VPS Performance Benchmarks 2026: Linode, Vultr, Hetzner, Contabo Tested

# VPS Performance Benchmarks 2026: Linode, Vultr, Hetzner, Contabo Tested

June 18, 2026 -- 11:47 PM, Berlin time. Coffee cold. Terminal windows stacked three deep.

I did this because I'm tired of reading "best VPS provider" lists that cite 2023 benchmarks or cherry-pick one test -- like "look how fast their NVMe is!" -- while ignoring what actually matters when your WordPress site chokes at 14:03 on a Tuesday because the database query took 800ms *again*. I run four production sites (a client's WooCommerce store, my own tech newsletter, a small SaaS dashboard, and a dev playground), and over the last 18 months, I've migrated between providers *six times*. Not for price -- I'll pay more for reliability -- but for *predictable performance*. So in early June 2026, I spun up identical $10/month tier VPS instances across Linode (Shared CPU, 2 vCPU / 4GB RAM), Vultr (Cloud Compute, 2 vCPU / 4GB RAM), Hetzner (AX41, 4-core AMD EPYC 9124 / 32GB RAM -- yes, *way* over-provisioned, but it's their entry-level dedicated-like cloud instance), and Contabo (VPS S, 4 vCPU / 8GB RAM / 160GB NVMe). All running Ubuntu 24.04 LTS, kernel 6.8.0-55-generic, fully patched. No custom tuning -- just clean installs, same PHP 8.3.12 + Nginx 1.24.1 + MariaDB 11.4.4 stack.

Here's exactly what I tested -- and how:

- CPU: Geekbench 6.4.0 (single and multi-core), plus 'sysbench cpu --threads=2 --cpu-max-prime=20000 run' (time to compute primes up to 20,000, lower = better)

- Disk I/O: 'fio' with 4K random reads/writes (IOPS) and 1MB sequential reads/writes (MB/s), direct=1, ioengine=libaio, queue_depth=32, runtime=60s -- repeated 3x, median taken

- Network latency: 'ping' from 5 global locations (New York, Tokyo, Frankfurt, Sao Paulo, Sydney) -- averaged over 100 pings per location

- Real-world web load: ApacheBench ('ab -n 5000 -c 100') hitting a stock WordPress 6.7 install (no caching plugins, default theme, 10 sample posts, WP_DEBUG off) served over HTTPS via Nginx + PHP-FPM (pm=dynamic, start_servers=4). Measured requests/sec, 90th percentile response time (ms), and error rate

All tests ran during off-peak EU hours (22:00-04:00 CEST) to minimize host noise. I waited 10 minutes after reboot before testing. No swap was enabled. Disk was formatted ext4 with 'noatime'. Full test scripts are on my GitHub (https://github.com/serverpicks/vps-bench-2026), if you want to replicate.

CPU Benchmarks

Geekbench favors consistent cores; sysbench stresses real-world threading. Hetzner's AX41 surprised me -- not just raw speed, but *consistency*. Linode's shared CPU showed noticeable variance between runs (+-8% on multi-core Geekbench), while Vultr felt snappy but capped hard at ~2.8GHz under load.

Winner: Hetzner -- no contest. That EPYC 9124 delivers desktop-tier single-core and serious parallel throughput.

Loser: Linode -- not broken, but noticeably less headroom under sustained load. You feel it when cron jobs pile up.

NVMe Disk I/O (fio, 4K random read/write)

This is where Contabo *should* shine -- and it did... until I looked closer. Their advertised "NVMe" is real, but queue depth handling feels brittle. At QD32, they hit high IOPS, but drop sharply under mixed workloads. Hetzner's AX41 uses enterprise-grade NVMe (likely Samsung PM1743), and it *stays* stable. Linode's SSDs are predictable but modest -- fine for blogs, tight for DB-heavy apps.

Winner: Hetzner -- best-in-class across the board, especially write IOPS. That 22.4K random writes? My WooCommerce cart updates *fly*.

Loser: Linode -- solid, but not competitive here. If you're running Redis or heavy logging, you'll notice the gap.

Network Latency (ms, median ping)

I used 'ping' from DigitalOcean NYC, AWS Tokyo, Scaleway Frankfurt, AWS Sao Paulo, and AWS Sydney. All providers routed through standard peering -- no magic BGP tricks. Vultr's global anycast DNS helped *some*, but physical distance dominates. Hetzner's Frankfurt node is stellar for EU traffic, but Asia latency is its weak spot. Contabo's network feels... unoptimized. Packets sometimes take bizarre paths (I saw traceroutes hop via Warsaw -> Chicago -> Frankfurt).

Winner: Vultr -- consistently lowest across *all* regions. Their edge POP density pays off.

Loser: Contabo -- not unusably slow, but 17ms to NYC vs. Vultr's 9.8ms adds up under API-driven apps.

Real-World Web Load (WordPress, ab -n 5000 -c 100)

This is the test I care about most. No synthetic stress -- just "what happens when 100 people land on your homepage at once?" I measured three things: requests/sec (throughput), 90th %ile response time (real user experience), and failed requests (anything >10s timeout).

Hetzner pulled ahead cleanly -- fastest throughput, lowest tail latency. But here's the real story: I re-ran the test *with MariaDB query cache disabled* (simulating dynamic content). Hetzner's advantage widened (112 req/sec vs. Vultr's 94). Contabo's two failures? Both were DB connection timeouts -- confirmed in 'journalctl -u mariadb'. Their MySQL config defaults to 'max_connections=151', and with 100 concurrent ab workers + background processes, it choked. I had to manually raise it to 256. Linode and Vultr handled it out-of-the-box.

Final Verdict: Who Wins Where?

- Best Overall Performer: Hetzner. It's not just speed -- it's *consistency*. That AX41 instance delivered top-tier CPU, disk, *and* web throughput without fiddling. Yes, Asia latency stings, but for EU/US-focused projects? It's the new gold standard. Downsides: support tickets take 12-24h, and their portal feels like it's running on a Raspberry Pi. But performance? Unmatched.

- Best Global Network: Vultr. If your users are everywhere, and you need low-latency APIs or real-time dashboards, Vultr's infrastructure is mature, predictable, and well-documented. Their CLI and Terraform provider just *work*.

- Most Reliable Baseline: Linode. Nothing dazzles, but nothing breaks. Perfect for "set-and-forget" staging environments or low-traffic sites where uptime > peak speed. Their monitoring and alerting are genuinely useful.

- Most Value-Heavy (With Caveats): Contabo. You get raw specs that embarrass the others -- *if* you're willing to tune. Their defaults assume you know what you're doing. I spent 90 minutes tweaking 'sysctl.conf', MariaDB buffers, and Nginx worker connections. Worth it? For a personal project or homelab -- absolutely. For client-facing production? Only if you have ops bandwidth.

I'm migrating my newsletter and SaaS dashboard to Hetzner next week. The WooCommerce store stays on Vultr -- its Tokyo latency matters for our Japanese subscribers. Linode keeps the staging env. And Contabo? I'm keeping it as my "break stuff and learn" sandbox. Because honestly? Watching 'fio' hit 42K IOPS on a 7.99/mo instance still makes me grin.