Liquid-cooled NVIDIA HGX B200 GPU rack with manifolds
HGX B200 rack design guide

Liquid-cooled B200 rows from CDU to acceptance

Design DLC-ready HGX B200 training rows—cold plates, manifolds, fabric, and Fremont soak—without changing facility standards mid-PO.

Rack design guide

DLC-ready HGX B200 without mid-PO facility surprises

NTS designs DLC-ready B200 rows from CDU to acceptance—compute, cooling, fabric, storage attach, and Fremont burn-in on one accountable BOM.

Design inputs include GPUs per rack, north/south NIC plan, facility supply temperature and flow, power density, weight limits, and contract vehicle for hardware plus integration CLINs.

Avoid common failure modes: under-sized CDU, NIC counts that starve collectives, and shipping without leak/flow evidence.

Design & configure

Architecture elements and next steps

Move from reference architecture into GPU SKUs, liquid hubs, staging, and procurement.

HGX B200 compute hosts Compute HGX B200 compute hosts Dual-socket EPYC or Xeon HGX B200 hosts with NVLink/NVSwitch for large-model training.
  • 8-GPU HGX density
  • Documented PCIe/NIC
  • ETO BOMs
Explore HGX B200 compute hosts →
Cold plates, manifolds & CDU Cooling Cold plates, manifolds & CDU Cold plates, manifolds, QDs, and CDU sizing matched to node TDP and facility plant.
  • Flow & QDs
  • CDU redundancy
  • kW worksheets
Explore Cold plates, manifolds & CDU →
Liquid-cooling racks Racks Liquid-cooling racks Advance liquid-cooling rack SKUs with elevations ready for dense B200 rows.
  • Row design
  • Manifold geometry
  • Contract quoting
Explore Liquid-cooling racks →
Fremont burn-in & soak Integration Fremont burn-in & soak Thermal soak, fabric loopback, and leak/flow evidence before freight.
  • Acceptance records
  • Serial manifests
  • Nationwide rack-stack
Explore Fremont burn-in & soak →
NDR / RoCE east-west Fabric NDR / RoCE east-west InfiniBand NDR or RoCE collectives with management OOB kept separate.
  • NIC count planning
  • Loopback soak
  • Cluster bring-up
Explore NDR / RoCE east-west →
HGX B300 buyer’s guide Related HGX B300 buyer’s guide Air vs DLC decision framework that carries forward from B200 to B300 programs.
  • Cooling choice
  • TCO drivers
  • Facility first
Explore HGX B300 buyer’s guide →
SEWP V / ITES-4H / GSA Procurement SEWP V / ITES-4H / GSA Multi-vehicle CLIN/BOM strategy for GPU AI infrastructure from NTS.
  • Vehicle selection
  • Hardware + DLC
  • Integration CLINs
Explore SEWP V / ITES-4H / GSA →
Request a rack design worksheet Next step Request a rack design worksheet Share GPUs/rack, supply temperature, fabric preference, and vehicle for a scoped B200 DLC design.
  • Facility inputs
  • BOM validation
  • Acceptance plan
Explore Request a rack design worksheet →

Frequently asked questions

GPUs per rack, NIC plan (ConnectX/BlueField), facility supply temperature and flow, rear-door vs direct-to-chip preference, kW/rack and weight limits, plus contract vehicle for hardware and integration.

  • Facility plant data
  • Fabric topology
  • Vehicle for CLINs

Dual-socket HGX B200 compute, cold plates/manifolds/QDs/CDU, InfiniBand NDR or RoCE east-west with separate OOB, NVMe plus parallel FS or object tiers, and Fremont thermal soak with fabric loopback.

  • Compute + cooling
  • Fabric + storage
  • Integration evidence

Under-sized or non-redundant CDU for 24/7 training, NIC counts that starve GPU collectives, and shipping without leak/flow evidence that causes on-site acceptance failure.

  • Plant headroom
  • Collective bandwidth
  • Pre-ship soak

Quote compute, cooling, and L11/L12 on SEWP V, ITES-4H, or GSA MAS with BOM validation—preferably the same vehicle for hardware and integration.

  • CLIN-aligned quotes
  • Same-vehicle preferred
  • Architecture review first