Gigabyte recently announced something pretty incredible: the TRX50 AI TOP (trillions of operations per second), a new motherboard specifically designed to train and refine AI locally.
The company says it designed the TRX50 to help utilize the maximum capacity of your CPU, GPU, memory and SSD. With the right hardware, the company says it can support up to 236B parameter large language models (LLMs) of AI training and tuning, all on the device, which can be useful for privacy and security.
The TRX50 is designed to work with AMD’s Threadripper and Threadripper Pro CPUs, so unsurprisingly some of the highest-quality parts on the market. The E-ATX board has four PCIe x16 slots, four NVMe M.2 slots, eight DIMM slots, two 10 Gbps Ethernet interfaces, WiFi 7 via a Qualcomm QCNCM865 module, and more. There’s also surge protection to ensure things don’t get too crazy.
Get to the level of OpenAI
Training LLMs is notoriously difficult and resource intensive, which is one reason OpenAI and its startup rivals are currently raising hundreds of millions of dollars. To that end, Gigabyte offers a convenient user interface in addition to new workflows and real-time progress monitoring.
The TRX50 supports many of the best open source models out there, with a special focus on models aimed at beginners. The AI TOP Tutor can also provide advice on setting up, customizing and generally playing with a model out of the box.
Moreover, Gigabyte also offers other DIY-friendly features such as EZ-Debug Zone to help debug LEDs and control buttons, the WIFI EZ-Plug design to consolidate Wi-Fi antenna plugs, and more.
Democratizing AI training
Like the Bitcoin mining of recent years, it’s hard not to see the AI frenzy as a similar pattern of huge buzz and excitement. Of course, crypto is mainly about the increasing value of coins, so there are important differences, but the hardware needs are quite similar.
Many of the Nvidia GPUs and various CPUs from various manufacturers that found their way into bitcoin mining operations now are repurposed for AIwhich coincided with a general reduction in interest and investments in crypto as a sector.
While Gigabyte is a big name, the TRX50 is clearly a very niche product. At the frontier, Apple built its M-series laptops to handle many AI-related workloads, including training, and the M4 chip comes with a variety of AI features.