External graphics docks that let you add discrete graphics to a laptop or desktop computer without opening the chassis have been around for a few years. But with a few exceptions, most eGPUs are big, power-hungry devices that look more like a desktop computer than an accessory.

The new ADLINK Pocket AI could be the most compact eGPU to date. It stuffs an NVIDIA RTX A500 GPU with Ampere graphics into a body that measures 106 x 72 x 25mm (4.2″ x 2.8″ x 1″), making it smaller (but thicker) than a smartphone. In other words, it’s a discrete GPU that you can easily slide into a pocket.

NVIDIA’s RTX A500 GPU is an entry-level discrete GPU meant for professionals. It’s tuned for content creation rather than gaming, but as Tom’s Hardware points out, you can use it for gaming or other tasks, and it should offer comparable performance to an NVIDIA GeForce RTX 3050 GPU.

The GPU has a base clock speed of 435 MHz and support for speeds up to 1335 MHz and 2,048 CUDA cores, 64 NVIDIA Tensor cores, and 16 NVIDIA RT cores. It features 4GB of GDDR6 memory and 112 GB/s memory bandwidth and a 25-watt TGP.

While it’s certainly not the most powerful eGPU around, the 250 gram (8.8 ounce) device is easily one of the most portable, allowing you to easily take a graphics processor home with you from the office without carrying your whole PC, for example.

The ADLINK Pocket AI features two USB Type-C ports: one is a Thunderbolt 3 port for high-speed 40 Gb/s connections to your computer and the other is a power input jack. ADLINK says you should be able to use the USB-C power banks or power adapters rated for 40W or higher USB Power Delivery.

ADLINK hasn’t announced pricing yet, but the company says the Pocket AI will go up for pre-order in April and begin shipping in June, 2023.

press release

Support Liliputing

Liliputing's primary sources of revenue are advertising and affiliate links (if you click the "Shop" button at the top of the page and buy something on Amazon, for example, we'll get a small commission).

But there are several ways you can support the site directly even if you're using an ad blocker* and hate online shopping.

Contribute to our Patreon campaign


Contribute via PayPal

* If you are using an ad blocker like uBlock Origin and seeing a pop-up message at the bottom of the screen, we have a guide that may help you disable it.

Subscribe to Liliputing via Email

Enter your email address to subscribe to this blog and receive notifications of new posts by email.

Join 16,184 other subscribers

Join the Conversation


Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

  1. “it should offer comparable performance to an NVIDIA GeForce RTX 3050 GPU.”

    Tom’s Hardware is likely wrong on this front, this thing comes with a halved memory bus, there’s no way that it’s comparable in performance to an RTX 3050, it’s likely closer to a GTX 1650 going from the available benchmarks of the mobile A500.

  2. Sounds like someone noticed the desire to run neural network applications on your local machine, even when all you’ve got is a laptop.
    But 4 gb of vram just isn’t enough for any of the party tricks that anyone cares about. Nvidia knows this, that’s why they’re making sure you’ll be paying nearly a thousand dollars to get 12gb or more.

    1. A2000 is cheap nowadays, I saw lot of people pair it with sff builds. Around $200 to $600.

      1. I’m seeing the new prices have dropped from USD $700, now it’s down to USD $400. That’s too expensive. At least for this card, the RTX A2000. It’s not worth dropping into SSF Builds like the Dell OptiPlex.

        Those older OptiPlex units will bottleneck the dGPU anyway. It maxes out with the i7-3770 on the 9010, and the other models have the PCie port moved so these cards can only fit into the x4-port. So then that bottlenecks it, even if the newer system had an i7-7700 CPU.

        The best Office PC system for conversion is now the Lenovo ThinkCentre M920s. Get it with i7-8700, 32GB RAM, 1TB SSD for about USD $300 but it’s a bit pricey. It will fit the A2000 again that’s pricey, total is around USD $700, or likely more.

        Better to wait for the M920s to drop to around USD $200 price, and the A2000 to also drop to around USD $250 price or lower. The total should be roughly USD $400-ish for it to make sense for the amount of hardware you’re getting. Frankly I don’t see this happening in 2023, which will mean it’s even more outdated by the time the price drops to acceptable levels. It’s taking too long. We might even see a Valve SteamDeck v2 that’s just as powerful at the same price.

        This hobby started as early as 2008 with the Dell 745, Core2Duo, and GeForce 6200 against the likes of the Xbox 360. It died in 2018 with the Dell 9010, i7-3770, and GTX 1650-LP. It was a good run for those years, but the party’s over.