MINIX tackles on-device LLM and AI processing with new T4000 and T5000 Generative AI Mini Workstation
MINIX has officially unveiled the T4000 and T5000 Generative AI Mini Workstation, positioning both systems as compact yet powerful solutions for professionals working with generative AI, local large language model inference, and enterprise-grade on-premise deployments, while also supporting content creation and lightweight model training in a significantly reduced desktop footprint.
Built around the NVIDIA Jetson AGX Thor modules and leveraging the latest Blackwell architecture, the T4000 delivers up to 1200 Sparse FP4 TFLOPs of AI performance while the T5000 pushes that ceiling to 2070 TFLOPs, supported by 1536 to 2560 GPU cores and fifth-generation Tensor Cores, alongside Multi-Instance GPU capabilities that enable efficient parallel workloads and a dedicated PVA 3.0 vision processing engine. T
This hardware foundation allows the systems to handle local inference for large language models ranging from 7B to 70B parameters, making low-latency, private AI processing increasingly viable for both businesses and individual creators.
To support these demanding workloads, the systems integrate high-core-count Arm-based CPUs, featuring a 12-core configuration on the T4000 and a 14-core setup on the T5000, paired with up to 128 GB of LPDDR5X memory running at 4266 MHz and delivering bandwidth up to 273 GB/s, ensuring sufficient headroom for multi-modal AI tasks and concurrent processing without bottlenecks.
This is complemented by a robust connectivity suite that includes dual 10GbE Ethernet ports, Wi-Fi 6E, Bluetooth 5.3, dual HDMI 2.1 outputs for 4K displays, and a mix of USB 3.2 ports, enabling flexible integration into professional workflows.
Despite its compact size – measuring just 139.3 × 131 × 76.8 mm and weighing 1.42 kg – the workstation incorporates a 1 TB PCIe 4.0 NVMe SSD with upgrade options up to 4 TB, alongside a durable chassis and a twin turbo intercooler system designed to sustain high-load performance without thermal throttling, all while maintaining a plug-and-play desktop deployment model.
On the software side, the systems ship with Ubuntu 24.04 LTS pre-installed alongside NVIDIA JetPack 7.1, offering full compatibility with CUDA, TensorRT, NIM, and containerized AI workflows, while also emphasizing security through fully local processing that avoids cloud dependency, supported by industrial-grade reliability features such as a programmable watchdog system, ESD protection, and wide environmental tolerance.
