EC-OrinNX
CSR2-N72R3399
AIBOX-OrinNX
CT36L/CT36B
AIBOX-OrinNX
38 * 38 mm Mini Size
6 TOPS Edge Computing Power
Private Deployment of Large Models
Meets Industrial Requirements
New-generation octa-core 64-bit high-performance AIOT processor RK3576 adopts a big.LITTLE architecture (4×A72 + 4×A53) , with main frequency of up to 2.2 GHz, providing robust support for high-performance computing and multitasking. Equipped with the Mali-G52 MC3 GPU, its 145G FLOPS GPU enables efficient heterogeneous computing, meeting the demands of graphics-intensive applications.
Built-in 6 TOPS NPU supports INT4 / INT8 / INT16 / FP16 / BF16 / TF32 operations, dual-core collaboration or independent operation, and parallel multitasking and multi-scenario processing. It enables intelligent data processing, voice recognition, and image analysis, meeting the edge AI computing needs of most terminal devices.
Supports the private deployment of ultra-large parameter models based on Transformer architecture, such as the Gemma series, ChatGLM series, Qwen series, Phi series, and other large language models.
Traditional Network Architectures
Multiple Deep Learning Frameworks
Docker Containerization
Supports 8K@30fps / 4K@120fps (VP9, AVS2, AV1) and 4K@60fps video decoding (H.264/AVC), as well as 4K@60fps video encoding (H.264/AVC).
The System On Module adopts a BTB interface design with high-speed industrial-grade connectors, offering ultra-strong transmission capability, high-frequency stability, and solder-free convenience. The System On Module measures only 38 × 38 mm, saving valuable space.
Features industrial-grade capabilities such as real-time networking, Flexbus, hardware resource isolation, and DSMC, meeting the needs of various industrial applications.
Supports RTLinux kernel with excellent real-time performance, widely used in industrial scenarios. Also supports Android 14, Linux OS, and Buildroot, providing a secure and stable system environment for product development and production.
Offers carrier board reference designs and complete technical documentation, enabling efficient secondary development and rapid product customization.
Suitable for edge computing, localized large models, machine vision, industrial cameras, industrial control hosts, automotive electronics, and other industries.
iCore-3576Q38 | ||
Basic Specifications |
SOC |
Rockchip RK3576 |
CPU |
Octa-core 64-bit processor (4×A72 + 4×A53) with a maximum frequency of 2.2GHz |
|
GPU |
G52 MC3@1GHz, support OpenGL ES 1.1/2.0/3.2, OpenCL 2.0, Vulkan 1.1, embedded with high-performance 2D acceleration hardware |
|
NPU |
6 TOPS NPU, It supports INT4/INT8/INT16/FP16/BF16/TF32 operations, supports dual-core collaborative or independent work, and supports multi-task and multi-scene parallelism |
|
ISP |
Built-in 16 million pixel ISP, support low-light noise reduction, support RGB-IR sensor, support up to 120dB HDR, AI-ISP to improve low-noise image effect |
|
Decoding/ Encoding |
Decoding: 8K@30fps/4K@120fps (VP9, AVS2, AV1), 4K@60fps (H.264/AVC) Encoding: 4K@60fps (H.264/AVC) Image codec: 4K@60fps MJPG |
|
RAM |
LPDDR4/LPDDR4x (4GB/8GB/16GB optional) |
|
Storage |
eMMC (16GB/32GB/64GB/128GB/256GB optional), UFS2.0 (Optional) |
|
Power |
5.0V (voltage tolerance ± 5%) |
|
OS |
It supports RTLinux kernel and has excellent real-time performance, which is widely used in industrial application scenarios Support Android14, Linux OS, Buildroot, provide a safe and stable system environment for product research and production It has new industrial features such as real-time network, Flexbus, hardware resource isolation, and DSMC to meet the needs of different industrial applications |
|
AI Performance |
Support the privatization deployment of ultra-large-scale parametric models under the Transformer architecture, such as Gemma 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. It supports the real-time object detection algorithm YOLO (You Only Look Once), which is fast and real-time compared with traditional object detection methods, and can accurately identify and locate multiple target objects in images or videos, powering AI applications. |
|
Interface type |
BTB (2 × BTB Socket(100Pin)) |
|
Size |
38.0mm × 38.0mm × 5.68mm |
|
Weight |
≈10g |
|
Environment |
Operating Temperature: -20℃- 60℃, Storage Temperature: -20℃ ~ 70℃, Storage Humidity: 10%~90%RH (non-condensing) |
|
Interface Specifications |
Internet |
2 × GMAC (providing RMII or RGMII interfaces to connect external Ethernet PHY; Supports 10/100/1000 Mbps speeds), expandable WiFi, Bluetooth, 4G, 5G through USB, PCIE, SDIO, UART |
Video input |
2 × MIPI DPHY CSI (Supports MIPI V1.2 version; 1 × 4 Lanes or 2 × 2 Lanes) 1 × MIPI DCPHY CSI RX (DPHY supports V2.0 version with 4Lane/2Lane/1Lane modes; CPHY supports V1.1 version with 0/1/2 Trio modes) |
|
Video output |
1 × HDMI2.1(4096×2160@120Hz)/eDP1.3(4096×2160@60Hz, supports 1Lane/2Lane/4Lane modes) 1 × DP1.4 (4096×2160@120Hz) 1 × EBC Output interface (support E-ink EPD (Electronic Paper Display), 2560×1920) |
|
Audio |
4 × SAI (4T/4R) 3 × SAI (1T/1R) Supports I2S/TDM/PCM mode, and supports sampling rates up to 192KHz |
|
PCIe/SATA |
1 × PCIe 2.1/SATA 3.1/USB 3.2 Gen1 Combo interfaces 1 × PCIe 2.1/SATA 3.1 Combo interfaces |
|
USB |
1 × USB3.2 Gen1 OTG0 (multiplexed with DP1.4) 1 × USB3.2 Gen1 OTG1 (multiplexed with PCIe 2.1/SATA 3.1) |
|
SDIO |
2 × SDIO3.0 |
|
PWM |
14 × PWM |
|
SPI |
4 × SPI (Supports serial master and serial slave modes, software configurable) |
|
I2C |
9 × I2C (Supports 7-bit and 10-bit address modes, data rates up to 100kbps in standard mode and up to 400kbps in fast mode) |
|
I3C |
1 × I3C (I2C compliant, SDR mode supported, up to 10 devices supported) |
|
UART |
12 × UART (Support automatic flow control mode, support RS485 function) |
|
CAN |
2 × CAN FD (Supports 8192 bit receive FIFO) |
|
SARADC |
3 × SARADC + 1 × SARADC (boot only), supports 12-bit resolution, up to 1MS/s sampling rate |