For a small studio, a startup or a university lab, the running cost of AI is usually a cloud bill: rented GPUs, per-token charges and data leaving the building. The GIGABYTE AI TOP 100 B850 is a desktop built to move that work back onto hardware you own. Announced on 10 November, the local AI workstation pairs a 16-core processor and 128GB of memory with a choice of graphics setups sized for running, and even fine-tuning, large language models at your desk.
Contents
GIGABYTE AI TOP 100 B850 specs
Both versions share the same core. Inside is AMD's Ryzen 9 9950X with 16 cores and 32 threads, 128GB of DDR5 memory and a 2TB PCIe Gen4 SSD for datasets and model files. The difference is the graphics, and that choice decides what kind of AI work the machine suits.
- NVIDIA GeForce RTX 5090 model: aimed at fast local inference with CUDA software. GIGABYTE quotes up to 64.9 tokens per second when running a 32-billion-parameter model, based on its own internal testing.
- Dual AMD Radeon AI PRO R9700 model: two cards with a combined 64GB of video memory. GIGABYTE says this setup can run models of up to 235 billion parameters and fine-tune models of up to 110 billion.
Put simply, the RTX 5090 version trades memory for speed, which suits chat assistants and agents where quick replies matter. The dual Radeon version trades speed for room, which suits teams that want to load bigger models or train their own variants. GIGABYTE notes that real results depend on the model, precision, software and settings.
Built to run for hours, and to be upgraded
Local AI work tends to mean long sessions at full load, so the cooling and power are sized for it. The system uses a dedicated AI TOP motherboard, liquid cooling for the processor and a high-airflow case, fed by a 1600W 80 PLUS Platinum power supply with headroom for more hardware later.
That headroom is part of the pitch. The AI TOP 100 B850 is built from standard desktop parts rather than a sealed appliance, with two PCIe 5.0 slots for graphics cards and room to add memory and storage as models grow. For connectivity it has two 10GbE network ports for moving large datasets to shared storage, Wi-Fi 7, and USB Type-C at 20Gbps.
On the software side, GIGABYTE's AI TOP Utility handles system monitoring, memory offloading and dataset creation in one interface, aimed at making training and fine-tuning of text and multimodal models less of a command-line job.
Why a local AI desktop matters
GIGABYTE first showed the AI TOP 100 B850 at Computex in June as part of its INFINITY Series and AI TOP showcase. This release is the full launch, with the configurations and figures now set out.
It also lands as the wider PC industry pushes the idea that AI should run on the machine itself. Microsoft made that the theme of its October event, though our own data shows Malaysian readers searching for AI PCs still mostly click on gaming and value laptops. The AI TOP 100 B850 sits at a different end of that market: it is not a laptop with an AI badge but a workstation for people who build and test models.
For a Malaysian business handling customer records, health data or unreleased product work, the main argument is control. Keeping a model and its data on a machine in the office avoids sending sensitive material to an overseas server and replaces a monthly token bill with a one-off hardware cost.
Price and availability in Malaysia
GIGABYTE has not announced Malaysian pricing or a local availability date for the AI TOP 100 B850. Given the parts list, with an RTX 5090 or two workstation graphics cards plus 128GB of memory, expect it to sit well into professional workstation territory rather than gaming PC budgets.
For developers and small teams already paying for cloud GPUs every month, the AI TOP 100 B850 is worth watching. Once a ringgit price lands, the sum is simple: compare it with a year of cloud bills, and decide whether owning the hardware wins.
Image courtesy of GIGABYTE.
