Renting a dedicated server is a different kind of purchase from renting a VPS. There is no share of a core to reason about and no overcommit to discover at 21:00. You are buying one physical chassis, with a fixed processor model, a fixed memory population and a fixed set of drives, in a specific building in a specific country — and you are committing to it for longer than a month.
That makes the selection process concrete. This guide is about choosing an offshore dedicated server from the things you can verify before ordering: the silicon, the memory, the storage layout, the network, the out-of-band access and the site. It deliberately does not repeat the general comparison between virtual and physical machines, which is covered in VPS versus dedicated server.
What a dedicated server commits you to
A dedicated server is the whole machine. Nothing is shared with another tenant: the sockets, the memory modules, the drives and the network card belong to your operating system. In exchange, three things become true at once:
- The specification is fixed. You cannot add a vCPU next month. You can usually add drives or memory if the chassis has room, and that is a conversation with the operator, not a dropdown.
- The failure domain is yours. A dead drive, a bad memory module or a failed fan is a hardware event, not a support abstraction. Redundancy is something you design and monitor.
- The delivery is manual. Every server here is provisioned after payment review. Nothing is imaged by an unattended system the moment a transaction confirms, and a dedicated chassis may be built to order.
None of that is a disadvantage if the workload needs the machine. It is a disadvantage if what you actually needed was a bigger VPS.
Read the processor model first
Core counts are the least useful column in a dedicated server listing on their own. The model determines the instruction set, the memory generation, the clock behaviour and the age of the platform.
| Offer | Processor | Cores / threads | Clock | Memory | Storage | Network | Monthly |
|---|---|---|---|---|---|---|---|
| Bucharest E-2288G | Intel Xeon E-2288G | 8c / 16t | 3.7–5.0 GHz | 64 GB ECC | 2 × 1 TB NVMe · RAID 1 | 20 TB · 1 Gbps | 89 USD |
| Amsterdam E-2388G | Intel Xeon E-2388G | 8c / 16t | 3.2–5.1 GHz | 64 GB ECC | 2 × 960 GB NVMe · RAID 1 | 20 TB · 1 Gbps | 99 USD |
| Amsterdam Ryzen 7950X | AMD Ryzen 9 7950X | 16c / 32t | 4.5–5.7 GHz | 128 GB ECC | 2 × 1.92 TB NVMe · RAID 1 | 20 TB · 1 Gbps | 169 USD |
| Miami 2× Silver 4314 | 2 × Intel Xeon Silver 4314 | 32c / 64t | 2.4–3.4 GHz | 256 GB ECC | 2 × 3.84 TB NVMe · RAID 1 | 30 TB · 10 Gbps · 2 IPv4 | 299 USD |
| Zurich EPYC 9354P | AMD EPYC 9354P | 32c / 64t | 3.25–3.8 GHz | 256 GB ECC | 2 × 3.84 TB NVMe · RAID 1 | 30 TB · 10 Gbps | 349 USD |
| Singapore EPYC 9354P | AMD EPYC 9354P | 32c / 64t | 3.25–3.8 GHz | 256 GB ECC | 2 × 3.84 TB NVMe · RAID 1 | 30 TB · 10 Gbps | 399 USD |
The pattern worth noticing is that the two cheapest machines are the highest-clocked ones. An E-2288G at 5.0 GHz will beat a 32-core EPYC on a single-threaded PHP request, a Node process or a lightly loaded database, because those workloads are limited by one fast thread rather than by parallel capacity. The EPYC earns its price when the workload actually runs in parallel: many workers, many containers, many concurrent database sessions.
The Miami machine is the other extreme: two sockets, 32 cores, but a 2.4 GHz base clock. It is a consolidation platform — a large memory pool and many threads for virtualisation — not a low-latency single-thread machine.
| Workload shape | Better starting point | Reason |
|---|---|---|
| One busy web application, moderate database | E-2288G or E-2388G | High clock, 8 fast cores, 64 GB is plenty |
| Many workers, queues, build jobs | Ryzen 7950X | 16 modern cores at 4.5–5.7 GHz |
| Consolidation, many VMs, large datasets | EPYC 9354P or 2× Silver 4314 | 32 cores and 256 GB of ECC memory |
| Memory-bound analytics | EPYC 9354P | 256 GB with room for page cache |
| Lowest entry cost with EU jurisdiction | Bucharest E-2288G | 89 USD, EU member state |
Memory: ECC and the capacity you will actually use
Every offer here ships ECC memory, which matters more on a dedicated server than on a laptop: a single flipped bit in a long-lived process is an outage nobody can explain, and ECC turns it into a corrected event in the log.
Capacity is a sizing question, not a badge. 64 GB is enough for a LAMP stack, a Node application and a database with a working set that fits in memory. 128 GB is the comfortable middle for a busy database plus application plus cache. 256 GB is for consolidation: several virtual machines, a large page cache or an analytics workload that scans more data than it fits on disk.
The mistake is buying 256 GB because it is the top of the list and then running a workload that never exceeds 12 GB. The other mistake is buying 64 GB for a database whose working set is 90 GB and discovering that every query goes to disk.
Storage: RAID 1 is availability, not backup
All six offers use two NVMe drives in hardware RAID 1. That protects against a single drive failure: the machine keeps running on the surviving member, and the array can be rebuilt after a replacement. It does not protect against deletion, ransomware, a bad migration or a corrupted filesystem, because every write — including the destructive one — is mirrored.
Three operational rules follow:
- Back up off the machine. A RAID 1 array is one chassis. A backup on the same chassis is not a backup.
- Watch the drives. SMART attributes, NVMe wear indicators and the RAID controller's event log are the difference between a planned replacement and a 03:00 rebuild.
- Plan the replacement. A drive swap on dedicated hardware involves the operator, a maintenance window and a rebuild. Budget for it in the same way you budget for the monthly fee.
The offer with 1 TB drives is not worse than the one with 3.84 TB drives for a workload that stores 200 GB. It is worse if you intended to keep years of logs and snapshots on the same pair.
Network: port, traffic and addresses
Two numbers decide the network line on a dedicated offer: the port speed and the included traffic. A 1 Gbps port with 20 TB suits a normal web estate. A 10 Gbps port with 30 TB suits a workload that moves large files, serves media or replicates a dataset.
The Amsterdam machines offer a 10 Gbps upgrade with 50 TB for +40 USD per month, described in the catalogue as a shared uplink, which is an honest label: the port is 10 Gbps, and the upstream is shared with other tenants. Zurich, Miami and Singapore include 10 Gbps and 30 TB as standard. Bucharest is 1 Gbps only.
IPv4 addresses differ too: Miami includes two, the others one. If you need a second address for mail separation, a VPN endpoint or a staging interface, that is a reason to look at Miami beyond its memory.
DDoS protection is not advertised here, and this article will not invent a number for it. If mitigation matters to your workload, ask the operator what is in front of the machine before you order, and treat an unverified claim as no claim.
Out-of-band access is a line item worth checking
Every offer includes out-of-band access: IPMI 2.0 on the Amsterdam, Zurich, Miami and Singapore machines, KVM over IP in Bucharest. This is the ability to see the console, power-cycle the machine, mount an ISO and reinstall the operating system without opening a ticket.
It is the single most useful feature on a dedicated server, and it is also a security surface. A baseboard management controller exposed to the internet with a default password is a well-known route to full compromise. Practical rules:
- Keep the BMC on a management network or behind a VPN; do not expose it publicly.
- Change the default credentials before the machine does anything else.
- Disable unused services on the controller and keep its firmware current.
- Treat BMC access as equivalent to physical access, because it is.
If a provider cannot tell you how the management interface is reached, that is an answer in itself.
Location changes price, latency and law
The same 32-core EPYC costs 349 USD in Zurich and 399 USD in Singapore, and the difference is geography rather than hardware. At the entry end, Bucharest is 89 USD and Amsterdam is 99 USD for effectively the same class of machine, with the EU applying in both and the latency profile differing by region.
Three things move with the site:
- Latency. Distance sets a floor. A server in Singapore is roughly 160–180 ms from Europe regardless of network quality. The location guide works through the arithmetic.
- Jurisdiction. Amsterdam and Bucharest sit in the EU under GDPR; Zurich under Swiss FADP; Singapore under the PDPA; Miami under US state law. A dedicated chassis is a physical object in that country, which makes the legal question more concrete than it is for a VPS.
- Capacity. Singapore is currently sold build-to-order and Miami is marked limited, which affects lead time rather than price.
Delivery, terms and the real monthly cost
Dedicated servers are provisioned manually after payment review. There is no automated handover, and a build-to-order machine is exactly what the label says. Plan a migration window around a human process rather than around a countdown.
The term ladder is the same as everywhere else in the catalogue: one month at list price, six months at 10% off the monthly rate, twelve months at 20% off. A twelve-month Bucharest E-2288G is 71.20 USD per month; a twelve-month Zurich EPYC is 279.20 USD per month. The discount is real, and so is the commitment.
Two lines are easy to forget when comparing the monthly figure:
- Windows Server 2022 adds 25 USD per month if the workload needs it. Linux is included.
- Your operating costs — backup storage, monitoring, a second machine for redundancy, and the time to patch and watch the hardware — are not in the catalogue price and are usually larger than the difference between two offers.
A checklist before you order
- Match the processor shape to the workload: clock first for single-threaded applications, cores and memory first for consolidation.
- Confirm the memory capacity against the working set, not the peak ever recorded.
- Decide how much local NVMe you need and whether 1 TB or 3.84 TB drives are the right size for the retention you plan.
- Check the port speed and the traffic allowance against real egress, and note whether the uplink is shared.
- Count the IPv4 addresses you need before choosing the site.
- Confirm the out-of-band access and how it is reached, then secure it on day one.
- Read the jurisdiction note and decide whether that country is where the data should live.
- Add the Windows licence, backup storage and your own time to the monthly figure before comparing offers.
- Plan for manual delivery, and treat "build to order" as a lead time rather than a stock status.
The catalogue lists every offer with its full hardware specification on the dedicated servers page. If the entry machine is the right shape but the site is wrong, the Bucharest E-2288G and the Zurich EPYC 9354P show how far the same decision can move on price and jurisdiction. Whichever you pick, the first hour after delivery should follow the hardening baseline — key-only SSH, a default-deny firewall, updates and a backup you have restored — before anything production touches the machine.