Tiny Corp: the world’s first AMD eGPU solution over USB3

2025 May 12. 05:30   

Tiny Corp has reached a major technology milestone by creating the world’s first external GPU (eGPU) system that works over a standard USB3 connection. This brings GPU acceleration to platforms that could not use it before, especially Apple Silicon devices.

Until now, external GPUs needed either a dedicated PCIe slot or a Thunderbolt/USB 4 port with PCIe tunneling. These requirements kept many systems—most notably Apple Silicon Macs and MacBooks, which lack native eGPU support due to driver limits and architectural differences—from using eGPUs. Tiny Corp’s solution also supports Windows and Linux, opening the door for anyone who couldn’t consider an external GPU before.

This new eGPU is especially important for AI and machine learning developers who need high computing power for tasks such as prompt processing or model training. By running AMD GPUs over USB 3, Tiny Corp has unlocked faster computing on devices that had few options until now.

The system uses libusb, an open-source library for USB communication across platforms. This method avoids the usual kernel-level GPU drivers, instead using custom user-level drivers to link the host computer with the external GPU.

Rather than trying to duplicate all GPU functions, the solution sends specific compute jobs (kernel executions) to the eGPU, letting AI models use dedicated graphics hardware.

Although this breakthrough is impressive, it has clear limits. The main technical hurdle is bandwidth. USB 3’s maximum theoretical speed is 10 Gbps—much lower than direct PCIe links or Thunderbolt/USB 4 ports. As a result, data transfers to and from the GPU will be slower, especially when loading large AI models or moving a lot of data.

Another limitation is that only AMD GPUs are supported right now, specifically those in the RDNA 3 and RDNA 4 families. There are signs that RDNA 2 support may come later, but current choices are still limited.

Today’s implementation is likely just the start. Tiny Corp has said it will share more detailed technical information once the developers finish refining the code.

Possible future improvements include:

  • Adding support for older AMD GPU architectures

  • Optimizing performance to use USB 3 bandwidth fully

  • Simplifying installation and setup

  • Supporting more types of hardware adapters

If this development succeeds and gains interest, other companies may adopt similar technology. However, because of today’s bandwidth limits, it remains to be seen how widely it can be used. 

Share this post
Thinkless: Fight against the Growing Resource Demands of AI
2025 May 23. Attila Fodor

Thinkless: Fight against the Growing Resource Demands of AI

In recent months, major tech companies have announced a series of reasoning features in their models. However, the immense resource requirements of these systems quickly became apparent, causing the prices of such subscription services to soar. Researchers at the National University of Singapore (NUS) have developed a new framework called "Thinkless", which could significantly transform how large language models (LLMs) handle reasoning tasks. This innovative approach, created by Gongfan Fang, Xinyin Ma, and Xinchao Wang at the NUS xML Lab, enables AI systems to dynamically choose between simple and complex reasoning strategies—potentially reducing computational costs by up to 90%. The framework addresses a critical inefficiency in current AI reasoning methods and represents a major step toward more resource-efficient AI.
Gemini Advanced Strengthens GitHub Integration
2025 May 15. Attila Fodor

Gemini Advanced Strengthens GitHub Integration

 There is no shortage of innovation in the world of AI-based development tools. Google has now announced direct GitHub integration for its premium AI assistant, Gemini Advanced. This move is not only a response to similar developments by its competitor OpenAI, but also a significant step forward in improving developer workflows.
Fake AI video-generation platforms pose a serious threat
2025 May 13. Attila Fodor

Fake AI video-generation platforms pose a serious threat

 Cybersecurity researchers have discovered a new, sophisticated attack method aimed at users of AI tools. Known as the Noodlophile Stealer, this previously unknown malware is spread through fake AI video-generation websites promoted in Facebook groups and on other social media. The attack plays on growing interest in AI content creation and tricks victims with promises of advanced video editing so they will install the malware. It combines social engineering with technical tricks to steal sensitive information—like browser passwords and cryptocurrency wallet data—via a Trojan.
Could the age of the smartphone soon be over?
2025 May 08. Attila Fodor

Could the age of the smartphone soon be over?

 Google’s antitrust trial is now under way, and Eddy Cue, Apple’s senior vice-president of services, has been called to testify. During his testimony, Cue made an unexpected and exciting remark: he suggested that in ten years, we might not need an iPhone—just as surprising as that sounds, it could follow the same path as the iPod.