Server Buying Guide

Buying a server is one of the most significant IT investments your business will make. Get it right and it will serve you reliably for five to seven years. Get it wrong and you'll be back at square one within eighteen months. This guide covers every dimension of the decision — from workload profiling and form factor through to power, cooling, warranty, and total cost of ownership.

At Server Warehouse, we supply brand-new HPE, Dell, and Lenovo servers exclusively. Everything below is written to help you arrive at the right specification before you buy.

1. Define Your Workload First

Every hardware decision flows from what the server will actually do. Common workloads and their primary demands:

2. Form Factor — Tower, Rack, or Blade

If you don't yet have a rack, factor in the cost of a server cabinet, PDU (power distribution unit), and cable management when comparing tower vs. rack pricing.

3. Processor (CPU)

• ›Core count: More cores handle more simultaneous tasks. Virtualisation and databases benefit most from high core counts (16–32+ cores per socket). File servers need far fewer.
• ›Clock speed: Higher GHz helps workloads that run single-threaded tasks (some ERP and accounting software). Less important than core count for virtualisation and parallel workloads.
• ›Single vs. dual socket: Entry servers have one CPU socket. Mid-range and enterprise servers have two — doubling potential CPU and RAM capacity. Consider a dual-socket model even if you only populate one CPU initially.
• ›Generation: Newer CPU generations deliver better performance per watt. An older generation at the same price is rarely the better deal once energy costs over five years are factored in.
• ›Intel Xeon vs. AMD EPYC: Both are excellent enterprise choices. AMD EPYC generally offers more cores per rand at mid-range. Intel Xeon has broader ecosystem support and specific advantages for some enterprise software licensing models.

4. Memory (RAM)

• ›Buy more than you need today: RAM is one of the cheapest ways to prevent performance problems. Provision at least 30% more than your current peak usage.
• ›ECC RAM is mandatory: Error-Correcting Code (ECC) memory detects and corrects single-bit memory errors silently. Desktop (non-ECC) RAM has no place in a server — a single uncorrected memory error can corrupt a database or crash a VM host.
• ›Check DIMM slots and maximum capacity: A server with 8 slots × 64GB = 512GB maximum. A server with 16 slots × 128GB = 2TB maximum. Choose a platform that gives you room to grow without replacing DIMMs you've already bought.
• ›DDR4 vs. DDR5: Newer platforms use DDR5 which is faster and more energy-efficient. DDR4 remains widely available and cost-effective. Both are enterprise-grade when ECC variants are used.
• ›Typical starting points: File server 16–32GB · ERP/accounting 32–64GB · Database 64–256GB · Virtualisation host 128GB–1TB · AI/analytics 256GB+

5. Storage

RAID — Always Configure Redundancy

• ›RAID 1: Two drives mirror each other. Survives one drive failure. Simple and reliable for OS volumes.
• ›RAID 5: Minimum 3 drives, one parity stripe. Good balance of performance, capacity, and redundancy. Survives one drive failure.
• ›RAID 6: Like RAID 5 but survives two simultaneous drive failures. Recommended for arrays with 6 or more drives.
• ›RAID 10: Mirrored pairs striped together. Excellent performance and redundancy. Requires an even number of drives. Best for high-transaction databases.
• ›Hardware RAID controller: Always use a dedicated hardware RAID controller (HPE Smart Array, Dell PERC, etc.) rather than software RAID. Hardware controllers have a battery-backed write cache that prevents data corruption during a power failure mid-write.

6. Networking & Connectivity

• ›1GbE vs. 10GbE vs. 25GbE: Most servers ship with 1GbE onboard. If your workload moves large amounts of data between servers — backups, VM live migration, NAS traffic — 10GbE or 25GbE makes a significant difference. Factor in switch compatibility.
• ›Dual NICs for redundancy: Bond two network interfaces together (NIC teaming / LACP) for failover and increased throughput. If one NIC fails, the server stays online.
• ›Out-of-band management (iLO / iDRAC / XClarity): A dedicated management port lets you access, reboot, and troubleshoot a server remotely — even if the OS is unresponsive. This is not optional for a production server.
• ›PCIe expansion slots: Verify the server has enough PCIe slots for additional NICs, HBAs (for SAN connectivity), or GPU cards you may need.

7. Power Supply & Redundancy

• ›Redundant PSUs: Always configure dual power supply units. If one fails, the server continues running on the other. A single PSU failure on a non-redundant system means immediate downtime.
• ›PSU wattage: Size your PSUs to run at 50–70% of rated capacity under full load. A PSU running at 100% continuously degrades faster and runs hotter. If your server draws 600W peak, use 800W PSUs.
• ›80 PLUS certification: Look for Titanium or Platinum efficiency ratings. An 80 PLUS Titanium PSU is 94%+ efficient at 50% load. Over five years the electricity saving on a 500W server is measurable in thousands of rands.
• ›UPS (Uninterruptible Power Supply): Every production server needs a UPS. South Africa's grid instability makes this non-negotiable. Size the UPS to provide at least 10–15 minutes of runtime to allow for a graceful shutdown or generator transfer.

8. Cooling & Environment

• ›Ambient temperature: Server rooms should be maintained between 18°C and 27°C. Every 10°C above optimal operating temperature roughly halves the expected lifespan of electronic components.
• ›Hot aisle / cold aisle: In rack environments, alternate rack orientations so server exhaust (hot air) faces a hot aisle and intakes face a cold aisle. This prevents hot and cold air mixing and dramatically improves cooling efficiency.
• ›Noise: Rack servers run their fans at high speed — they are loud. Do not locate rack servers in open-plan offices. Tower servers are quieter and more office-friendly.
• ›Dust and humidity: Servers require low-dust, low-humidity environments. Garages and storerooms are not suitable. Dedicated IT rooms or sealed server cabinets with filtered airflow are the minimum standard.

9. Operating System & Software Compatibility

• ›Hardware Compatibility List (HCL): Before finalising your specification, verify that your chosen server model appears on the HCL for your intended OS and any critical applications (VMware, Microsoft SQL Server, SAP, etc.).
• ›Windows Server licensing: Microsoft licenses Windows Server per physical core. A 16-core server requires 16 core licences at minimum. Factor this in — it can equal or exceed the hardware cost itself.
• ›VMware vSphere licensing: Also core-based. If virtualisation is your goal, budget for hypervisor licensing alongside the hardware.
• ›Linux is free: If your workload supports it, Ubuntu Server, RHEL, or Rocky Linux eliminate OS licensing costs entirely. Many enterprise applications run well on Linux.

10. Warranty & Support

• ›Standard warranty on new servers is typically 1–3 years, extendable to 5 years via support contracts.
• ›On-site vs. return-to-depot: For production servers, on-site next-business-day parts replacement is the minimum. Return-to-depot means days without a server.
• ›Firmware and driver support: New servers from major brands receive firmware and security patches throughout the warranty period — a meaningful security advantage over used or end-of-life hardware.

11. Scalability & Future-Proofing

• ›Buy a server with empty RAM slots, empty drive bays, and an empty CPU socket — even if you don't populate them now. The ability to double RAM or add drives without replacing the whole unit is worth paying for at purchase.
• ›Hot-swap drive bays allow you to replace a failed drive without taking the server offline. All production servers should have hot-swap storage as a minimum.
• ›Virtualisation is generally the most flexible approach for growth — add resources to VMs as needed without procuring new hardware immediately.
• ›Plan for a five- to seven-year hardware lifecycle. Servers that are too small on day one will be replaced far sooner — a false economy that costs more in total than buying correctly upfront.

12. Total Cost of Ownership (TCO)

The purchase price is only part of what a server costs. Over five years, factor in:

13. New vs. Used — Why New Wins for Production

• ›Unknown history: A used server may have run at 100% load for three years before you buy it. You inherit the wear, not just the hardware.
• ›No manufacturer warranty: Used servers typically have no warranty. A failed drive, dead PSU, or faulty RAID controller means out-of-pocket repair costs and unplanned downtime.
• ›End-of-life risk: Used servers are often discontinued models. Firmware security patches stop, replacement parts become scarce, and you may face forced replacement sooner than expected.
• ›Energy inefficiency: Older generation servers draw significantly more power for the same workload. The energy cost difference over five years often negates the lower purchase price entirely.

Summary Checklist