What a Dedicated Server Actually Is: Specs, Workloads and How to Choose

A dedicated server is a single physical machine serving one customer, which changes three things versus VPS or shared hosting: the CPU is yours (no steal time), the disks are local (predictable IOPS), and you get hardware-level management (IPMI/IP-KVM, BIOS access, RAID control). This page covers what those three advantages mean in practice, which specs matter for gaming versus high-traffic web workloads, and what to verify before you commit to a monthly contract.

The sections below follow the order in which most buyers get it wrong: they pick the provider first, then discover the hardware cannot do the job. Start with workload, then map to specs, then choose the facility.

What “Dedicated” Actually Buys You

On a VPS, the hypervisor allocates your vCPUs against an oversubscribed host. Most of the time this is fine; when a neighbour runs a compile job or a crypto workload, your processes are preempted and latency spikes. Steal time of 5% is noticeable in a game server tick loop and invisible in a WordPress site. On bare metal, no other workload shares your cores.

Storage is the second difference. Cloud block storage typically offers burst IOPS with a much lower sustained baseline, and shared NVMe arrays are rate-limited per tenant. Local NVMe in a dedicated chassis delivers 200,000+ random read IOPS sustained, which is what makes database-heavy or chunk-based game workloads behave predictably.

The third is control. IPMI or IP-KVM lets you reinstall an OS, adjust BIOS power settings, and recover a machine that will not boot — without opening a ticket. For game servers, disabling C-states and forcing performance governor in BIOS is worth several milliseconds of tick jitter.

Spec Axes: Where to Spend and Where to Save

ResourceGame server priorityWeb/app priorityNotes
Single-thread clockCriticalModerate3.7 GHz+ boost for tick-based games
Core countModerate (8–16)High for PHP workersMatch to concurrency, not prestige
RAMHigh (world state)High (cache)ECC preferred, cheap to over-provision
StorageNVMe, RAID1NVMe, RAID10Never SATA SSD for new builds
Network1 Gbps, low jitter1–10 Gbps, high throughputUnmetered vs commit matters
DDoSAlways-on, inlineOn-demand acceptableAsk for scrubbing capacity per IP

Two rules of thumb worth keeping: do not pay for 64 cores to run a 32-slot game server, and do not buy a 4-core box to run a busy WooCommerce store. In both cases the money is in the wrong resource.

Matching Workload to Hardware Class

Game servers are latency-bound: they care about single-thread performance, memory access latency, and network jitter. A 3.8 GHz 8-core EPYC or Xeon with 64 GB RAM will outperform a 2.4 GHz 32-core machine every time for Minecraft, Palworld, Rust, or FiveM, because the simulation loop is effectively single-threaded per world.

High-traffic web workloads are throughput-bound: PHP-FPM workers, database connections, and TLS handshakes all scale with cores and with cache. A busy Magento or Laravel application with a local MySQL instance wants cores, RAM for buffer pool, and fast sequential writes. The two profiles rarely overlap, which is why “gaming” and “enterprise” dedicated plans diverge in the market.

High-traffic and batch workloads sit between: video transcoding wants cores and fast scratch disks, analytics wants RAM and NVMe, and anything doing sustained writes needs to know its RAID level and write-amplification characteristics before it signs a twelve-month term.

Facility and Network Choices

Location is chosen by the user, not by the provider’s marketing map. Put the server within 20 ms of the majority of your users; beyond that, TCP and UDP retransmit behaviour degrades faster than most people assume. For a European game community, Frankfurt, Amsterdam, and London cover almost everyone within 30 ms; for US-wide audiences, one East and one Central node beat a single West Coast machine.

Network terms matter as much as location. Ask whether bandwidth is unmetered or metered, whether the port is a true 1 Gbps or a shared uplink, and what the commit is on a 10 Gbps offering. Metered plans with a low per-TB overage can cost more than a slightly pricier unmetered plan during a launch spike.

Pre-Signature Verification

  • Request a test IP in the exact facility and run mtr from your users’ networks
  • Get the CPU model, RAM type and speed, and disk model in the quote
  • Confirm IPMI/IP-KVM and whether it is included or an add-on
  • Ask how many IPv4 addresses you get and whether reverse DNS is self-service
  • Confirm reinstall and rescue procedures and their turnaround time
  • Read the acceptable-use and DDoS policy: some hosts null-route on first attack
  • Check whether hardware replacement is guaranteed in hours and what SLA credit applies

A one-month initial term is worth the higher monthly rate. It gives you a full billing cycle to run acceptance tests and load the machine with real traffic before committing to annual pricing.

Cost Model: What You Are Actually Paying For

Roughly 40–55% of a dedicated server’s price is hardware depreciation, 20–30% is power and cooling, 10–20% is space and cross-connect, and the remainder is support, network, and margin. This is why the same CPU costs 30–50% more in a high-power-cost metro and why older hardware is discounted so aggressively — the operator has already recovered the capex.

The practical implication: if a quote is dramatically below market for the stated specs, one of those components is being cut. Usually it is power (older, inefficient CPUs), network (shared uplink), or support (no remote hands). Ask which.

Next Step

The fastest way to a good decision is to write down your two most demanding workloads, the number of concurrent users or slots, and your latency target — then compare them against published specs rather than sales copy. For a starting point, review the dedicated server plans and specs overview, and keep the acceptance-test checklist from the 30-minute node verification guide ready for the day the machine is handed over.

Leave a Reply