Fogwise AIRBox Q900 AI box is an upgrade to the Fogwise Airbox powered by a Qualcomm IQ-9075 SoC with up to 200 TOPS (sparse) of AI performance, 36GB RAM, and 128GB UFS storage.
Radxa says its new AI micro-server competes directly against the NVIDIA Jetson Orin NX 16GB, offering cheaper overall system cost, similar performance, and higher efficiency. Other benefits include Cortex-R52 real-time cores, 2.5GbE networking, and separate GPU, NPU, and DSP.
Fogwise AIRBox Q900 specifications:
- SoC – Qualcomm DragonWing IQ-9075
- CPU
- Octa-core Kryo Gen 6 (Cortex-A78C-based) application cores @ up to 2.36 GHz
- Quad-ore Cortex-R52 real-time cores @ up to 1.85GHz
- GPU – Adreno 663 GPU delivering up to 1.2 TFLOPS FP32 with secure GPU compute; supports Vulkan 1.2, OpenGL ES 3.2, OpenCL 2.0 FP, Adreno NN Direct
- VPU – Adreno VPU 765
- Video Decode
- AV1 / HEVC / H.265 / H.264 / VP9 / MPEG-2
- 1x 8Kp60 / 2x 8Kp30 / 4x 4Kp60 / 8x 4Kp30 / 16x 1080p60 / 32x 1080p30
- Video Encode:
- H.264 / H.265 / HEIF / HEIC
- 2x 4Kp60 / 4x 4Kp30 / 8x 1080p60 / 16x 1080p30
- Concurrent 2x 4Kp60 encode + 2x 4Kp60 decode
- Video Decode
- NPU
- 200 TOPS @ INT8 (Sparse Computing Power)
- Hexagon Tensor Processor (HTP) with quad HVX and dual HMX
- Supports deep-learning frameworks: TensorFlow, PyTorch, ONNX, Paddle, Caffe, DarkNet, etc.
- CPU
- System Memory – 36GB 96-bit LPDDR5 @ 6400MT/s (with ECC support)
- Storage
- 32MB SPI Flash
- UFS & eMMC connector
- 128GB UFS 3.1 (Gear4 x2)
- M.2 M Key Slot (PCIe Gen4 x4)
- Video Output – HDMI 2.0 port up to 4Kp60
- Networking
- 2 x 2.5GbE RJ45 ports with TSN
- Wi-Fi 6 / BT 5.4 via mini PCIe socket
- 4G / 5G cellular via mini PCIe socket + Nano SIM slot
- Up to 3 external antennas
- USB
- USB 3.1 Gen2 Type-A port
- USB 3.1 Gen2 OTG Type-A port
- USB Type-C serial console port
- Misc
- Power and EDL buttons
- PWM fan connector (PWM fan built-in)
- RTC battery connector
- Power Supply – 12V / 5.4A DC via 5.5 x 2.5 mm DC jack
- Dimensions – 104 x 84 x 45 mm (aluminum alloy enclosure)
- Temperature Range – 0°C ~ 60°C
Radxa mentions support for Ubuntu and Yocto Linux, as well as TensorFlow, PyTorch, ONNX, and other AI frameworks. It can run GPT, LLaMa, and Stable Diffusion locally. It still relies on CasaOS for easy deployment, like the original Fogwise Airbox. Performance-wise LlaMA-7B inference takes 0.6 seconds for the first token, and it then speeds up to about 12 tokens per second. More details can be found in the announcement, as the documentation website is not up just yet, as I write this post.
The company also provides a quick comparison between the AIRbox Q900 micro-server and the NVIDIA Orin NX 16GB module:
- Price – Similar range, but Q900 is lower — The Q900 is a complete system going for $600, while Orin NX 16GB module alone costs $600.
- Memory – The Q900 has 36 GB, and the Orin NX has 16 GB. However, the NVIDIA module has faster memory bandwidth: Q900 78 GB/s vs. Orin NX 100 GB/s (higher).
- Storage – The Q900 includes a 128 GB UFS device for the OS, while the Orin NX requires an external SSD.
- AI Performance – 200 TOPS (sparse) for Q900 vs. 157 TOPS for Orin NX (with Super mode enabled).
- GPU/NPU – The Q900’s GPU is independent and does not share performance with the NPU, whereas Orin NX’s GPU shares compute within its 157 TOPS.
- Real-time Cores – The IQ-9075 SoC has 4x Arm Cortex-R52 cores for real-time tasks.
- DSPs – The IQ-9075 includes dedicated audio DSP and general-purpose DSPs, enabling parallel workloads.
- Networking – The Q900 supports 2.5G TSN natively, while Orin NX requires additional expansion.
- Video Codec – The Q900 delivers far stronger encode/decode performance — at least twice as high as the Orin NX 16 GB.
- Power Consumption – The Q900 runs under 20 W in normal operation, while the Orin NX 16GB requires 45 W to achieve 157 TOPS of performance. When the Power Mode is set to 25W, it can achieve around 100 TOPS.
One thing that’s not listed above is that the NVIDIA module might be better for robotics, since it supports a range of GPIOs and MIPI CSI interfaces, which are nowhere to be found on the Qualcomm IQ9 device.
The Fogwise AIRBox Q900 AI micro-server can be purchased now for $599 on Arace. More details may also be found on the product page.
Jean-Luc started CNX Software in 2010 as a part-time endeavor, before quitting his job as a software engineering manager, and starting to write daily news, and reviews full time later in 2011.
Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress. We also use affiliate links in articles to earn commissions if you make a purchase after clicking on those links.
