This combined hardware-software setup allows you to train AI models at your desk

By necessity, most AI model-training takes place in the cloud, where data centers can support the required storage, compute capacity, and electricity. Companies, let alone individuals, rarely pack the in-house computing punch required. But advancements in tech are rapidly lowering the barrier to entry.

With a solution that combines leading-edge hardware with specialized software, GIGABYTE is making it possible to train your own AI on your desk, unlocking benefits like flexibility, upgradeability, privacy, and security while controlling for the unpredictable costs associated with training in a data center. The company launched its AI TOP solution earlier this year at Computex, combining a suite of AI-ready hardware with GIGABYTE’s AI TOP Utility software. Together, the products make it possible — and cost-effective — for small and medium-sized businesses, research centers, and even individuals of any experience level to train their own AI using household electricity systems.

The AI TOP Utility software was introduced to reinvent users’ workflows with a user-friendly interface. With its upgraded release in October, GIGABYTE has brought users a range of new capabilities, including its capacity to account not only for large language models (LLMs), but also for growth in large multimodal models (LMMs). These are poised to quickly become more efficient and accessible over the next year, as LMMs can digest multiple forms of data like images, text, and video simultaneously. GIGABYTE’s upgraded AI TOP Utility allows users to fine-tune LMMs, meaning AI models are more useful than ever, with the ability to generate images and short videos, not just text. The upgrade also means the software now supports the latest 405B-parameter LLMs, with a range of features that elevate the model.

Outputs improve with AI TOP Utility’s ability to use retrieval-augmented generation (RAG) to choose which model should be selected for a given project. The software takes local AI fine-tuning to the next level by creating custom datasets and offering various preset fine-tuning strategies. Users can schedule the training to run at any time, monitor system hardware through a real-time dashboard, and validate models within the app for instant performance feedback.

The growth of the average AI training model has strained local training capacity, so GIGABYTE applies memory offloading technology, allowing large amounts of data to offload to system memory and SSDs, breaking through the limitations of VRAM and enabling massive scale in a cost-effective way. Finally, the software now supports Linux and Windows Subsystem for Linux (WSL) to round out the feature stack.

Working in tandem with AI TOP Utility is GIGABYTE’s AI TOP hardware, which brings the processing power required to train AI models efficiently and unlock full performance. GIGABYTE’s AI TOP 100 and AI TOP 500 setup kits offer options suited to specific training needs — the AI TOP 100 provides a startup solution perfect for small businesses or personal AI training, while the AI TOP 500 enhances capacity to fit the needs of medium-sized businesses in sectors like finance, engineering, education, and medical research. Either way, GIGABYTE combines a powerful motherboard with a power supply and a chassis, allowing users to customize it with GIGABYTE’s AI TOP graphic cards and SSDs to meet their specific needs.

Maximum performance: GIGABYTE for the first time added the latest generations of Intel and AMD motherboards to the equation. Its AI TOP hardware mix includes the Z890 AORUS XTREME AI TOP, Z890 AORUS MASTER AI TOP, Z890 AI TOP, and X870E AORUS XTREME AI TOP, calibrated for compatibility with the AI TOP Utility. They’re also optimized for dual-GPU configuration and come with cutting-edge Thunderbolt 5 technology, meaning users are equipped with the machines not only for local AI training, but also to maximize their gaming experience.

Enhanced speed and durability: AI TOP SSDs deliver up to 219,000 TBW for intensive AI workloads — nearly 20 times more than enterprise SSDs and 156 times more than consumer SSDs. The solution delivers up to 7,200 MB/s for sequential read and up to 6,500 MB/s for sequential write — enabling ultra-fast processing.

Better efficiency and power tracking: AI model training can require hefty power consumption. GIGABYTE has upgraded its AI TOP power supply to support up to four PCIe Gen 5 graphics cards. Meanwhile, users can, through the Power Meter feature, track accumulated power consumption by kilowatt-hour and see current consumption on the power supply unit’s LCD monitor, helping them create better workloads and manage efficiency.

GIGABYTE also enables users to significantly boost speeds and expand computing power by connecting up multiple AI TOP setup kits through ethernet or Thunderbolt ports, accounting for the rapid growth in model size.

With a suite of powerful hardware to complement its upgraded software, GIGABYTE’s AI TOP solution offers a seamless experience for at-home AI model-training for both AI experts and novices. And, whether they’re seeking the personal AI training experience of AI TOP 100 or desire the enhanced capacity of AI TOP 500, users can pick the model that’s right for them. Either way, users get a powerful local AI training desktop solution delivered in a seamless and flexible package.