Superior energy efficiency! Support the private deployment of
mainstream large models. Bring private AI capability to meet
individual AI deployment needs.
The private deployment of large models
The box is equipped with an ARM Mali G52 MC3 GPU, delivering up to 6 TOPS of computing power. This enables advanced intelligent data
processing, speech recognition, and image analysis, effectively fulfilling the AI application demands for edge computing
on a wide range of terminal devices.
RK3576, the new octa-core 64-bit high-performance AIOT processor, features a big.LITTLE architecture (4×A72 +4×A53), an advanced
lithography process, and a frequency of up to 2.2 GHz. It ensures powerful support for high-performance computing and multitasking.
The Mali-G52 MC3 GPU, delivering 145G FLOPS, is capable of supporting efficient heterogeneous computing
to meet the demands of graphics-intensive applications.
This device supports 8K@30fps / 4K@120fps decoding (H.265 / HEVC, VP9, AVS2, and AV1),
4K@60fps decoding (H.264 / AVC), and 4K@60fps encoding (H.265 / HEVC, H.264 / AVC).
With dual 1000Mbps Ethernet, the AI box ensures high-speed and stable network communication,
meeting the needs of various application scenarios.
The AI box features an industrial-grade all-metal enclosure with an aluminum alloy structure for thermal conduction. The side of the
top cover features a grille design for external airflow and efficient heat dissipation, ensuring computing performance
and stability even under high-temperature operating conditions.
Its top cover is a porous hexagonal design, combining elegance with high efficiency. The compact, exquisite device operates stably
and meets the needs of various industrial-grade applications.
Support Linux OS. This provides a safe and stable system environment for product research and production. We offer SDKs, tutorials, technical
documentation, and development tools to streamline and improve the development process.
n deployment of large models, data security, and privacy protection.
Specifications | ||
Basic Specifications | SOC |
Rockchip RK3576 |
Octa-core 64-bit processor (4×A72 + 4×A53), up to 2.2GHz |
||
GPU |
G52 MC3 @ 1GHz, supporting OpenGL ES 1.1/2.0/3.2, OpenCL 2.0, Vulkan 1.1 Built-in high-performance 2D acceleration hardware |
|
NPU |
6 TOPS NPU, supporting INT4/8/16/FP16/BF16/TF32 mixed operations |
|
ISP |
The integrated 16-megapixel ISP supports low-light noise reduction, an RGB-IR sensor, and up to 120dB of HDR. The AI-ISP technology enhances image quality with reduced noise. |
|
VPU |
Decoding: 4K@120fps: H.265/HEVC, VP9, AVS2, AV1; 4K@60fps: H.264/AVC Encoding: 4K@60fps: H.265/HEVC, H.264/AVC |
|
RAM |
LPDDR4 (4GB/8GB optional) |
|
Storage |
eMMC (16GB/32GB/64GB/128GB/256GB optional), UFS 2.0 (optional) |
|
Storage Expansion |
1 * M.2 (optional with USB 3.0 lower layer interface, default USB 3.0 function) (internal to the host), 1 * TF Card |
|
Power |
DC 12V/3A(DC 5.5 * 2.1mm) |
|
Consumption |
Standard power consumption:1.2W(12V/100mA) Maximum power consumption:7.2W(12V/600mA) Sleep power consumption:0.072W (12V/6mA) |
|
OS |
Linux OS(Ubuntu) |
|
Software Support |
・ The private deployment of ultra-large-scale parameter models under the Transformer architecture, such as Gemma-2B, LlaMa2-7B, ChatGLM3-6B, Qwen1.5-1.8B. ・ Traditional network architectures such as CNN, RNN, and LSTM; a variety of deep learning frameworks including TensorFlow, PyTorch, MXNet, PaddlePaddle, ONNX and Darknet; custom operator development ・ Docker container management technology |
|
Size |
93.4mm * 93.4mm * 50mm |
|
Weight |
≈ 500g |
|
Environment |
Operating temperature: -20℃~60℃ Storage temperature: -20℃~70℃ Storage humidity: 10%~90%RH (non-condensing) |
|
Interfaces | Network |
2 * 1000Mbps(RJ45) |
Video output |
1 * HDMI2.1(4K@120fps) |
|
USB |
2 * USB3.0(Limit 1A current) |
|
Other interfaces |
1 * Type-C (AIBOX-3576: Burning)、1 * Console (Debug)、 1 * Power button、1 * MaskRom |
欢迎反馈问题,您的意见与建议是我们的动力!
Copyright © 2014 - 2023 FIREFLY TECHNOLOGY CO.,LTD | 粤ICP备14022046号-2