and complex application scenarios.

Super Parallel Computing

Integrates 8 ultra-wide parallel computing AI cores A100, delivering 60 TOPS of AI computing power. Supports 1024-bit RVV 1.0 parallel computing and the Vector Crypto standard. Equipped with TCM cache and DMA acceleration, it efficiently improves data processing and transmission performance.

RISC-V Hardware Virtualization

Supports RISC-V hardware virtualization, providing CPU and memory virtualization capabilities through the RVH 1.0 extension, interrupt virtualization capabilities through the RV AIA extension, and peripheral virtualization capabilities through the RV IOMMU extension.

High-Level Defense Technology

Adopts complete RISC-V security architecture with M/S/U three‑level privileges, providing hardware‑level defense against Spectre and Meltdown attacks. Equipped with RISC-V PMP/ePMP and dedicated IOPMP mechanism for high‑level protection. Supports secure boot, secure storage, and signature verification, with integrated hardware encryption (AES/SHA/RSA&SM2/SM3/SM4) for full lifecycle security management.

32GB LPDDR5 Memory

Supports up to 32GB of high-bandwidth LPDDR5 memory, paired with UFS2.2 high-speed storage, offering larger capacity, higher bandwidth, and lower power consumption for faster data read/write. It meets the requirements for memory capacity and real-time response speed in private deployment of large AI models.

Support 4K Video Codec

Supports 4K@120fps H.264/VP9 high-definition video decoding, enabling smooth playback of high-bitrate ultra-high-definition video streams with clear and stutter-free visuals. It also supports 4K@60fps H.264 hardware video encoding, delivering efficient video compression, recording, and transmission, saving storage space and bandwidth while maintaining excellent image quality.

Native Linux Mainline Kernel

Powered by world's first RVA23-compliant K3 chip, natively supports Linux mainline kernel (v7.0-rc1). Fully compatible with Ubuntu 26.04 and more. Ecosystem compatibility, stability, and long-term maintainability meet mainstream computing platform standards—ready for desktop productivity, edge LLM inference, and edge server deployments.

Rich Expansion Interfaces

Adopts 260-pin gold finger interface, with physical form and electrical pin definitions compliant with the NVIDIA Jetson Orin Nano standard. Features a wide range of expansion interfaces including I2C, I2S, SPI, UART, CAN, PWM, GPIO, PCIe 3.0, USB 3.0, USB 2.0, MIPI-DSI, MIPI-CSI, and DP, meeting peripheral expansion requirements across various scenarios.

Supports Running 30B Large Models

CORE-K3JD4 delivers 60 TOPS of general AI computing power, supports up to 32GB LPDDR5 memory, and is compatible with the RVV instruction set, native FP8 inference, and multimodal acceleration. It can run 30B-parameter large models offline at the edge, with measured throughput exceeding 10 tokens per second for 30B models, enabling efficient on-device private LLM deployment and low-latency intelligent inference.

Application Scenarios

Widely applicable in edge computing, computing power services, data security, AI education, private deployment of large language models, robotics, AI e-commerce and retail, smart home, AI computing server and intelligent vision inspection.

Edge Computing

Computing Power Services

Data Security

AI Education

Private Deployment of Large Models

Smart Home

Interfaces

Specifications

See also

Technical questions

Ask a question

No published question about this part yet. Questions are reviewed before publication.

Request a quote