What Is a Rack Server?

A rack server is a server in a flat metal chassis made to bolt into a standard 19-inch equipment rack. Rails let it slide out for service, and a full-height rack, often 42U, holds dozens of servers alongside switches, storage and power distribution.

The rack dimensions come from the EIA-310 standard. The front panel is 19 inches (482.6 mm) wide, the opening between the rails is about 450 mm, and height is counted in multiples of the rack unit. One rack unit, written 1U, is 1.75 inches or 44.45 mm. A 2U server is twice that height, a 4U server four times.

Inside, a rack server looks like any other server: processors, ECC memory, drives and network cards. What sets it apart is the packaging. Drives load from the front in hot-swap caddies, power supplies slide out from the back, and fans pull air front to back so a whole row of racks can share a cold aisle and a hot aisle.

Height decides how much fits inside. A taller chassis takes more drives, taller expansion cards and larger heatsinks, and it cools high-power parts more quietly.

Card and drive counts vary widely by model, so treat the expansion column as typical rather than a rule. The rack also needs depth: standard server racks are commonly 1,000 mm or more deep, while short-depth chassis fit shallower telecom and wall-mount cabinets. Leave room in the plan for cable management, a top-of-rack switch, power distribution and a rack-mount UPS, which all take their own share of the rack height.

Rack servers are one of three main server form factors. Each suits a different kind of site.

A high-density server puts more compute into each rack unit. Two common designs are multi-node servers, which fit two or four independent servers into one 2U chassis with shared power supplies, and GPU-dense servers, which fit four to eight accelerators into 2U to 4U.

Density moves the problem to power and cooling. A rack of GPU servers can draw tens of kilowatts, far more than many older server rooms were designed for, so the rack’s power feeds, cooling and airflow have to be planned with the servers. Remote management through a baseboard management controller (BMC) becomes essential when there are too many machines to visit one by one.

Rack servers also run in control rooms, vehicles, ships and telecom cabinets, where the rack is the better form factor for the installation. These sites change the specification: shallower chassis, DC power input, wider operating temperatures and resistance to shock and vibration.

Our rack mount servers cover that range, from 1U edge AI servers with redundant AC or DC power to a 2U fanless rack server rated for wide temperatures and a 2U defence server tested to MIL-STD-810G and MIL-STD-461. For GPU-heavy inference near the machines, our edge AI servers fit several high-power GPUs in a rugged chassis. Our guide to AMD EPYC Embedded for edge servers covers processor choice. A rack mount KVM console gives local screen and keyboard access in the same rack.

We build rack servers with the processors, memory, GPUs, storage and operating system fitted as standard, and validate each one before it ships. Tell us your rack depth, power feeds and workload, and we will specify the server to match.