How to Choose the Right Dedicated Server Configuration
By The Atlas Hoster team4 min read
Configuring a dedicated server means choosing a processor, memory, disks, and network that match a specific workload. Too small, and the server saturates; too large, and you pay for unused capacity. This guide reviews the four components to size, in the order that matters.
Configure a dedicated server: the four levers
A configuration comes down to four choices, made in this order: the processor, memory, disks, and network. Each addresses part of the workload, and all four must be consistent with each other. It is useless to have a powerful processor if the disk bottlenecks it, or lots of memory if the network cannot keep up.
Define the workload: base, virtualization, computation
Before choosing hardware, describe what the server will do. Workloads fall into three categories, with opposing needs:
- Database: memory and fast disk intensive, sensitive to data integrity.
- Virtualization: requires many cores and memory to host multiple virtual machines. Computation: processor-bound, less demanding on the rest.
A database and a computation server are not configured the same way. Write down the workload first, then choose each component accordingly. To determine if a dedicated server is truly necessary, read Dedicated server: when you really need one.
If you are still hesitating between a dedicated server and a VPS, the comparison dedicated server or VPS helps you decide.
Processor: cores and frequency
The processor is judged on two axes: the number of cores and the frequency.
- A large number of cores serves parallel workloads: virtualization, high-traffic web server, file processing.
- High frequency serves sequential workloads: a database executing long queries, an application waiting for a fast response.
The reference server processor families are Intel Xeon and AMD EPYC, available in ranges according to the number of cores and frequency. The classic mistake is to over-dimension cores “for later”: unused cores are lost money. Start by measuring your current workload (number of requests, average load) rather than guessing.
ECC memory: when it is essential
ECC (Error-Correcting Code) memory detects and corrects single-bit errors, and detects double-bit errors. On standard memory, an isolated error may go unnoticed and silently corrupt data.
For a database or file system hosting critical data, ECC is highly recommended: silent corruption of a client record or transaction is difficult to detect and repair. For a simple reconfigurable web server, the stakes are lower, but quality dedicated servers are almost always equipped with ECC by default.
Disks: NVMe and RAID
The disk is often the real bottleneck. Two choices:
- NVMe: fast SSDs, directly connected, much more performant than SATA for the random access of a database.
- RAID: combining multiple disks for performance or fault tolerance.
Common RAID levels:
| RAID | Disks | Effect |
|---|---|---|
| RAID 1 | 2 | Mirror: each piece of data is duplicated, one disk failure causes no loss |
| RAID 10 | 4+ | Mirror + aggregation: performance and redundancy |
| RAID 5 | 3+ | Aggregation with parity: survives one failed disk |
| RAID 6 | 4+ | Dual parity: survives two failed disks |
For a database, RAID 10 on NVMe offers both speed and security. Remember that RAID is not a backup: it protects against disk failure, not accidental deletion or compromise.
Network: bandwidth and IP addresses
The network connects your server to your visitors. Check:
- Link bandwidth: 1 Gbit/s is common and sufficient for most uses; 10 Gbit/s is justified for large volumes.
- Number of IP addresses: one main IPv4 address, and additional IPs if you host multiple services or certificates.
- Location: a server close to your visitors reduces latency, detailed in Latency and hosting.
Installed and hardened system at delivery
Configuration does not end with hardware. Check what the provider delivers with the machine:
- an installed and updated operating system;
- initial hardening (restricted SSH access, basic firewall);
- documented access, with a machine name and specific credentials.
A “bare” machine imposes all this initial work. A delivered machine that is installed and supervised saves you hours, even days. The machine name and access are detailed in What is my hostname and How to connect to the server via SSH.
Once the machine is delivered, it remains to be administered daily: this is the subject of managing a dedicated server.
To summarize
A good dedicated server configuration starts from the actual workload: cores for parallelization, ECC for data integrity, NVMe in RAID for speed and resilience, and a network suited to the target audience. Measure first, choose second, and have the machine delivered installed and hardened.
For a dedicated server configured to measure, delivered installed and supervised, consult Atlas Hoster dedicated server offers.
Frequently Asked Questions
How many cores do I need for a database? It depends on the workload, not a fixed rule. Measure the average load and simultaneous requests: a lightly used database gets by with a few cores, a heavy database may require dozens.
Does RAID replace backups? No. RAID protects against disk failure, not accidental deletion, software error, or compromise. Backups remain essential.
Is ECC memory really useful? Yes for critical data: it prevents silent corruption that standard memory allows to pass. On a server hosting a database, it is highly recommended.