Most dedicated server disputes come down to a handful of fields in the contract that were never written down. The sales page says 32 cores; the server has 16 physical cores and 32 threads. The listing says 1 Gbps; the contract says “up to” 1 Gbps with an 11 TB monthly cap. Before you sign, verify these eight specification fields in writing and match them against the provider’s actual hardware handout. You can use a comparison of dedicated server hosting providers to see which vendors publish this detail up front.
1. CPU: model, core count, and thread count
Ask for the exact model string, for example “2x Intel Xeon Gold 6338” or “1x AMD EPYC 9354P”. A model string lets you look up core count, base and turbo clock, and memory channels yourself. Reject contracts that only state “32 cores” or “high-frequency Xeon”, because those can map to a 2014 E5-2660 v3 or a current EPYC 9124 with very different performance.
Also confirm whether the contract counts physical cores or threads. A provider advertising “64 threads” for an EPYC 7763 is selling 64 physical cores, but a provider advertising “64 threads” for two E5-2680 v3 chips is selling 48 physical cores. The difference matters for per-core licences in cPanel, Plesk, SQL Server, and many Java or .NET workloads.
2. RAM: usable capacity and module population
The contract must state usable RAM after any hypervisor or IPMI reservation, which is usually 200 to 600 MB less than the installed total. It must also state how the RAM is populated. 128 GB as 8 x 16 GB ECC on an eight-channel EPYC board is a very different machine from 128 GB as 4 x 32 GB on the same board, because the latter halves memory bandwidth.
Verify the speed too. A DDR4 platform rated for 2,933 MT/s that is populated with 2,400 MT/s DIMMs runs at 2,400. Most providers do not list this, so ask for the output of dmidecode -t memory after delivery, or require the RAM speed in the specification table.
3. Storage: model, interface, and RAID level
NVMe is not a spec, it is a class. A contract that says “2 x 512 GB NVMe” can mean consumer-grade drives with 0.3 DWPD or enterprise drives with 1 to 3 DWPD. Ask for the drive model and endurance rating. For write-heavy workloads such as database replicas, check the TBW figure and confirm whether the provider replaces failed drives on a 4-hour or 24-hour SLA.
Confirm the RAID level and whether the controller is hardware or software. Software RAID 1 with mdadm is fine for most cases, but it consumes CPU on rebuild and is subject to the provider’s monitoring. Hardware RAID with a battery-backed cache changes write performance and failure behaviour. If you expect to add disks, also confirm spare bays and whether they are already cabled and powered.
4. Network: port speed, commit, and transfer allowance
“1 Gbps port” means the link negotiates at 1 Gbps, not that you can push 1 Gbps continuously. Ask for the guaranteed commit, typically 100 Mbps to 1 Gbps, the burst allowance, and the monthly transfer cap. Many European providers use unmetered within the EU and 1 to 3 TB per month to North America. Confirm the overage rate and whether it is charged per GB or throttled.
Also confirm the IP allocation, the IPv6 prefix size, and whether addon IPs are free or billed. A /29 IPv4 block is common on dedicated servers and costs nothing at good providers; a /28 may carry a monthly fee. For IPv6, insist on at least a /64 and preferably a /48 if you plan VLANs or sub-allocations.
Also confirm whether the monthly figure includes the hardware RAID battery unit, the second NIC port, and the IPv6 prefix, or whether each is a separate line item. The line items that appear only on the invoice are the ones that break budgets.
5. Power: included watts and amp limits
Most dedicated contracts include a fixed power budget, often 300 to 500W per server or per rack unit. Above that, the provider bills per kWh or requires an upgrade. Confirm the included wattage, the per-kWh overage rate, and the amperage of the circuit, because a dual-socket 400W server plus two switches can exceed a standard 208V 10A feed.
Check whether your provider meters power at the PDU or estimates it. PDU-level metering is more accurate but sometimes billed at a higher rate. If you plan to add GPUs, this is the field that decides whether the contract can be extended at all.
6. Remote access: IPMI, KVM, and rescue mode
Confirm that out-of-band access, usually IPMI or iDRAC/iLO with HTML5 KVM, is included and not an addon. Ask about the number of concurrent sessions, whether you can mount an ISO from a remote URL, and whether the KVM is available during network outages. Some providers route IPMI on a separate management VLAN, which is safer.
Rescue mode matters more than most buyers think. Confirm that a rescue environment is available without a ticket, that it supports at least one current Debian and Ubuntu image, and that it can mount the disks for data recovery. A rescue system that requires a 24-hour ticket response is functionally unavailable.
7. SLA: uptime, credit formula, and escalation
Read the SLA text, not the badge. A 99.9 percent uptime commitment allows 43 minutes of downtime per month. Note the measurement window (calendar month or rolling 30 days), what counts as downtime (network outage versus scheduled maintenance), and the credit formula. Credits of 5 to 20 percent of the monthly fee are typical; a full refund is rare.
Also confirm the ticket response times and escalation path. The numbers that matter are first response for hardware faults, replacement time for disk or RAM failures, and the network operations contact for outage updates. If the provider cannot name a hardware replacement window, the SLA is marketing.
8. Delivery, cancellation, and data handling
Commit to a delivery window in writing. “Within 24 hours” is standard, but a 30-minute build time is a real differentiator for urgent migrations. Confirm the billing cycle start, whether partial months are prorated, and the notice period for cancellation. Most providers require 30 days, and some auto-renew for 12 months unless you give notice.
Finally, agree on what happens to your data at termination. Manual reinstall, disk retention, and secure wipe are the three options. A provider that keeps your disks for 7 to 30 days after cancellation gives you a recovery window; one that wipes immediately does not. Put this in the contract, then verify each of these eight fields on the first day using dmidecode, ip addr, and a quick disk and network benchmark. A short comparison of dedicated server contracts and specifications is a useful checklist alongside this list.
Leave a Reply
You must be logged in to post a comment.