Dedicated servers mean different things to different providers. Some sell you a slice of a virtualization cluster and call it dedicated. Others sell you a single machine but rent the hardware from a wholesale upstream, and your "dedicated" box ends up sitting in the same rack as customers from completely unrelated networks. We do neither. Every machine on this page is physical hardware we bought, racked in datacenters we contracted with directly, and operate without any third-party virtualization layer between you and the metal.
What you get: predictable per-core performance because no hypervisor schedules around you, NVMe IOPS that match the spec sheet because no neighbour competes for the disk subsystem, kernel and BIOS access through IPMI for operations that need it, and rDNS aligned to your sending domain because we own the IP allocation. Pay in cryptocurrency, no identity verification, full cryptographic isolation between your operation and ours.
Three workload categories produce materially different outcomes on dedicated bare-metal versus virtualized alternatives. Understanding which category your work falls into determines whether dedicated infrastructure is worth the price difference, or whether VPS is the more economical path.
I/O-intensive workloads. Database servers, mail servers handling 50K+ daily messages, log aggregation, and storage backends all generate sustained random I/O patterns that hypervisors handle inconsistently. NVMe drives rated for 700K random read IOPS deliver 200-400K under virtualization because the hypervisor's I/O scheduler queues requests across tenants. The same drives in a single-tenant configuration deliver close to spec, often within 5-10% of the manufacturer's published numbers. For a PowerMTA queue processing transactional mail at 200 messages per second, the difference between virtualized and dedicated I/O translates directly into queue depth and delivery latency.
CPU-intensive sustained workloads. The virtualization tax on CPU varies dramatically by hypervisor generation, host overcommit ratio, and the specific vCPU scheduling algorithm. A vCPU advertised as equivalent to one physical core typically delivers 60-85% of physical throughput depending on tenant load. For batch processing, rendering, scientific computation, or any workload where the CPU is the bottleneck, the virtualization tax is real money. Dedicated bare-metal eliminates the variability; your 32-core EPYC delivers 32 cores worth of work at the frequency the silicon was designed for.
Compliance and isolation requirements. Regulated workloads (financial services, healthcare, legally exposed content), audit-traceable operations, and single-tenant data residency requirements often disqualify shared infrastructure outright. The compliance argument is not about whether the hypervisor leaks data between tenants in practice; it is about what an auditor or court will accept. A single-tenant physical machine with full chain-of-custody from manufacturer to deployment is straightforward to attest. A vCPU sliced from a shared host requires explaining the hypervisor's isolation guarantees, the host operator's own controls, and the residual risk surface. For some operators that is fine; for others, dedicated is the only path that closes the attestation cleanly.
What dedicated does not buy you: better network performance to internet-distant endpoints (network latency to Gmail is the same for VPS and dedicated when they sit in the same rack), better support response times (response is a function of provider operations, not of hardware tier), or magical reputation lift (sender reputation is independent of whether your IP sits on bare-metal or virtualized infrastructure). Dedicated buys predictable per-machine performance and single-tenant isolation. That is the entire value proposition.
Tell us what the server will run and which jurisdiction fits your operation. The configurator returns the matching baseline, expected price range, and the configuration knobs most operations adjust from baseline.
Bulk email at moderate sustained load benefits from the E5's per-core throughput plus 64 GB RAM for PowerMTA queues and MailWizz database. NVMe RAID-1 protects against drive failure. Bulgaria balances EU jurisdiction, Tier 1 transit, and competitive pricing.
Configurator output is a starting recommendation, not a final order. Confirm exact stock, current promo pricing, and any custom requirements (specific NIC, GPU, dual CPU, large RAM) on Telegram before placing the order. We maintain reserves of the most-ordered configurations; unusual specs may take 24-72 hours to stage.
Three reference configurations cover the majority of email infrastructure and general-purpose workloads. Each is customisable from the baseline shown.
All three baselines are starting points. Need 256 GB RAM, dual EPYC sockets, RAID-10 across 8 drives, GPU pass-through, or a 25 Gbps NIC? We configure custom hardware on request. Pricing scales linearly with components; non-standard configs take 24-72 hours to stage depending on stock.
Hardware in Sofia and hardware in Singapore answer to different legal frameworks, transit different networks, and serve different audiences with different latency. The right jurisdiction depends on your threat model, your audience geography, and the regulatory exposure of your operation.
| Location | Datacenter tier | Transit carriers | Median latency to Gmail | Legal framework | Best for |
|---|---|---|---|---|---|
| BG Sofia | Tier III+ | Cogent, NTT, Telia | 22 ms | EU member, GDPR, low hosting cost | Default EU choice for cost-effective stable jurisdiction |
| RO Bucharest | Tier III+ | Cogent, NTT, Telia | 26 ms | EU member, GDPR, favorable hosting law | EU operations needing strong network capacity |
| MD Chișinău | Tier III | Cogent, NTT | 30 ms | Non-EU EEA-adjacent, no automatic EU directives | EU-adjacent without full EU regulatory exposure |
| PA Panama City | Tier III | Cogent, Telxius | 54 ms | Limited US MLAT exposure, multi-jurisdiction independence | Americas-focused operations, US legal threat model |
| HK Hong Kong | Tier III+ | NTT, Cogent, CN2 | 24 ms (China) | Separate legal system, post-2020 NSL considerations apply | East Asia operations, China-adjacent latency |
| SG Singapore | Tier IV | NTT, Telia, Cogent | 22 ms (APAC) | Stable common-law jurisdiction, strong privacy framework | Premium APAC hub, regulated operations |
| UA Kyiv | Tier III | Cogent, NTT | 30 ms | Non-EU, active conflict zone considerations apply | Cost-sensitive operations comfortable with the risk |
Latency figures are medians measured from the datacenter to Gmail's nearest edge over rolling 30-day windows. Network conditions vary; corporate Outlook tenants and smaller receivers route differently. For operations where 5-10 ms latency matters (real-time API), measure from your specific endpoints rather than relying on these medians.
Every dedicated server provisions with the same base configuration. Add-ons stack on top; nothing essential is locked behind tier upgrades.
We buy the hardware, rack it in datacenters we contract with directly, and assign it to one customer. No shared CPU, no virtualization layer, no wholesale reseller arrangement where you discover six months in that the machine is co-located with unrelated tenants.
One IPv4 from a clean /24 block at provisioning. IPv6 /64 included. Custom rDNS aligned to your sending domain set within 4 hours of provisioning. Additional IPv4 at €8/month each. /28 and larger blocks available with operational justification.
Full IPMI access for emergency console, OS reinstall, BIOS configuration, virtual media mounting. IPMI traffic firewalled to your authorised IPs at the network level for security. Credentials provided at provisioning; rotation on request.
Multiple uplinks via Cogent, NTT, Telia (Arelion), GTT, and CN2 depending on location. Direct peering with Google, Microsoft, Apple at most locations. BGP announcements support our /24 IP ranges; we do not run on customer-of-customer arrangements.
Disk SMART monitoring, RAID array health, ECC RAM event tracking, CPU thermal monitoring, NIC status, PSU health. Failures detected within 60 seconds, replacement initiated immediately. Typical replacement windows: NVMe 2-4h, RAM 4-6h, CPU 8-12h, motherboard 12-24h depending on datacenter.
L3/L4 mitigation included on all servers. Scrubbing capacity 100-500 Gbps depending on location. L7 (application-layer) filtering available as €49/month addon. For sustained large-scale attacks, dedicated scrubbing capacity is configurable on request.
AlmaLinux 9, Debian 12, Ubuntu 24.04 LTS, Rocky Linux 9, FreeBSD 14, Windows Server 2022. Default Linux images ship with our hardened CIS-aligned baseline (SSH key-only, fail2ban, firewall tuned for SMTP, automatic security updates). Custom ISO mounting via IPMI for specialised distributions.
Pay in BTC, Lightning, XMR, ETH, USDT (ERC/TRC/BEP), USDC, Solana, Litecoin, TRON, Bitcoin Cash, Dogecoin, DAI through self-hosted BTCPay. No third-party payment processors. No identity verification required for standard plans. Email address only for account management.
Tickets and Telegram conversations route to engineers, not first-tier scripts. Median response 15-30 minutes during operating hours. We read your ticket, look at your server, and respond with technical specifics. We do not gatekeep behind escalation tiers.
The price tag is not the relevant comparison. Sustained workload economics, hidden fees, performance variability, and operational overhead all factor into actual TCO. Below is a working model for the four common architecture choices applied to a representative sustained workload.
| Hardware | €169/mo |
|---|---|
| Extra IPv4 (3) | €12/mo |
| L7 DDoS | €49/mo |
| Setup (one-time) | €0 |
| Year 1 total | €2,760 |
Performance predictable at spec. No virtualization tax. Hardware monitoring + replacement SLA included. No surprise overage charges.
| VPS-4 equivalent | €119/mo |
|---|---|
| Extra IPv4 (3) | €9/mo |
| DDoS addon | €39/mo |
| Setup | €0 |
| Year 1 total | €2,004 |
Saves ~€750/year. Per-core performance variability under sustained load. NVMe IOPS 50-70% of bare-metal spec. Hypervisor scheduling jitter affects PowerMTA queue latency.
| m5.2xlarge or n2-standard-8 | ~€340/mo |
|---|---|
| Bandwidth (15 TB outbound) | ~€1,200/mo |
| EBS / persistent disk (500 GB) | ~€55/mo |
| Elastic IP | ~€4/mo |
| Year 1 total | ~€19,200 |
Bandwidth alone exceeds dedicated total. KYC required. Email-related operations frequently restricted by AUP. Migration off cloud is the most common reason operators end up with us.
| Hardware capex (amortized 36mo) | ~€95/mo |
|---|---|
| Rack + power + cooling | €180/mo |
| 1 Gbps unmetered transit | €90/mo |
| IPv4 /29 lease | €18/mo |
| Hardware capex (year 1) | €3,400 |
| Year 1 total | ~€7,756 |
Capex commitment, depreciation risk, replacement labour, on-call engineering. Makes sense for very specific hardware needs (custom NICs, GPU, very large RAM); rarely makes sense otherwise.
Year-1 totals exclude operational labour cost on the customer side, which is similar across dedicated and VPS and significantly higher for self-colocation. Cloud totals exclude reserved-instance discounts (typically 30-50% if you commit 1-3 years). The directional point holds: dedicated bare-metal in offshore jurisdictions sits in the right TCO range for most sustained workloads.
The Iron-E3 baseline ships with Intel Xeon E3-1245v6, four cores at 3.7 GHz boost. The E3 platform retains relevance for workloads where per-core throughput matters more than core count. PowerMTA at low-to-mid volume, MailWizz with modest customer counts, single-tenant transactional mail, general web application servers all fit within the E3 envelope. ECC RAM, server-grade reliability, RAS features present.
Iron-E5 ships with Intel Xeon E5-2680v4, fourteen cores at 2.4 GHz base, 3.3 GHz boost, 35 MB L3 cache. The E5-v4 platform remains the most economical option for sustained mid-volume bulk email work. Per-core throughput is lower than newer Scalable parts but core count compensates; PowerMTA queue throughput at this configuration sustains 1,500-2,500 messages per second to mixed receiver destinations.
Iron-EPYC ships with AMD EPYC 7402P, twenty-four Zen 2 cores at 2.8 GHz base, 3.35 GHz boost. EPYC 7002 (Rome) generation provides the best price-performance ratio in our current fleet. Sixty-four cores per socket available on EPYC 7702P configurations for customers needing the density. PCIe 4.0 throughout, allowing eight-drive NVMe arrays without bandwidth constraint. EPYC's chiplet architecture means workloads that fit within a single CCD (eight cores) avoid cross-CCD memory latency; pinning processes to specific cores helps workloads sensitive to this.
For workloads requiring newer silicon (Sapphire Rapids, Granite Rapids, EPYC Genoa, EPYC Turin), we configure on request. Pricing scales accordingly; 32-core EPYC 9354 configurations land around €399/month, 64-core EPYC 9554 around €549/month. The mainstream offshore market has not fully migrated to Genoa/Turin yet because the price-per-core has not crossed the threshold where it beats Rome; that crosses through 2026 and we will refresh baselines accordingly.
All baselines ship with NVMe storage. NVMe-only fleet since 2023; we deprecated SATA SSD and HDD options because the price-performance gap closed and operational consistency improved. Drives are enterprise NVMe (typically Samsung PM983, Micron 7300 PRO, Intel D7-P5520 depending on stock), rated for 1.0-1.3 DWPD endurance, PLP (power-loss protection) on all units.
Default RAID configuration is RAID-1 (mirror) for two-drive arrays. RAID-10 across four drives available as upgrade (+€90/month including extra drives). RAID-Z and ZFS configurations supported on FreeBSD and Linux with ZFS modules; we set them up on request but do not provide managed ZFS administration. Hardware RAID controllers (LSI MegaRAID 9460-16i) on EPYC configurations; software RAID (mdadm) on Intel configurations unless hardware RAID is requested.
For workloads requiring extreme IOPS, we configure all-NVMe arrays of 8+ drives. Sequential reads in those configurations reach 25 GB/s; random IOPS exceed 5M across the array. The cost scales (eight-drive configurations sit at €450-600/month depending on capacity) but the performance ceiling is high enough for serious database work.
Each datacenter hosts its own BGP-announced IP space we control. We do not sub-allocate from upstream provider ranges, which means our IP allocations follow our policies, our rDNS configuration, and our reputation management discipline. Major locations announce /22 to /20 blocks with IPv4 capacity in the low thousands per region; IPv6 announcements are /36 to /32 with effectively unlimited /64 allocations.
Transit upstreams vary by location. Bulgaria runs Cogent + NTT + Telia (Arelion). Romania runs Cogent + NTT + Telia. Moldova runs Cogent + NTT. Panama runs Cogent + Telxius. Hong Kong runs NTT + Cogent + CN2 (China-direct). Singapore runs NTT + Telia + Cogent. Ukraine runs Cogent + NTT. Direct peering with Google, Microsoft, Apple at most EU locations; APAC peering through carriers' existing relationships.
Default uplink is 1 Gbps per server with traffic policies per plan (some plans capped at 30 TB monthly, others unmetered). 10 Gbps upgrade available as €80/month addon. 25 Gbps for high-throughput operations on request. Multi-NIC configurations (LACP bonding for aggregate throughput, separate management network) supported on EPYC and dual-socket Intel platforms.
DDoS scrubbing capacity ranges from 100 Gbps in Moldova and Panama to 500 Gbps in Bulgaria, Romania, Singapore. Mitigation runs at the network edge before traffic reaches your server. L3/L4 mitigation included; L7 (application-layer) available as €49/month addon. For sustained large-scale attacks, dedicated scrubbing capacity is configurable on request; pricing depends on attack profile and capacity required.
Hardware monitoring runs continuously. Disk SMART metrics polled every 60 seconds. RAID array health monitored via out-of-band controller telemetry. CPU thermal sensors read continuously; alarm thresholds trigger before silicon throttling activates. ECC RAM events logged; correctable errors above threshold trigger DIMM replacement. NIC link state, transit utilization, packet loss tracked. PSU health monitored via IPMI sensors.
Customer notifications happen through Telegram and email. Maintenance windows scheduled with 72-hour notice minimum, generally during regional low-traffic windows (02:00-06:00 local). Emergency maintenance (responding to active hardware failure) initiated immediately with customer notification within 15 minutes of action. Status page reflects ongoing maintenance and recent incidents; historical data available going back 12 months.
Backup is customer responsibility by default. We do not run backup of customer-owned servers. For customers wanting managed backup, we offer Backup-as-a-Service addon (encrypted offsite backup to second jurisdiction, €19/month for up to 500 GB). Recovery testing on request.
Single-tenant means the entire physical machine is yours. No virtualization layer, no shared CPU, no neighbour stealing IOPS from your NVMe array. The hardware is dedicated to one customer. Compare with VPS where 10-50 customers share one physical host through a hypervisor; performance is variable because the hypervisor schedules CPU time across all tenants.
Jurisdiction defines which legal framework governs your hardware, your data, and complaint resolution. A US-based server is subject to DMCA takedown notices and US judicial process. A Bulgaria-based server is subject to Bulgarian law and EU directives, with no automatic enforcement of US instruments. For operators handling content with potential legal exposure across multiple countries, jurisdiction selection is the first architectural decision.
Standard configurations: 4-8 hours from confirmed payment. Custom hardware (specific CPU, more RAM than baseline, RAID-10 arrays, dual NICs, GPUs): 24-72 hours depending on stock. We confirm stock and timing on Telegram before invoicing.
Yes. Default OS images: AlmaLinux 9, Debian 12, Ubuntu 24.04 LTS, Rocky Linux 9, FreeBSD 14, Windows Server 2022. Custom ISO mounting is available via IPMI for any other OS. Operators running specialised distributions (Gentoo, Arch, custom kernels for forensics or networking) provision through IPMI without restriction.
Bitcoin (BTC and Lightning), Monero (XMR), Ethereum (ETH), USDT (ERC-20, TRC-20, BEP-20), USDC, Solana, Litecoin, TRON, Bitcoin Cash, Dogecoin, DAI. All payments processed through self-hosted BTCPay Server. No third-party payment processors. No KYC for standard plans. Identity verification is not required to provision a server.
Hardware monitoring detects failures (disk SMART errors, RAID degradation, RAM ECC events, thermal anomalies, NIC failures) within 60 seconds. We initiate replacement immediately. Typical replacement windows: NVMe drives 2-4 hours, RAM 4-6 hours, CPU 8-12 hours, motherboard 12-24 hours. SLA credits apply per the published SLA schedule for downtime exceeding the replacement window.
Volumetric DDoS protection (L3/L4) is included on all servers across all locations. Capacity ranges from 100 Gbps in Moldova and Panama to 500 Gbps in Bulgaria, Romania, and Singapore. Application-layer (L7) protection is available as a recurring addon at €49/month. For operations expecting sustained large-scale attacks, dedicated scrubbing capacity is configurable on request.
Yes. Bring-your-own-IP (BYOIP) is supported across all locations with operational verification. You provide LOA documentation showing IP block ownership, RPKI configuration if applicable, and reverse DNS delegation. We announce your block via BGP from the chosen datacenter. BYOIP setup typically takes 5-10 business days depending on registry response times. Useful for operators with established IP reputation that should follow them across hosting transitions.
Iron-E3 includes 30 TB monthly traffic on 1 Gbps uplink, then 100 Mbps unmetered. Iron-E5 includes unmetered traffic on 1 Gbps. Iron-EPYC includes unmetered on 10 Gbps. 10 Gbps upgrade available on E3/E5 plans for €80/month. 25 Gbps on request. We do not surprise-bill bandwidth overages; the policy is clear at provisioning.
Technically yes. PowerMTA's VMTA architecture supports unlimited sending domains on one server. Reputation-wise, the answer depends on volume and traffic mix. Sustained operations at 100K+ daily messages per server benefit from per-domain IP rotation; lower volumes work fine on a single IP across multiple sending domains. We help work out the right architecture during onboarding consultation.
Cancel anytime, no contracts, no penalty. Data migration to another provider or location is your responsibility; we cooperate with reasonable migration timelines (30-day grace period after cancellation request, IP reverse DNS preserved through transition). Migration to another of our locations: free. Migration with IP allocation transfer requires BYOIP-style setup at the destination.
Telegram order takes 10 minutes. Stock confirmed in real time. Server provisioned within 4-8 hours of confirmed payment. IPMI credentials and rDNS aligned to your sending domain delivered with provisioning notification. Cancel anytime, no contracts.
# Median Telegram response: 12 minutes during operating hours