NVIDIA DGX Station A100, 4x A100 (160GB)

It is a deskside machine, not a rackmount server. The DGX Station A100 is built to sit beside a desk in an office or lab: it runs on a standard wall outlet, cools itself quietly without data center air handling, and needs no rack. Every other AI machine in this catalog is data center hardware; this is the one you can put in a normal room and use like an appliance.

A person, team, or department that wants real multi-GPU AI compute they control, without building infrastructure. AI researchers iterating without queuing for cluster time, data science teams wanting a dedicated training machine, and departments standing up AI capability in their own space. It was designed as a personal AI supercomputer, and that is exactly the role it fills.

A standard wall outlet. The single 1500W supply runs on ordinary 100 to 240V AC, and the machine uses a self-contained refrigerant cooling system quiet enough for an office, so there is no rack, raised floor, three-phase power, or special air handling required. That ordinary-environment design is the whole reason the DGX Station exists.

Four A100 Tensor Core GPUs at 160GB total, a 64-core AMD EPYC 7742, and 512GB of DDR4 RAM. Storage is a 7.68TB NVMe U.2 data drive plus a separate 1.92TB NVMe M.2 OS drive, networking is dual 10GbE, and the whole machine runs from a single 1500W supply. It weighs about 43kg and ships pre-installed with DGX OS.

160GB total across four GPUs is the 40GB-per-GPU A100 configuration. The DGX Station A100 was offered in both 40GB and 80GB versions, and this is the 40GB one. The exact memory configuration of your unit is confirmed in the quote, and MillionMiner can advise if your workload needs the larger per-GPU memory.

Yes, and it was designed for it. The A100's Multi-Instance GPU feature partitions the four GPUs into as many as 28 independent instances, each with isolated memory and compute, so several developers can work on one Station simultaneously as if it were multiple smaller machines. For a small team, that sharing is a large part of the value.

Secondary-market. NVIDIA discontinued the DGX Station line, so these are pre-owned units, which is the realistic way to acquire one now. MillionMiner verifies each unit before shipment and confirms its condition and the 2-year warranty term in the quote, the same secondary-market discipline applied across the catalog's A100 inventory.

For most development and training work, yes. The A100 remains a highly capable Tensor Core GPU, and four of them with 160GB is substantial compute for model development, fine-tuning, and inference. For a team that wants a controlled, in-house machine rather than cloud time or a data center deployment, it is a strong and economical choice on the secondary market.

Yes. It ships pre-installed with DGX OS and NVIDIA's deep-learning frameworks and tools, so it is productive on arrival rather than after a setup project. Plug it into a wall outlet and a network connection, and it is ready to train, which is part of why the DGX Station suits teams without dedicated infrastructure staff.

Submit your requirements through the quote form. A MillionMiner specialist confirms the unit's configuration and memory variant, its condition and warranty, and delivery. Each Station is verified before shipment and shipped worldwide DDP with duties handled, and MillionMiner can advise on rackmount A100 alternatives if a data center deployment suits your situation better.