Pocket AI – A portable NVIDIA RTX A500 eGPU for AI developers, embedded & industrial applications

We’ve previously seen it’s possible to connect an eGPU to a mini PC through a PCIe x16 to M.2 NVMe adapter or a Thunderbolt 3 port, but while it’s fine to install on your desk for gaming or develop AI applications, the eGPU being larger than most mini PCs, it’s a little too big to integrate into products, and potentially inconvenient to carry around.

ADLINK Pocket AI portable eGPU changes that with an NVIDIA RTX A500 GPU housed in a 106 x 72 x 25mm box that’s about the size of a typical power bank and connects to a host through a Thunderbolt 3 connector. The company says the upcoming eGPU is mostly designed for AI developers, professional graphics users, and embedded industrial applications, but can also be for gaming.


Pocket AI specifications:

  • GPU – NVIDIA RTX 500
    • Architecture – NVIDIA Ampere GA107
    • Base clock: 435 MHz Boost clock: 1335 MHz
    • 2,048x CUDA Cores
    • 64x NVIDIA Tensor Cores
    • 16x NVIDIA RT Cores
    • Performance
      • 100 TOPS (INT8) inference
      • 6.54 TFLOPS single-precision floating point
    • 1x NVENC, 2x NVDEC
    • Memory – 4 GB GDDR6 @ 6001 MHz, 112 GB/s
    • TGP – 25 W
  • Host Interface – Thunderbolt 3.0 (PCI Express 3.0 x 4)
  • Power Supply – USB Power Delivery 3.0+ via Type-C connection on 15V and 40W+ supports
  • Dimensions – 106 x 72 x25mm
  • Weight – 250 grams
  • Temperature Range – 0 to 40°C

Pocket AI Yolov4

The portable eGPU can be connected to a laptop or mini PC with a Thunderbolt 3.0 port and running either Windows 10, Windows 11, or Linux, albeit hot plugging is not supported for the latter. You’ll also need a 40W+ USB-PD power adapter to power the thing.

ADLINK says both NVIDIA CUDA X GPU-Accelerated Libraries for AI and HPC and NVIDIA RTX platform are supported with the latter offering support for ray tracing (OptiX, Microsoft DXR, Vulkan),  AI-accelerated features (NGX), rasterization (Advanced Shaders), simulation (CUDA 10, PhysX, Flex), and asset interchange formats (USD, MDL). I can also see there are running Yolov4 demo from the NVIDIA TAO toolkit.


NVIDIA Jetson platforms are still better suited to design embedded AI systems from the ground up, especially if low power is required, but the small Pocket AI RTX eGPU looks great for developers or to add higher-end AI capabilities to existing x86 systems with a Thunderbolt 3 port. Embedded industrial applications may benefit from 3D rendering, AI-enhanced image processing, and sensor fusion pairing on a mobile workstation.

ADLINK did not disclose pricing yet but says the Pocket AI will provide a route to GPU acceleration at a fraction of the cost of a laptop with equivalent GPU power. Pre-order will start this month, and shipping is scheduled to start in June 2023. Further details may be found on the product page.

Share this:

Support CNX Software! Donate via cryptocurrencies or become a Patron on Patreon

ROCK Pi 4C Plus
Notify of
The comment form collects your name, email and content to allow us keep track of the comments placed on the website. Please read and accept our website Terms and Privacy Policy to post a comment.
Weller PCB manufacturer