Orange Pi has announced the "OrangePi AI Studio Pro" featuring dual Ascend 310 chips. With 352 TOPS of AI computing power and up to 192GB LPDDR4X, it connects to the host via USB4 (Type-C) as an "AI box." This is more of an external AI accelerator than a traditional SBC.
Orange Pi has announced the OrangePi AI Studio Pro. Product information is available on the official Ascend Community (Powered by Ascend) page. This unit is an AI box equipped with two Huawei Ascend 310 chips, delivering 352 TOPS of AI computing power with up to 192GB LPDDR4X memory. It connects to the host PC via USB4 (Type-C), functioning as an "external AI accelerator" for offloading AI inference processing. Rather than connecting sensors via GPIO and network like a typical SBC, this product is designed for operations where the host manages networking and storage while AI processing is delegated to this device.
Table of Contents
Specifications
| ■ OrangePi AI Studio Pro | |
|---|---|
| AI Processor | Ascend 310 × 2 (16 CPU cores, 16 AI cores) |
| AI Performance | 352 TOPS |
| Memory | LPDDR4X 96GB / 192GB (up to 4266Mbps) |
| Local Storage | SPI Flash 32MB × 2 |
| Main I/O | USB4 (Type-C) × 1 |
| Power | DC 12V, 20A (240W class, 5.5×2.5mm jack) |
| Cooling | 12V 4-pin fan connector × 2 |
| Size | 207.7 × 132.6 × 40mm (6mm feet) |
| Supported OS | Ubuntu 22.04.5 (Kernel 5.15.0.126), Windows (planned) |
352 TOPS with dual Ascend 310 and up to 192GB LPDDR4X massive memory packed into a USB4-connected enclosure. However, primary information about network and storage expansion is unconfirmed, with only one USB4 (Type-C) port in a streamlined configuration. It's worth checking if I/O details are published before deployment.
Features
AI Processor (Dual Ascend 310, 352 TOPS)
The primary source confirms that this device features two Huawei Ascend 310 chips, achieving a combined 352 TOPS of AI computing power. The total configuration includes 16 CPU cores and 16 AI cores.
For comparison, NVIDIA Jetson Orin NX (70-100 TOPS class) and Intel Movidius series (tens of TOPS class) are competitors, but this product significantly exceeds them with 352 TOPS. However, since the TOPS calculation conditions (INT8/FP16/Sparse, etc.) are not specified in the primary source, simple numerical comparisons should be avoided.
The official page lists use cases including OCR, object detection, facial recognition, search recommendations, and large language models (multimodal). Combined with the large memory capacity (up to 192GB), this configuration is well-suited for LLM inference and multi-model parallel processing.
Memory and Storage
Memory comes in two configurations: LPDDR4X 96GB or 192GB, with speeds up to 4266Mbps. This is an exceptional capacity for an edge AI board, providing ample room for LLMs and running multiple models simultaneously.
Local storage is listed only as SPI Flash 32MB × 2, with no primary information available about OS placement or update procedures. The operation might assume using host-side storage via USB4, but details await official documentation.
Power and Cooling
Power is supplied via DC jack (5.5×2.5mm) at 12V, 20A, with a 240W class adapter. Given this power consumption range, active cooling with fans is essential by design. Indeed, two 12V 4-pin fan connectors are provided.
The enclosure features ventilation slots, indicating this product is designed for desktop placement with adequate airflow. Noise-sensitive environments should take note of fan acoustics.
Network and Expansion
The only "expansion" elements confirmed in primary sources are the RGB light interface and fan connectors.
Ethernet, Wi-Fi, Bluetooth, GPIO, M.2 (for NVMe/Wi-Fi), and PCIe typically expected on SBCs are not confirmed in primary sources. Since communication with the host occurs through a single USB4 (Type-C) port, an operation model where networking and storage are provided by the host while this device handles AI processing exclusively is the practical approach.
Exterior
The enclosure measures 207.7 × 132.6 × 40mm (6mm feet), a "box-type" device designed for desktop placement. No photos of the bare board are available in primary sources, but the enclosure design suggests a simple industrial equipment aesthetic.
Images of the port side and bottom are not available in official primary sources, so detailed port layout awaits publication.
Additional Notes
OS and Boot
The supported OS is Ubuntu 22.04.5 (Kernel 5.15.0.126), with Windows listed as "planned for future support."
However, OS image download locations and installation procedures are not confirmed in primary sources. The role of SPI Flash (bootloader only or OS storage) is also unclear. We recommend verifying that setup procedures are documented before purchase.
Availability and Pricing
Price and availability information is not confirmed in primary sources. The Orange Pi official site (orangepi.org) and OrangePi China site (orangepi.cn) sales pages were inaccessible at the time of writing due to access issues.
Summary
The OrangePi AI Studio Pro is an "AI box" with an exceptional configuration for the edge AI board market: 352 TOPS from dual Ascend 310 chips and up to 192GB LPDDR4X. It specializes in connecting to hosts via USB4 (Type-C) to offload AI processing.
It's ideal for those who want to run inference with large models or multi-model parallel processing externally—addressing workloads too demanding for typical multi-TOPS AI accelerators but not quite requiring data center deployment.
It's not suited for those who want to connect sensors using GPIO or Ethernet like a traditional SBC. Primary I/O expansion information is scarce, and currently only a host-dependent operation model is apparent.
This is a product whose evaluation will improve as official documentation covering boot procedures and I/O details becomes available. We look forward to further updates.
Related Links
- Ascend Community (Powered by Ascend) Product Page: https://www.hiascend.com/poweredbyascend/hardware/xunlong
- Official PDF (Hardware Characteristics & Key Specifications): https://www.hiascend.com/s/ascendstatic/lst/as3.0/pba20250414/apn/1/pdf_1.pdf
- Orange Pi Official Site: https://www.orangepi.org/