Frequently asked questions

A KVM virtual machine with hardware-assisted virtualisation, running on hardware we own or operate. You get full root, your own kernel, your own firewall rules and console access. It is not a container and not a shared hosting account: nothing is sandboxed away from you.

A median of 47 seconds from confirmed payment to a reachable SSH port. Median forty-seven seconds from confirmed payment to a reachable SSH port. Windows images take three to four minutes because of sysprep. If you selected priority provisioning, the build starts within twenty seconds of confirmation instead of joining the queue.

Cores on the PULSE and VAULT families are fair-share, so they can burst above their allocation when the node is idle. Every other family gives you pinned physical cores. Memory is never overcommitted on any plan, ballooning is disabled in the hypervisor configuration, and each node keeps twenty percent of both CPU and RAM unallocated as headroom.

Yes, in both directions within a family, and across families when the target has at least as much disk. Vertical resizes are a reboot. Cross-family moves are a live volume migration and usually cost a few seconds of network interruption.

No, and we would rather be straight about it than run a crippled one. What exists instead is the first-order guarantee: if a service does not perform as described within seventy-two hours of your first order, we credit the full amount to your balance, no argument. Start on the monthly cycle, run your own benchmarks on the exact machine you would buy, and move to annual once the numbers convince you.

Because settlement has a fixed cost. Because every crypto settlement carries a fixed network cost that does not care how small the invoice is. Hourly billing would fragment one payment into hundreds, and we would have to price that overhead back into the hourly rate. Two cycles instead: monthly, or annual at thirty percent off. The annual price is the real price of the platform, and the monthly one carries the cost of billing you twelve times.

Yes: 3% off from 3 servers, 6% from 5, 10% from 10, stacked on the annual 30%. Yes, automatically, with no negotiation and no sales call. Three servers on one invoice take another three percent off, five take six percent, ten take ten percent, and it stacks on the annual discount. A ten-server annual order therefore settles at thirty-seven percent below the monthly list price. Above fifty servers, open a ticket and an engineer will price the capacity directly.

Card networks require merchants to run identity verification and give the acquirer the right to reverse a settled payment for months afterwards. Accepting cards would mean asking you for documents, so we do not accept cards. Crypto settlement through OxaPay removes that intermediary entirely and takes roughly fifteen percent out of our cost base, which is reflected in the price list.

Bitcoin including Lightning, Monero, Ethereum on mainnet, Arbitrum and Base, USDT on Tron, Ethereum, BNB Chain, TON and Solana, USDC, Litecoin, Tron, Toncoin, Solana, Dogecoin, BNB and Dai. The invoice page lets you switch asset and network before you send anything.

1 confirmation for Lightning, Tron, TON and Solana; 2 for Litecoin and Dogecoin; 3 for Bitcoin on-chain; 10 for Monero. One for Lightning, Tron, TON and Solana, which is effectively instant. Two for Litecoin and Dogecoin. Three for Bitcoin on-chain and ten for Monero. Provisioning starts the moment the threshold is reached, without a human step.

Underpayments below two percent are accepted and the shortfall is deducted from your balance. Larger underpayments sit on the invoice as a partial payment and you can top it up from the same page. Overpayments are credited to your account balance automatically and can be spent on anything, including renewals.

Seven days before a service expires we email an invoice. If your balance covers it and automatic renewal is enabled, we take it from the balance and nothing else happens. If not, pay the invoice from the dashboard. Services are suspended three days after expiry and destroyed seven days after suspension.

Yes. Add the details in the billing settings and every future invoice carries them, with reverse-charge applied for valid EU VAT numbers. The fields are optional, and leaving them empty does not restrict anything.

Within seventy-two hours of a first order, if the service does not perform as described, we credit the amount to your account balance. Crypto settlement is one-way, so we cannot send funds back on-chain, and the balance is the mechanism we have. On annual cycles we credit unused whole months if you cancel more than sixty days before renewal.

No. CrestVPS never asks for identity documents, a phone number, an address or a card. Never. There is no field in our systems to store one and no support workflow that requests one. Account creation is an email address and a password. If anyone contacts you claiming to be CrestVPS support and asks for a document, it is not us.

Web access logs with IP addresses, rotated after seventy-two hours. Hypervisor event logs, rotated after fourteen days. Your email address, invoices, server allocations and tickets, kept while the account exists. We do not perform deep packet inspection on customer uplinks and we do not log DNS queries on the resolver we provide.

We check whether the request is valid legal process from the jurisdiction where the hardware physically sits. If it is, we comply with exactly what it compels and nothing more, which in practice is an email address, a payment reference and a server allocation. If it is not, we refuse. Both outcomes are counted in the transparency report we publish every six months.

VPNs and proxies for your own use or your own users are fine and are one of the most common workloads on the platform. Tor middle relays and bridges are allowed everywhere. Tor exit nodes are allowed only in Amsterdam, Reykjavik and Bucharest, and require you to open a ticket first so we can flag the address with our upstreams.

Add your PGP public key in settings and every message we send, including invoices and root credentials, is encrypted to it. If you lose that key you will need to remove it from the account before you can read our mail again.

Then we cannot recover the account, because the email address is the only identifier we hold for you. This is the direct cost of not collecting anything else. Enable two-factor authentication, store the recovery codes offline, and consider using an address you control the domain for.

Around fifty images, including Debian, Ubuntu, AlmaLinux, Rocky, Fedora, Arch, Alpine, NixOS, Gentoo, Void, Devuan and Slackware; FreeBSD, OpenBSD, NetBSD, DragonFly and HardenedBSD; Windows Server 2019 through 2025 and Windows 11 Pro; Proxmox, TrueNAS, OPNsense, pfSense, VyOS and RouterOS; and a set almost nobody else offers, such as 9front, Haiku, SmartOS, OpenIndiana, MINIX 3 and ReactOS. You can also upload an ISO of up to eight gigabytes or netboot with iPXE.

Upload an ISO from the dashboard and it appears as a virtual CD on your machine, or boot netboot.xyz and pull whatever you want over the network. The web console gives you a graphical installer, so anything that installs on ordinary KVM will install here.

Yes, at no cost, on any plan with dedicated cores. It is what makes Proxmox, nested KVM, Docker Desktop and Windows Sandbox work. Enable it in the builder or toggle it later from the server page; it takes effect on the next boot.

Two manual snapshot slots are included on every plan, taken at the block layer with no downtime. Paid tiers add nightly automatic backups with seven or thirty restore points. Off-site replication encrypts the volume on the node before shipping it to a different country, so the destination facility never holds a readable copy.

Select LUKS2 with remote pre-boot unlock. The volume header is on your disk, the passphrase never touches our systems, and after each boot the machine sits in an initramfs with Dropbear waiting for you to SSH in and unlock it. We hold no key and cannot recover the data if you lose the passphrase.

Open the web console. It is a noVNC session attached to the virtual display, plus serial-over-SSH, and it works regardless of what you did to the network stack. There is also a rescue mode that boots a live image with your volume attached but unmounted.

Yes. These are full virtual machines with their own bootloader. Compile what you want, load whatever modules you want, patch the scheduler if that is your idea of an evening. Nothing is loaded from the host into your guest.

Full PCIe passthrough of a physical card, not a time-sliced vGPU. You get the whole device, its full memory and direct driver access. CUDA 12 and ROCm images are prebuilt, and NVLink pairs are available on request in Amsterdam, Frankfurt and Ashburn.

One is included, up to 16 individually, or a routed /29. One is included. You can add up to sixteen individually or take a routed /29. Larger allocations need a short justification because of RIPE policy, not because of us. Every address is checked against the major blocklists before it is handed out.

A routed /64 comes free with every server and a /48 delegation is three euros a month. Reverse DNS is self-service for the whole range from the dashboard.

It is closed by default, which is what keeps our ranges clean. Ask for it in a ticket, tell us roughly what you are sending, and we open it. Ranges with a warmed-up sending history are available for legitimate transactional mail.

It runs inline at the edge of each PoP, always on, on the same path your normal traffic takes. There is no detection window to wait through and no rerouting to a third-party scrubbing centre, so there is no latency penalty and no BGP flap when an attack starts. Volumetric floods are absorbed upstream of your uplink; the L7 tier adds protocol signatures for game and voice traffic.

Yes. We run AS204871 and will announce your RIPE or ARIN prefix, or set up a full eBGP session so you announce it yourself. You will need a valid LOA and a matching route object. Anycast across six of our PoPs is available on top of that.

The port drops to 100 Mbps until the cycle rolls over. Nothing is billed by surprise and nothing is suspended. You can lift the cap by buying extra traffic from the dashboard at any point, or move to an unmetered plan.

99.9% network and power availability per calendar month as standard, or 99.99% with the upgrade. 99.9% network and power availability per calendar month as standard, measured per server, with credits starting at ten percent of the monthly fee for the first tenth of a percent missed. The 99.99% upgrade multiplies credits tenfold. Credits are applied to your balance automatically when our own monitoring detects the breach; you do not have to claim them.

Any hour of any day, with a median first reply of 11 minutes over the last 30 days. Always. Median first reply over the last thirty days is eleven minutes, and the person who answers has root on the hypervisor your server runs on. There is no first-line script and no ticket bouncing.

Two hours of engineering per month for eighty-nine euros, used for hardening, migrations, performance work or anything else you would otherwise do yourself. It is not a monitoring-and-patching contract; it is direct access to the people who build the platform.

Yes, at no charge, for any order above fifty euros a month. Send us access details for the source machine in a ticket and we handle the copy, the DNS cutover plan and the verification. Most migrations complete inside two hours.