Skip to content

Component hub

AI / ML Builds

High-VRAM GPUs, Gen4 NVMe storage and supplies sized for sustained multi-card load rather than burst gaming load.

VRAM first, then bandwidth

Model size determines the memory a system actually needs; the finder maps a model to a VRAM and bandwidth requirement before it proposes hardware, so the recommendation starts from the workload instead of the price list.

Power under sustained load

Multi-GPU systems are sized on continuous draw: estimated system wattage, recommended supply minimum with headroom, connector counts for each card, and the annual electricity cost at South African tariffs shown alongside the build.

Lanes and spacing, not just watts

Gen4 NVMe drives share lanes with everything else on the board. Where a board documents lane sharing — a storage slot disabled when a particular SATA port is used, or a graphics slot dropping to half width when a second slot is populated — the limitation is recorded as an attribute of that board and reported by the engine instead of being discovered at boot.

Multi-card systems are also checked physically: card length against case clearance and slot pitch, cooler height against case maximum, and radiator size against what the case actually supports.

Connectors are enumerated per graphics card rather than totalled across the system, because a supply with enough watts and too few plugs is not compatible at all.

Parts on this hub

No parts are listed on this hub yet. Prices and stock appear here as the catalogue is loaded.

Service on this catalogue