Curated Builds
Pre-assembled, compatibility-checked bundles spanning startup to full datacenter scale. Select up to four to compare side by side, or hand any of these straight to the advisor.
Startup
Cost-optimized, get-started builds for small teams.
Solo Inference Rig
The cheapest real seat at the AI table.
- 1× L4
- 1× EPYC 9654
- 2× DDR5 64GB RDIMM
1 compatibility note(s)
Fine-Tune Starter
Enough card to fine-tune, enough network to not bottleneck it.
- 1× L40S
- 1× EPYC 9654
- 2× DDR5 128GB RDIMM
- 1× ConnectX-7
Small Training Box
Two Hopper GPUs — a real multi-GPU training rig without a DGX price tag.
- 2× H100 PCIe
- 1× EPYC 9654
- 4× DDR5 128GB RDIMM
- 2× ConnectX-7
Midsize
Multi-GPU servers and small clusters, built with room to grow.
Single DGX Pod
One DGX H100. Eight GPUs, one NVLink domain, ready on day one.
- 1× DGX H100
- 1× 3-Phase PDU
- 2× ConnectX-7
Blackwell Growth Server
DGX B200 plus a real fabric and storage — room to grow into a cluster.
- 1× DGX B200
- 1× 3-Phase PDU
- 1× Quantum-2 Switch
- 2× ConnectX-7
- 1× Flash Storage Node
Small Cluster (2-Node)
Two DGX H100 nodes on one fabric — your first real cluster.
- 2× DGX H100
- 1× Quantum-2 Switch
- 2× 3-Phase PDU
- 4× ConnectX-7
- 1× Flash Storage Node
Enterprise / Full-Scale
Full rack-scale systems provisioned end-to-end for datacenter buildouts.
GB200 Single Rack
One full GB200 NVL72 rack — 72 GPUs acting as one, fully provisioned.
- 1× GB200 NVL72
- 1× Rack CDU
- 1× 3-Phase PDU
- 2× Quantum-2 Switch
- 2× Flash Storage Node
2 compatibility note(s)
GB300 Flagship Rack
The most powerful rack NVIDIA ships, fully provisioned end-to-end.
- 1× GB300 NVL72
- 1× Rack CDU
- 1× 3-Phase PDU
- 2× Quantum-2 Switch
- 1× Spectrum SN4700
- 2× Flash Storage Node
2 compatibility note(s)
Multi-Rack Datacenter Buildout
Four GB200 NVL72 racks — a real data-hall-scale buildout.
- 4× GB200 NVL72
- 4× Rack CDU
- 4× 3-Phase PDU
- 4× Quantum-2 Switch
- 2× Spectrum SN4700
- 4× Flash Storage Node
2 compatibility note(s)