Switching and adapters designed for AI traffic between GPUs and nodes.
AI servers move enormous amounts of data between each other. Ordinary networking becomes the bottleneck, so clusters use these switches and adapters instead.
Who it is for
Typical workloads
Previous: —
Next: —
The network switch that joins AI servers together. Standard office switches choke on AI traffic patterns; this one is designed for them.
Memory
Configurable
Deployment
Rack server
From £62,500
ex VAT
Available to order
The highest-performance fabric for training across many servers. It does some of the maths inside the network itself, which shortens training runs.
Memory
Configurable
Deployment
Datacentre
From £74,000
ex VAT
Allocation only
The card that lets one AI server talk to another at full speed. Without it, multi-server training spends its time waiting on the network.
Memory
Configurable
Deployment
Rack server
£4,250
ex VAT
Available to order
A smart network card with its own processors. It handles security and storage traffic so the server's CPUs stay free for AI work.
Memory
Configurable
Deployment
Rack server
£3,650
ex VAT
Available to order