Stories and announcements about Ai are everywhere you look, but you’d be forgiven for not knowing where to begin when it comes to getting involved. After ChatGPT and basic image-creation tools, the Ai learning curve accelerates somewhat but now Gigabyte is here to help. Its new AI TOP technologies allow beginners and enthusiasts alike to train Ai systems at their own desk whether that’s at home or in the office. Here’s what you need to know.
What is Gigabyte AI TOP?
Gigabyte’s AI TOP technologies are spread over three categories: the AI TOP Utility, dedicated AI TOP hardware and an AI TOP Tutor, called Iris. When combined, they enable you to create and train your own Large Language Models (LLMs) on your own premises meaning you can cut costs, optimise performance, guarantee privacy and keep your intellectual property secure, locally.
AI TOP Utility
Gigabyte’s AI TOP Utility is a user-friendly app that supports 236B-parameter LLMs (models with up-to 236 billion parameters) and comes with the top-20 most-popular Ai models (like Llama, BERT and GPT-2) built in.
It eliminates the need for coding by offering adjustable settings that take just a few clicks to configure and allows you to simply tailor your LLMs for Standard, High Precision or High Speed responses depending on your requirements. Experienced Ai users can still tweak models with the command console.
Training LLMs is a time-consuming process so the AI TOP utility provides an at-a-glance ‘loading bar’ and an ‘estimated time of arrival’ for training to complete. All system resources have their status monitored and displayed on a single page to identify any bottlenecks or performance issues. Problems like memory allocation failures can be easily identified meaning you can easily make adjustments to avoid repeated issues on subsequent runs.
The dashboard also has sections that validate your training settings, track whether the quality of your training is progressing as expected (using a loss chart) and log progress and problems when training.
The utility also offers scheduling so that you can train your models using cheaper, off-peak electricity.
Ultimately, it’s never been easier to create a powerful LLM with the AI TOP Utility and its simplified workflow.
AI TOP hardware
Gigabyte is also launching a range of AI TOP PC components and pre-built systems that embrace Gigabyte’s Ultra Durable manufacturing principles. These have all had their design optimised to work with Ai workloads and the AI TOP Utility to ensure maximum performance, reliability and efficiency.
It’s worth noting that these aren’t workstations with special power requirements. They’ve already been extensively tested with AI TOP workloads and so spending more on traditional, expensive, third-party workstation certifications, isn’t required.The AI TOP motherboards are built to maximise the configurations for Ai model training across the CPU, GPU, RAM and SSD(s). Units like the TRX50 AI TOP provide four x16 GPU sockets, support up to 2TB of RAM and can handle four SSDs running in RAID 0. There are also two 10GbE LAN ports with surge protection.The AI TOP motherboards are built to maximise the configurations for Ai model training across the CPU, GPU, RAM and SSD(s). Units like the TRX50 AI TOP provide four x16 GPU sockets, support up to 2TB of RAM and can handle four SSDs running in RAID 0. There are also two 10GbE LAN ports with surge protection.
The AI TOP motherboards are built to maximise the configurations for Ai model training across the CPU, GPU, RAM and SSD(s). Units like the TRX50 AI TOP provide four x16 GPU sockets, support up to 2TB of RAM and can handle four SSDs running in RAID 0. There are also two 10GbE LAN ports with surge protection.
The AI TOP graphics cards are optimised for four-card, dual-slot implementations. They have additional RAM, all-copper heatsinks and ‘turbo fan’ cooling across optimised airflow channels. Models like the GeForce RTX 4070 Super AI TOP 16G can operate in quad-arrays off a single CPU and provide a generous 64GB of VRAM for model training.
Furthermore, Gigabyte is AMD’s only professional graphics card partner and the W7900 AI TOP 48G and W7800 AI TOP 32G offerings have significantly more VRAM (48GB and 32GB respectively) than standard graphics cards. The potential of having a colossal 192GB of VRAM (across four cards) opens the door to creating even larger models.
The AI TOP SSDs support RAID and are also optimised for AI training workloads. The key model is the AI TOP 100E which is available in 1TB and 2TB versions and facilitates 150x Total Bytes Written (TBW) more than a regular SSD, for longevity.
Finally, the AI TOP power supply units use server-grade components and have been extensively tested with AI model-training workloads to ensure stability and reliability. Models like the UD1600PM PG5 AI TOP have double the lifespan of regular PSUs and can support up to four GPUs.
The optimised power efficiency means an AI TOP PC can generate over two-and-a-half tokens per Watt, which is dramatically higher than the less-than-half a token per Watt you can expect from a regular PC.
Everything comes with an extended warranty.
Finally, Gigabyte’s four, pre-built systems provide options to suit different price-versus-performance requirements stretching from consumer to enterprise. They range from a dual-card Nvidia 4070 Ti Super-based system (that suits 8B to 30+B parameter models) which can perform 100K-samples in 288 hours, up to a (70B to 236+B parameter) system which sports four W7900 AI TOP cards and can perform the same workload in just 15 hours.
AI TOP Tutor
Completing the AI TOP trifecta is Airus, the AI TOP Tutor. The cloud-based, human-like Ai can provide set-up assistance, technical support and even consultations with beginners and Ai professionals alike.
So, whatever your budget or level of Ai experience, Gigabyte has you covered with its AI TOP technologies: train LLMs securely and privately on your own desktop and save money on expensive cloud costs and unoptimised computing. To find out more click here.
