T-Chip Group
Innovate intelligent hardware and solutions.
Server products and efficient solutions.
StationPC NAS products and services.

ROC-RK3576-PC Motherboard

6 TOPS Computing Power NPU

· Equipped with Rockchip RK3576

· Private deployment of mainstream AI models

· 4K high-frame-rate video decoding

· Supports high-bandwidth LPDDR5 memory

· Built-in 16-megapixel ISP

· Credit card-sized compact design

· Compatible with multiple operating systems

· Provides development materials


Android

LinuxOS

Buildroot+QT

KylinOS

ROC-RK3576-PC Motherboard

Equipped with Rockchip octa-core AI processor RK3576, 6 TOPS NPU for computing power, supports private deployment of modern mainstream large models, and allows for massive storage expansion via M.2 PCIe/SATA SSDs. It boasts 4K high-frame-rate video decoding capabilities and features a compact credit-card-sized design.

AI Processor RK3576

Features the octa-core 64-bit high-performance AIoT processor RK3576, utilizing a big.LITTLE architecture with a main frequency of up to 2.2GHz. Equipped with the Mali-G52 MC3 GPU, its 145 GFLOPS performance enables efficient heterogeneous computing, providing robust support for high-performance computing and multitasking.

Private Deployment

Supports private deployment of ultra-large-scale parameter models based on the Transformer architecture, such as the Gemma series, ChatGLM series, Qwen series, Phi series, and other large language models.

4K Video Decoding

Supports 8K@30fps / 4K@120fps decoding (VP9, AVS2, AV1) and 4K@60fps decoding (H.264/AVC), as well as 4K@60fps encoding (H.264/AVC). Features HDMI 2.1 (4K@120fps), eDP 1.3 (4K@60fps), DP 1.4 (4K@120fps), and MIPI DSI (2560×1600@60fps) interfaces, supporting triple independent display.

LPDDR5 Memory

LPDDR5 offers larger memory capacity, higher bandwidth, faster data transfer rates, lower power consumption, and more advanced Error Correction Code (ECC) technology. Meets the memory space and response speed requirements for private deployment of large models.

Image Processing

Built-in 16-megapixel ISP with support for low-light noise reduction, RGB-IR sensors, and up to 120dB HDR. AI-ISP enhances image quality in low-noise conditions. Supports MIPI-CSI input (MIPI CSI D-PHY: 1×4 Lanes, 2×2 Lanes).

Credit Card Size

Overall dimensions of 93.00mm × 60.15mm, similar to a credit card in size. The compact structure is built with high-specification electronic components and multiple circuit optimization designs to enhance overall system stability.

Operating Systems

Supports Android, Linux OS, and Buildroot+QT, providing a secure and stable system environment for product development and production.

Development Materials

Includes complete source code, tutorials, technical documentation, and development tools, enabling users to efficiently perform secondary development and quickly create independently controllable products.
6 TOPS NPU Computing Power

Built-in NPU with computing power of up to 6 TOPS, supporting INT4/INT8/INT16/FP16/BF16/TF32 mixed operations. Capable of intelligent data processing, voice recognition, and image analysis, meeting the edge computing AI application needs of most terminal devices.

Interfaces
Specifications

ROC-RK3576-PC

Basic Specifications

SOC

Rockchip RK3576

CPU

Octa-core 64-bit processor (4×A72 + 4×A53), frequency is up to 2.2GHz

GPU

G52 MC3@1GHz, support OpenGL ES 1.1/2.0/3.2, OpenCL 2.0, Vulkan 1.1, embedded high-performance 2D acceleration hardware

NPU

6 TOPS NPU, support INT4/INT8/INT16/FP16/BF16/TF32 mixed operation

Decoding/ Encoding

Decoding: 8K@30fps/4K@120fps: VP9, AVS2, AV1, 4K@60fps: H.264/AVC Encoding: 4K@60fps: H.264/AVC

RAM

LPDDR4/LPDDR5 (4GB/8GB/16GB optional)

Storage

eMMC (16GB/32GB/64GB/128GB/256GB optional), UFS3.1 (Optional)

Storage expansion

1 × M.2 (Scalable 2242 PCIe NVMe/SATA SSD), 1 × TF Card

Power

DC 12V (5.5mm × 2.1mm, support 12V~24V wide voltage input)

OS

Android14, Linux OS, Buildroot+QT

Software support

Support the privatization deployment of ultra-large-scale parametric models under the Transformer architecture, such as Deepseek-R1 series, Gemma series, Llama series, ChatGLM series, Qwen series, Phi series and other large language models. It supports traditional network architectures such as CNN, RNN, and LSTM, and supports the import and export of RKNN models; Support a variety of deep learning frameworks, including TensorFlow, TensorFlow Lite, PyTorch, Caffe, ONNX and Darknet. It also supports the development of custom operators. Support Docker container management technology.

Size

93.00mm × 60.15mm × 12.49mm

Weight

≈ 50g

Environment

Operating Temperature: -20℃ ~ 60℃, Storage Temperature: -20℃ ~ 70℃, Storage Humidity: 10%~90%RH (non-condensing)

Interface Specifications

Internet

1 × Gigabit Ethernet (1000Mbps/RJ45), 2.4GHz/5GHz Dual-band WiFi (802.11a/b/g/n/ac), Bluetooth 5.0

Video input

1 × MIPI-CSI DPHY (30Pin-0.5mm,1×4lanes/2×2lanes)

Video output

1 × HDMI2.1 (4K@120fps), 1 × DP1.4 (4K@120fps), 1 × MIPI-DSI DPHY (2560×1600@60fps,1×4lanes,30Pin-0.5mm)

Audio

1 × 3.5mm Audio jack (Support MIC recording,American Standard CTIA), 1 × MIC (2Pin-1.25mm)

USB

1 × USB3.0, 1 × USB2.0, 1 × Type-C (OTG/DP1.4)

Watchdogs

External watchdogs

Extension interface

1 × FAN (4Pin-1.25mm), 1 × Debug(3Pin-2mm), 1 × Double-row pin headers (20Pin-2.0mm, Lead-out: USB2.0, I2C, SPI, SARADC, UART, LineOut)

Customization

Firefly team, with over 20 years of experience in product design, research and development, and production, provides you with services such as hardware, software, complete machine customization, and OEM server.