ADLINK introduced today the Pocket AI which is a plug-and-play AI accelerator based on the NVIDIA RTX A500 GPU with 4GB GDDR6 64-bit memory. The Pocket AI is compatible with NVIDIA CUDA libraries and it supports Windows 10/11 and Linux.

ADLINK’s product announcement indicates that the Pocket AI uses NVIDIA’s Ampere GA107 architecture which uses an 8nm process.

• NVIDIA Ampere GA107 — 2048 CUDA cores, 64 NVIDIA Tensor Cores (435MHz – 1335MHz), 16 NVIDIA RT Cores, up to 6.54 TFLOPS, 25W TGP

Pocket AI dimensions
(click image to enlarge)

The Pocket AI features a Thunderbolt 3.0 (up to 40Gb/s) interface so it can be used with most modern laptops equipped with USB Type-C ports.

The company states that the Pocket AI can be easily integrated into existing embedded systems so it can be used for other applications besides gaming, for example, machine learning and computer vision.

Pocket AI rear view
(click image to enlarge)

The product brief mentions that this device will be compatible with Windows 10, Windows 11 and Linux. However, its Hot Plug capability will not be supported under Linux.

Pocket AI front view
(click image to enlarge)

Other Pocket AI specifications:

• Memory:
  • 4GB GDDR6 (6001 MHz)
  • 112 GB/s Memory Bandwidth
• USB:
  • Thunderbolt 3.0 (PCIe 3.0 x4)
• Operating Temperature:
  • 0℃ to 40℃
• Mechanical:
  • 106 x 72 x 25mm
  • 250g

Further information

ADLINK hasn’t revealed the price for the AI Pocket, but it will be available for pre-order starting April 2023. Shipping will occur in June 2023 onwards; the product announcement can be found here.