Product image for illustration; configuration may differ.
NVIDIA DGX GB200
Rack-scale DGX with 72 Blackwell GPUs in Grace Blackwell Superchips, 13.5 TB HBM3e — priced on quote.
| GPU memory | 13.5 TB HBM3e |
|---|---|
| Networking | 72× ConnectX-7, 36× BlueField-3 (200 Gb/s InfiniBand and Ethernet) |
| Total fast memory | 30.2 TB |
| GPUs | 72× NVIDIA Blackwell (in Grace Blackwell Superchips) |
| AI performance | 1,440 PFLOPS FP4, 720 PFLOPS FP8 |
NVIDIA DGX GB200 is built to order and priced per project (configuration, delivery, installation and support).
Check a model against our machines before you buy.
Pick the model and quantisation you plan to run. We'll show whether it fits in memory, how much room is left for context, and the throughput to expect. Estimates come from memory bandwidth and are replaced by our own bench figures where we have them.
Machine
Quantisation
Comfortable. 70 GB spare — room for 128k context, a second small model, or a RAG index in memory. Expect roughly 3.8 tok/s single-stream.
Machines that run this configuration
Weights 42.8 GB at 4.85 bits/param · KV cache 5.0 GB at 16k (fp16, approximate) · 10 GB reserved for OS and runtime. Speed is single-stream decode, memory-bound; prompt processing and batching behave differently. Or WhatsApp an engineer on +44 7515 356 146.
A liquid-cooled, rack-scale DGX system built from Grace Blackwell Superchips: 72 Blackwell GPUs in one NVLink domain for trillion-parameter training and inference. Delivered as a project with facility planning.
Workstation runs this as a project: site survey, power and cooling design, rack layout, installation and cabling, then onboarding onto Base Command Manager or Mission Control, Kubernetes or Slurm. Ongoing monitoring and managed support contracts are available. Support comes from our own UK engineers, and Trade members get free tech support on every order.
Priced on quote: tell us what you need and we will confirm price, availability and lead time.