Gigabyte Launches AI TOP 100 B850 Workstation for Local AI Workloads
Gigabyte announced the AI TOP 100 B850 desktop system, powered by an AMD Ryzen 9 9950X processor and designed to support either a single NVIDIA GeForce RTX 5090 or dual AMD Radeon AI PRO R9700 GPUs. The setup is targeted at local AI tasks, achieving 64.9 tokens per second on 32B models with the single RTX 5090 configuration. As local AI execution grows in demand for data privacy and low-latency needs, pre-configured workstations like the AI TOP 100 B850 allow developers and researchers to fine-tune and run large language models on-premises without relying entirely on cloud infrastructure. Built on an AMD B850 motherboard with 16+2+2 phase VRM, the system features 128GB of DDR5 memory, a 1600W 80 PLUS Platinum power supply, dual 10GbE network ports, and Wi-Fi 7 support. Gigabyte claims the dual R9700 GPU configuration can handle inference for models up to 235B parameters and fine-tuning for models up to 110B parameters.
## BACKGROUND
Local AI workstations enable organizations to train and run machine learning models on dedicated hardware on-premises rather than renting cloud GPUs. Metrics such as tokens per second (TPS) measure generation speed during inference, indicating how fast a model output is produced.