Servers

12 products

Products

An AI server runs models on your own premises instead of renting time in the cloud, which keeps data in the building and the cost predictable. The shelf covers the whole range of the job: 8-GPU HGX rack servers, one of them liquid-cooled, for training and heavy inference, a 4U chassis that takes ten GPUs, a Xeon tower server for everyday business workloads, and a fanless Jetson box that runs video analytics next to the cameras.

Frequently asked questions

What is the difference between a GPU server and a tower server?

A GPU server is built around accelerators. The 8U HGX chassis takes eight SXM5 GPUs (H100 80 GB or H200 141 GB) on one baseboard, so it needs a rack, power planning and serious cooling. A tower server such as the PowerEdge T560 stands on the floor of an office or back room and runs business software and virtual machines on up to two Xeon processors.

Is a liquid-cooled GPU server worth it?

It is when the rack is dense or the room is warm. The liquid-cooled HGX H200 server fits eight GPUs in 5U instead of 8U and moves heat to a coolant loop rather than into the room. It needs a coolant distribution unit or a facility water loop, so plan that before ordering.

Which configuration does the listed price refer to?

Servers ship in many configurations of processor, memory, drives and GPUs. The price on each product page is for the configuration named there; send a Custom Order request with the workload and the parts you need, and we will come back with a quote for that build.

Can I order a single server?

Yes. Every server on this shelf starts from one unit. GPU and accelerator cards on their own are on the GPUs & AI accelerator cards shelf.