Web hosting & SaaS
Put client sites, a busy WooCommerce store or a multi-tenant SaaS on a dedicated server for web hosting, with ECC memory, mirrored NVMe and the control panel you prefer.
All three recommended servers use ECC RAM, which corrects single-bit errors before they reach a database page or a cached file.
Two NVMe drives in software RAID 1 out of the box, so a failed disk means a swap within hours, not a restore from backup.
Install cPanel & WHM or Plesk with your own license, an open-source panel such as HestiaCP, or plain nginx and PHP-FPM.
Set the PTR record from the dashboard so outgoing mail matches your hostname, one of the first checks receiving servers run.
Recommended servers
From a first deployment to a busy production setup. Live stock and prices; in-stock picks are online in minutes.
Agency or freelancer
6C / 12T · 3.4–4.8 GHz
Six cores up to 4.8 GHz, 32 GB DDR4 ECC and 2 × 512 GB NVMe: a lean panel server for brochure sites, staging and small WordPress installs.
Our pickBusy stores and panels
8C / 16T · 3.2–5.1 GHz
Eight cores up to 5.1 GHz and 64 GB ECC for uncached PHP: WooCommerce checkouts, cPanel with many accounts and MariaDB on 2 × 1 TB NVMe.
SaaS and multi-tenant apps
12C / 24T · 3.7–5.4 GHz
12 Zen 4 cores up to 5.4 GHz, 128 GB DDR5 ECC and 2 × 1.92 TB NVMe: app, queue workers, database and Redis on one box with room to grow.
Sizing
Site counts are rough guides for typical sites behind a page cache; one uncached plugin or a traffic spike changes the picture, so watch PHP-FPM queues and load after you move.
| Workload | Typical need | Suggested plan |
|---|---|---|
| Agency panel, 20–50 cached WordPress sites | 4–6 cores, 16–32 GB, page cache on every site | Intel Xeon E-2236 |
| WooCommerce store with steady checkouts | High clocks for uncached PHP, Redis object cache | Intel Xeon E-2388G |
| cPanel server with 100+ small accounts | Cores and RAM for PHP, mail and DNS together | Intel Xeon E-2388G |
| Laravel, Rails or Django SaaS with queues | App, workers, Redis and database on one host | AMD EPYC 4464P |
| Database-heavy multi-tenant SaaS | 256 GB for buffer pools, 10 Gbps to replicas | AMD EPYC 7443P |
Expert notes
Short, practical advice from our engineers for this workload.
Each PHP-FPM worker serves one request at a time, so pm.max_children is your real concurrency limit. Size it from two angles and keep the smaller result:
ps -o rss= -C php-fpm8.3 (adjust to your PHP version), instead of guessing.Give each site its own pool and system user so one compromised plugin can’t read its neighbors’ files. Use pm = ondemand for quiet sites and dynamic for busy ones, and size OPcache so it never evicts scripts.
Anonymous visitors should never reach PHP: a full-page cache (nginx FastCGI cache, LSCache on OpenLiteSpeed or a caching plugin) answers them from memory or disk. The expensive traffic is what can’t be cached: carts, checkout, account pages and logged-in users. Those requests run PHP and SQL every time, which is where high clock speeds earn their keep.
innodb_buffer_pool_size so frequently read tables and indexes stay in RAM, leaving room for PHP and the page cache.DISABLE_WP_CRON and run wp cron event run --due-now from a system timer, so scheduled jobs stop riding on visitor requests.Mail from a fresh server IP is judged on its setup first:
A mirror copies mistakes as faithfully as data: a deleted folder or a hacked site lands on both drives at once. Ship nightly, encrypted backups off the machine, for example with BorgBackup or restic to the backup space add-on, and test a restore before you need one.
Root access means you install exactly what you need. These are proven choices for this workload.
FAQ
Yes. Install a supported OS in one click, for example AlmaLinux or Rocky Linux for cPanel & WHM, then run the vendor’s installer over SSH. Licenses are bought from cPanel or Plesk directly. To avoid license costs, HestiaCP and Virtualmin cover web, mail and DNS for free, while CloudPanel focuses on running PHP, Node.js and Python apps on nginx.
It depends on what the sites do more than on how many there are. A server full of cached brochure sites mostly idles, while one busy WooCommerce store can keep eight cores working. Estimate peak load as uncached requests per second times the average PHP time per request: the result is roughly the number of cores busy at peak.
Rarely for SSL: with SNI, any number of HTTPS sites share one IPv4 address, and every server already includes one IPv4 plus a routed /64 of IPv6. Extra addresses help when you want to separate mail reputation or give a client a dedicated IP. You can add up to four per server without justification, and larger blocks with one.
No. You get full root and handle the operating system, panel, updates and site support yourself, or with your own team. Our engineers cover hardware, network, provisioning, IPMI access and DDoS mitigation around the clock, with a 30-minute first-response target when a server is down. For software questions we share docs and pointers on a best-effort basis.
Partly. Network-level floods such as SYN, UDP and amplification attacks are filtered upstream at no extra cost. Application-layer attacks, like waves of valid-looking HTTP requests against a login or search page, are outside that layer: put a reverse proxy, CDN or WAF in front of busy sites and add rate limits in nginx. Together, they cover both layers.
Pick one of the recommended in-stock servers and it can be online in minutes. Need something bigger or different? Our engineers quote custom builds within one business day.