The parts behind AI and compute..

Start with the workload

A useful hardware enquiry begins with the software, model size and intended use. Development, local inference, rendering and server workloads can have very different requirements. Include the software stack and any hardware guidance from its vendor.

GPU and accelerator requirements

Share the required GPU model or class, memory capacity, number of devices and any constraints around interconnects, card dimensions or cooling. Consumer graphics cards, workstation cards and data centre accelerators should be considered against their intended environments.

• Required accelerator and memory capacity
• Number of cards and expansion-slot layout
• Application and driver requirements
• Power connectors and available electrical capacity
• Chassis clearance, airflow and thermal design

The components around the accelerator

Processors, memory, motherboards, storage and networking determine how the wider system supports a workload. Include the platform and chassis details where you already have them.

• CPU socket, platform and PCIe requirements
• DDR generation, capacity and ECC/RDIMM requirements
• Storage interface, capacity and endurance needs
• Networking interface and link speed
• Power supply capacity, cabling and cooling approach

Compatibility needs a complete specification

A shared socket or connector alone does not prove compatibility. Confirm motherboard support lists, firmware requirements, memory organisation, expansion clearance and the software vendor's supported configuration before committing to an order.

Send a component requirement

Use the catalogue to prepare a list, or send an existing bill of materials with quantities and essential specifications. Describe the component supply requirement and identify where you need a particular manufacturer or exact part number.