Technology / Pod 32
The infrastructure is the machine.
Compute, networking, cooling, and power. Designed together as a factory-built module, with dedicated bare-metal access to the cluster.
System architecture
A closer look at the whole system.
A GPU cluster depends on the systems around it. Pacific brings those systems into a repeatable module, then confirms the configuration against the workload and the site.
The interactive model below is an illustrative design study. The reference specification and the agreed engineering plan define the deployment.
One machine. Every system.
Compute, cooling, and power distribution, designed together. A factory-built module replaces the complexity of a one-off construction project.
Integrated by designCompute + fabric
Dedicated hardware. A deliberate configuration.
Pod 32 is built around 32 Supermicro HGX B300 nodes and 256 NVIDIA B300 GPUs. Single-tenant bare metal gives the workload a dedicated physical cluster.
Rail-optimized 400G Ethernet connects the system. Fabric configuration is chosen with engineering, with measured NCCL criteria included in the acceptance program.
- Compute
- 32 Supermicro HGX B300 nodes
- GPUs
- 256 NVIDIA B300
- System memory
- 128 TB
- Storage
- 983 TB raw NVMe per module
- Fabric throughput
- Configurable 3.2 or 6.4 Tbps per node
Thermal + electrical
Density starts with the supporting systems.
Direct-to-chip liquid cooling and redundant coolant distribution units are part of the reference design. There is no site water requirement.
- Reference operating load
- 645 kW
- Rack count
- 4
- Cooling
- Direct-to-chip liquid, redundant CDUs
- Site inputs
- Prepared pad and power feed
Acceptance
The specification is only the starting point.
The 24-stage factory, site, and integration test program produces witnessed acceptance evidence. Defined gates include measured NCCL fabric-performance criteria, so handoff is grounded in recorded results.
The next conversation
Start with your site. Your workload. Your constraints.
Review the capacity, site requirements, and acceptance plan with Pacific.
Discuss your deployment