HuaKun AI infrastructure: workstation, server or cluster?
Choose between local inference, PCIe accelerators, modular platforms and clusters using memory, software fit, concurrency and a pilot.
ExploreAI compute designs from servers to multi-cabinet SuperPoD systems. SOFCONS connects accelerator topology, data networking, power and heat rejection in one project scope.
Prepare an enquiry
| Model | Applications | Key features |
|---|---|---|
| HuaKun AT900 A2 PoDcManufacturer details ↗ | Cabinet-scale liquid-cooled AI compute building block. |
|
| HuaKun AT9000 A3 SuperPoDManufacturer details ↗ | Liquid-cooled compute cabinets with air-cooled interconnect cabinets. |
|
| HuaKun AT958 B3Manufacturer details ↗ | Air-cooled AI server in standard and supernode versions. |
|
| HuaKun AT928 B5Manufacturer details ↗ | Air-cooled server with supernode interconnect options. |
|
| HuaKun AT9500 SuperPoDManufacturer details ↗ | Multi-cabinet system with liquid-cooled compute and interconnect cabinets. |
|
Choose between local inference, PCIe accelerators, modular platforms and clusters using memory, software fit, concurrency and a pilot.
ExploreCompare cold plate, spray and immersion through node fit, networking, heat rejection, U/OU planning and service procedures.
ExploreTraining data, model partitioning, memory requirements and concurrent jobs determine cluster scale. AT958 B3 and AT928 B5 cover air-cooled servers, while AT900 A2 PoDc and SuperPoD systems represent different cabinet-scale designs.
Accelerator interconnects, data networks and management networks have distinct roles. Storage reads, checkpoint writes and network topology are designed around the training software. Multi-node validation forms part of the acceptance plan.
Cabinet placement, floor loading, power distribution, cooling loops and service access are part of the design. AT9000 A3 SuperPoD combines air and liquid cooling; AT9500 SuperPoD uses liquid cooling for compute and interconnect cabinets.
Combine workload and site information in the enquiry to define compute and facility responsibilities.
Infrastructure, integration or software: start with the objective and the current situation.
Prepare an enquiry