Dedicated server listings look similar at first glance: a processor model, a memory figure, some drives and a bandwidth number. The details behind those lines decide whether the machine suits your workload or becomes an expensive compromise.
This dedicated server buying checklist covers the specifications and service terms worth comparing before you sign a contract.
Start with the workload
Hardware choices only make sense against a defined job. Before reading any specification sheet, write down:
- What the server will run: databases, web applications, virtualization, file storage, AI workloads
- Current resource usage, if you are migrating from an existing system
- Expected growth over the contract term
- How much downtime the business can tolerate
- Who will administer the machine
CPU: cores versus clock speed
Processor listings show a core count and a clock speed. Which one matters more depends on the software.
When core count matters
Workloads that do many things at once benefit from more cores. Examples include virtualization hosts running several virtual machines, web servers handling many simultaneous requests, containers, video encoding and build pipelines.
When clock speed matters
Some software does most of its work on one thread at a time. For these workloads, per-core performance decides how fast each task completes. Many game servers, some legacy business applications and individual database queries fall into this group.
Other CPU points to check
- Generation. A newer processor often outperforms an older one with the same core count and clock speed, and uses less power doing it. Compare model numbers, not just headline figures.
- Sockets. Dual-processor servers offer more cores and memory capacity. Some software is licensed per socket or per core, which can change the total cost significantly.
RAM
Memory shortages cause the most visible performance problems. When a server runs out of RAM, it starts swapping to disk and slows down sharply.
- Capacity. Databases perform best when their working data fits in memory. Virtualization hosts need the sum of all guest allocations plus overhead.
- ECC memory. Error-correcting code memory detects and corrects single-bit errors. It is standard on server-class hardware and important for data integrity. Confirm it is included.
- Expandability. Ask how much memory the server can hold for later upgrades.
Storage: NVMe versus SATA
Storage type affects application responsiveness more than most buyers expect.
| Type | Characteristics | Best suited to |
|---|---|---|
| SATA HDD | Spinning disks; lowest cost per gigabyte; slowest, especially for random access | Backups, archives, large media libraries |
| SATA SSD | Flash storage on the SATA interface; far faster than HDD | General-purpose workloads, boot drives |
| NVMe SSD | Flash storage connected over PCIe; much higher throughput and lower latency than SATA | Databases, virtualization, high-traffic applications |
The SATA interface was designed for spinning disks, and it limits how fast even a good SSD can go. NVMe was designed for flash and communicates over PCIe, which removes that bottleneck and handles many parallel operations.
RAID
RAID combines several drives so the server can survive a drive failure, run faster, or both.
- RAID 0: stripes data across drives for speed. No redundancy. One failed drive loses everything.
- RAID 1: mirrors two drives. Survives one failure. Usable capacity is that of a single drive.
- RAID 5: stripes with parity across three or more drives. Survives one failure.
- RAID 6: like RAID 5 with double parity across four or more drives. Survives two failures.
- RAID 10: mirrored pairs striped together, using four or more drives. Good performance and redundancy, at the cost of half the raw capacity.
RAID 1 and RAID 10 are common choices for databases and operating system volumes. Rebuilding a large parity array after a failure takes time and stresses the remaining drives.
One point is not negotiable: RAID is not a backup. It protects against hardware failure. It does not protect against deleted files, corrupted data, ransomware or a mistaken command, all of which are faithfully copied to every drive.
Bandwidth and port speed
These two terms are often confused.
- Port speed is the maximum rate at which the server can send and receive data at any moment, for example 1 Gbps or 10 Gbps.
- Bandwidth or transfer allowance is how much data you may move during the billing period.
Plans are usually either metered, with a fixed monthly transfer allowance, or unmetered, where you may use the port continuously. Read the terms on unmetered plans carefully. Check whether the port is dedicated to you or shared, and whether a fair use policy applies.
Also ask:
- What overage charges apply if you exceed the allowance
- How many IPv4 addresses are included and whether IPv6 is available
IPMI and remote management
Out-of-band management lets you control a server independently of its operating system, through a dedicated management controller on the motherboard. IPMI is the common standard, and hardware vendors offer their own versions, such as Dell iDRAC and HPE iLO.
With it you can:
- Power the server on, off or cycle it remotely
- Open a remote console to see the screen even when the operating system will not boot
- Mount an installation image to reinstall the system
- Read hardware sensors such as temperature and fan status
Management interfaces are a security risk if exposed. They should be reachable only through a VPN or a restricted private network, never directly from the public internet. Confirm how the provider secures access.
DDoS protection
A distributed denial-of-service attack floods a server with traffic to knock it offline. Any public-facing server can be a target.
Questions for the provider:
- Is protection included or an add-on?
- Is it always on, or activated after an attack is detected?
- Does it cover only network-layer floods, or also application-layer attacks against your website?
- What happens during an attack larger than the protection covers? Some providers null-route the targeted IP address, which takes your service offline.
Managed versus unmanaged
This choice affects your total cost and risk more than any hardware line item.
Unmanaged means the provider keeps the hardware, power and network running. Everything from the operating system up, including security patches, configuration, monitoring and troubleshooting, is your responsibility.
Managed means the provider takes on some or all of those tasks. The scope varies widely between companies, so ask for it in writing. Clarify who handles updates, monitoring, security hardening, backups, control panel support and incident response, and at what hours.
An unmanaged server is economical only if you have the skills and time to run it. If you lack in-house administrators, see what server management covers.
Backups
Decide the backup plan before the server goes live, not after the first incident.
- Location. Backups must be stored off the server, ideally in a different facility.
- The 3-2-1 rule. Keep three copies of your data, on two different types of storage, with one copy off-site.
- Frequency and retention. Match these to how much data you can afford to lose and how far back you may need to go.
- Restore testing. A backup that has never been restored is an assumption. Test recovery on a schedule.
- Protection from tampering. Backups reachable with the same credentials as the server can be encrypted or deleted in the same attack.
Key takeaways
- Define the workload and measure current usage before comparing hardware.
- Choose more cores for parallel workloads and higher per-core speed for single-threaded ones. Check the processor generation.
- Insist on ECC memory and room to expand.
- NVMe suits databases and busy applications. HDDs remain practical for bulk storage and backups.
- Use RAID for hardware redundancy, and never treat it as a backup.
- Distinguish port speed from transfer allowance, and read the terms on unmetered plans.
- Confirm secured IPMI access, the scope of DDoS protection and exactly what “managed” includes.
- Keep tested, off-server backups following the 3-2-1 rule.
iSky Creative supplies dedicated servers with optional management, and you can contact us to match a configuration to your workload.