RTX PRO 6000 server
Air-cooled GPU server, Blackwell Server Edition
| Memory | 96 GB GDDR7 per GPU (2 to 8 GPUs, up to 768 GB) |
|---|---|
| Compute | around 4 PFLOPS FP4 with sparsity per GPU |
| Bandwidth | 1.6 TB/s, PCIe Gen5 |
| Isolation | MIG up to four 24 GB instances |
| Power draw | 600 W per GPU, air-cooled |
| Density | 2 to 8 GPUs per server |
What the machine runs
Several models from 70 to 200 billion parameters in parallel, RAG and multi-GPU fine-tuning. Recommended runtime: vLLM. Backbone of the dwarfstar stack.
Positioning
The sovereign workhorse runs air-cooled on standard servers, with the best cost-per-token ratio.
Manufacturers
Certified servers from Dell, HPE, Lenovo, Supermicro, ASUS and Gigabyte.
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 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 per-card service-level agreement.