Introduction
On the day before Computex 2024 kicked off, GIGABYTE invited a gaggle of folks to a launch event without any further details beyond muttering something about being Unleashed. Assuming there would be food and beer served, it seemed like the perfect option for the afternoon between meetings that day so we rolled in to check it out.
At the start of the presentation, the big GIGABYTE boss, Eddie Lee, was on stage to introduce the overall concept of GIGABYTE’s AI Top structure which consisted of a Utility, Hardware and a Tutor. To support each of the points, various GIGABYTE folks walked through each of the key points.
You can watch the full presentation in the video below, or keep scrolling to see our summary of it:
AI TOP Utility
The first element of the AI TOP presentation related to their approach of reinventing AI Training. Specifically, GIGABYTE has created a utility that is a GUI wrapper around common text based AI tools to make it easy to manage the process of locally training a LLM at your business without involvement from the cloud. Based on conversations with GIGABYTE, at least one component of AI TOP hardware must be installed in the system to take advantage of the AI TOP Utility.
One of the differentiating factors that GIGABYTE pointed out with its software is that it also allowed the use of SSDs as part of the overall memory for processing AI models. This would allow for larger models to be processed than what might have been possible between GPU and system memory.
AI TOP Hardware
Moving on to the second category of hardware, GIGABYTE is launching “AI TOP” versions of a number of its components ranging from motherboards (of the Threadripper and Intel Workstation variety), GPUs, power supplies and SSDs. Each piece of hardware in the AI TOP line receives an additional year of warranty and is optimized for use in locally hosted AI rigs.
The AI TOP series motherboards are essentially workstation boards with features that better align with loading them up with AI kit. Features like having four PCIe x16 slots, room for four or more NVME SSDs, a pair of Thunderbolt 4 connections and workstation class processors.
On the GPU front, the focus was on the RTX 4070 Ti AI TOP which is produced using a blower fan. This GPU was highlighted because of its power profile and ability to run four of them in a single rig that could still work on a 110v 15A US outlet.
For SSDs, the AI TOP flavor focuses on both performance (towards the upper end of PCIe Gen 4 speeds) but more importantly, improved endurance. For the 2TB model, GIGABYTE is claiming a TBW that is 150x higher than a typical Gen4 PCIe drive.
Finally, on power supplies, GIGABYTE has logically gone for a 1600W flavor of its PSU to crown as the AI TOP PSU. They claim it has been bulked up for endurance and reliability with “server grade” components to be able to run AI workloads day in and day out.
The various hardware introduced included:
- TRX50 AI TOP Motherboard (Threadripper)
- W790 AI TOP Motherboard (Xeon)
- GeForce RTX 4070 Ti SUPER AI TOP 16G
- AMD RADEON PRO W7900 AI TOP 48G
- AMD RADEON PRO W7800 AI TOP 32GB
- GIGABYTE AI TOP 100E SSD (1TB/2TB)
- GIGABYTE UD1600PM PG5 AI TOP
AI TOP Tutor
For the final part of the presentation, GIGABYTE introduced an AI (TOP?) assistant that was, well, you guessed it, AI based and trained on GIGABYTE’s AI TOP product information as well as how to operate AI systems. The AI TOP Tutor could spec out AI systems based on your budget (using… GIGABYTE products, of course) and help you get your rigs up and running.
Overall, the AI TOP Tutor is clearly aimed at GIGABYTE trying to raise the overall education level about AI systems purchasing and is directed at the IT Managers that were told to “go buy some AI systems” without much other guidance.
Pricing
From a pricing perspective, GIAGBYTE is presenting price tiers that are abased on the model size that needs to be processed. Adjusting these rigs can be done with the aforementioned AI TOP Tutor.
For those that are curious, the 8B token model is based on a Threadripper 7960X system with 2x RTX 4070 Ti SUPER AI TOPs, 128GB RAM and a GIGABYTE TRX50 AERO D motherboard. At the top end, they’ve got a Threadripper 7975X selected with 4x W7900 AI TOP GPUs, 1GB of RAM and the GIGABYTE TRX50 AI TOP motherboard.
Conclusion
While the initial thought was that GIGABYTE simply performed a find and replace on “X” for “AI TOP” from its workstation kit when preparing the products for the AI TOP launch, it’s one of those things where the sum is greater than the parts as they’re targeting the small to midsize business that has a need for local AI processing. Given that we did not see their competition making a push for this particular market segment means GIGABYTE could be on to something (or… we’ll all forget about it by CES 2025).