MINIX has officially announced the launch of its T4000 and T5000 Generative AI Mini Workstations, designed to deliver professional-grade AI computing in a compact desktop form factor. These machines are built for tasks such as local large language model (LLM) inference, content creation, and lightweight model training.
Powered by NVIDIA Jetson AGX Thor modules with the Blackwell architecture, the T4000 provides 1200 Sparse FP4 TFLOPs of AI performance, while the T5000 reaches up to 2070 TFLOPs. This level of processing power enables smooth local inference for LLMs ranging from 7B to 70B parameters, allowing businesses to maintain data privacy without relying on cloud-based hardware.
Professional-Grade Hardware and Memory
To ensure efficient performance during intensive AI workloads, the workstations feature high-core-count Arm Neoverse-V3AE 64-bit CPUs, with 12 cores in the T4000 and 14 cores in the T5000. They are equipped with up to 128 GB of high-bandwidth LPDDR5X memory, offering 273 GB/s bandwidth to eliminate bottlenecks in large model operations. The compact units measure just 139.3 x 131 x 76.8 mm and weigh 1420 grams, making them ideal for desktop deployment without the need for server racks. A twin turbo intercooler is included to sustain high-load performance, and the storage system starts with a 1 TB PCIe 4.0 NVMe SSD that is upgradable to 4 TB.
Enterprise Security and Software Integration
The T4000 and T5000 are designed for secure on-premise deployment, ensuring all data and models are processed locally on the device. The workstations come pre-installed with Ubuntu 24.04 LTS and NVIDIA JetPack 7.1, providing full compatibility with the NVIDIA software ecosystem, including CUDA, TensorRT, and NIM. Connectivity options are extensive, featuring dual 10GbE Ethernet for rapid data transfers, Wi-Fi 6E, Bluetooth 5.3, and dual HDMI 2.1 ports for 4K displays. Additionally, the hardware is built for industrial-grade reliability with features such as a 255-level programmable watchdog and Grade 4 ESD immunity.
