Free tech learning for everyone, from basics to pro

Server ArchitectureSystems ArchitectureBeginner

Server hardware: CPU, ECC memory, storage, NICs and redundant power

In short: Server hardware is built to run all day, survive part failures and be managed remotely. ECC memory catches errors, hot-swap drives and power supplies let you replace parts without downtime, and a baseboard management controller gives you access even when the operating system is down.

Why server hardware is different

A server runs services other people depend on, often around the clock. So the hardware is chosen for reliability, serviceability (easy to repair) and remote management, not only speed. Many parts can be replaced while the machine keeps running.

Inside a serverCPU sockets1 or 2, many coresECC memorycorrects bit errorsStorage bayshot-swap drivesRAID / HBAdisk controllerNICs10-100 Gb/s portsDual PSUsredundant powerBMCiDRAC, iLO, IPMIFanshot-swap coolingTPMsecure boot keys
Figure 1. Server components look like PC parts but are built for redundancy, remote management and 24/7 use.

Processor and memory

Server CPUs, such as Intel Xeon and AMD EPYC, offer many cores, large caches and support for lots of memory channels. A dual-socket server has two physical CPUs. More cores help when you run many virtual machines or busy databases; higher clock speed helps single-threaded workloads. Software licensing is often charged per core, so check before buying the biggest chip.

ECC (error-correcting code) memory stores extra check bits with every word of data. It can detect and correct single-bit errors caused by electrical noise or cosmic rays, and detect larger ones. On a desktop a flipped bit might crash a game; on a database server it could silently corrupt records. Servers use registered (RDIMM) modules, which add a buffer so the board can carry many more modules.

Storage, network and power

ComponentOptionsWhat to consider
DrivesHDD (SATA/SAS), SATA SSD, NVMe SSDNVMe is far faster for databases and VMs; HDDs are cheap for bulk capacity and backups.
ControllerHardware RAID card, HBA, software RAIDRAID 1 or 10 for performance and safety; RAID 5/6 for capacity. RAID is not a backup.
Network cards1, 10, 25, 100 Gb/s; copper or SFP fibreUse at least two ports, teamed or bonded, connected to different switches.
PowerDual hot-swap PSUsPlug each supply into a different power feed or UPS, otherwise redundancy is wasted.

A NIC (network interface card) in a server often supports offload features that move work such as checksums from the CPU to the card. Teaming two ports gives both extra bandwidth and protection if a cable or switch fails.

Remote management and security chips

Desktop PC versus serverDesktop PCNon-ECC memorySingle power supplyFew drive baysManaged in personDesigned for office hoursServerECC memoryRedundant hot-swap PSUsMany hot-swap baysOut-of-band BMCDesigned for 24/7 load
Figure 2. The real difference is reliability and manageability, not raw speed.

Every serious server has a BMC (baseboard management controller): a small separate computer on the motherboard. Vendors brand it as Dell iDRAC, HPE iLO or Lenovo XClarity; the open standards are IPMI and the newer Redfish API. Through the BMC you can power the server on and off, watch the console, mount an install image and read hardware sensors, even when the operating system has crashed.

Tip: BMCs are powerful and have had serious vulnerabilities. Put them on a separate management VLAN, change default passwords, and keep their firmware patched.

A TPM (Trusted Platform Module) stores keys used for secure boot and disk encryption such as BitLocker. Together with UEFI Secure Boot it helps prove that the server started with trusted firmware and software.

When sizing a new server, start from the workload: how many VMs or users, how much data, what growth over three to five years. Leave headroom of around a third for peaks and growth, and buy a support contract that matches how quickly you need parts replaced.

Practice

What does ECC memory protect against?
Single-bit memory errors, which it detects and corrects automatically; it also detects many multi-bit errors so they do not silently corrupt data.
A server has two power supplies, both plugged into the same power strip. What is wrong?
The strip is a single point of failure. Each supply should connect to a separate feed or UPS.
Why should the BMC be on its own management network?
It gives full low-level control of the server, so exposing it on the user or internet network would be a major security risk.

Key takeaways

  • Server hardware prioritises reliability, serviceability and remote management.
  • ECC memory detects and corrects memory errors; servers use registered modules.
  • Choose drive type and RAID level for the workload; RAID is never a backup.
  • Team NICs and split dual PSUs across separate feeds to remove single points of failure.
  • Secure the BMC (iDRAC, iLO, IPMI, Redfish) on an isolated management network.
Tracks your learning on this device only. See your progress on the learning paths.

About the author

Mohamed Fayas

Founder of FM TecHub. CCNA certified, with a BTEC HND in Network Engineering, and BEng (Hons) in Computer Networking and Cloud Security. He writes practical, plain-English guides on networking, security and cloud.

More about the author

Original content written by Mohamed Fayas for FM TecHub. © 2026 FM TecHub. You're welcome to share a link to this page; please don't copy the text elsewhere without permission.

Stay up to date by email

New notes, tools and security news from FM TecHub, explained simply. Free.

We only use your email to send these updates. See our privacy policy.