QDNASales and integration of LLM inference and training platforms, on-premises or hybrid

DGX Station

Workstation, GB300 Desktop superchip

DGX Station
Memory784 GB advertised, 748 GB usable (252 HBM3e and 496 LPDDR5X)
GPUBlackwell Ultra GB300 Desktop and Grace 72-core
Computearound 20 PFLOPS FP4
NetworkConnectX-8 SuperNIC 800 GbE
Modelsup to 1000 billion parameters
Form factorWorkstation

What the machine runs

Large models on premises with comfortable fine-tuning headroom. Recommended runtime: vLLM with Unsloth.

Positioning

SMBs get a full rack in a single chassis to develop and serve without a server room. A complete station comes in at around 97,000 euros as an estimate.

Manufacturers

The DGX Station is assembled by ASUS, Dell, Gigabyte, MSI and Supermicro around the GB300 board supplied by NVIDIA.

GPUDirect Storage

On platforms fitted with ConnectX cards, GPUDirect Storage technology establishes a direct path between NVMe or NVMe over Fabric storage and GPU memory, bypassing the CPU buffer. Sustained throughput reaches around 50 gigabytes per second per link. Compatible storage arrays, such as NetApp, VAST, DDN or WEKA, feed training and large-scale RAG at full speed.

Support

Open stack supported by QDNA and the communities. Optional NVIDIA AI Enterprise (NIM, NeMo, Triton) with a service level agreement, per card.

Configure this hardware

A no-commitment call to size your platform.

Book a call