MINIX has announced the T4000 and T5000 Generative AI Mini Workstations, both as compact desktop systems designed for local AI workloads including large language model (LLM) inference, content generation, and enterprise deployment without reliance on external cloud infrastructure.
MINIX T4000 & T5000 Workstations
The systems are powered by NVIDIA Jetson AGX Thor modules based on the Blackwell architecture, delivering up to 1200 Sparse FP4 TFLOPs on the T4000 and up to 2070 Sparse FP4 TFLOPs on the T5000, alongside GPUs with 1536 to 2560 cores and fifth-generation Tensor Cores. The platforms also include Multi-Instance GPU (MIG) capability for parallel processing and a dedicated NVIDIA PVA 3.0 vision processing engine, enabling support for local inference of models ranging from 7 billion to 70 billion parameters.
Both models are equipped with Arm Neoverse-V3AE processors, offering 12 cores on the T4000 and 14 cores on the T5000, paired with up to 128GB of LPDDR5X-8533 memory (up to 273GB/s bandwidth) to support multi-modal AI workloads and concurrent processing. Storage is handled by a 1TB PCIe 4.0 SSD, with support for expansion up to 4TB. Connectivity options include dual 10GbE LAN ports, Wi-Fi 6E, and Bluetooth 5.3, along with multiple USB ports and dual HDMI TMDS outputs supporting 4K displays.
The MINIX workstations are configured for on-device AI processing, allowing data and models to remain local without requiring cloud access. They ship with Ubuntu 24.04 LTS (NVIDIA JetPack 7.1) and support NVIDIA software tools including CUDA, TensorRT, and containerized AI workflows. Additional features include 255-level programmable watchdog functionality, Grade 4 electrostatic discharge protection, and “wide environmental tolerance.”
Pokdepinion: I imagine it won’t be cheap, but on-premise processing is an important capability to have.