I Benchmarked Three VPS Servers So You Don’t Have To: Netcup vs Oracle Free Tier
Every few months I run through my server stack and put the candidates through their paces. This time I had three contenders sitting in front of me: my existing Netcup VPS in Virginia, a freshly provisioned Netcup root server with DDR5 RAM in Nuremberg, and the Oracle A1 Free Flex instance I finally managed to claim after months on the waitlist. I ran the same Yet Another Bench Script across all three, and the results were… revealing.
Netcup DDR5 (Germany)
Best overall
Netcup VPS (Virginia)
Runner-up
Oracle Free A1
Caveats apply
Netcup DDR5 DE
Oracle A1 Free
★ = category winner
The Contenders
Before we get into numbers, here’s what we’re working with:
Netcup VPS (Virginia, US): 4 vCPU on an AMD EPYC-Genoa processor, 7.7 GB RAM, 503 GB storage, running Ubuntu 22.04. Located in Manassas, Virginia.
Netcup Root Server DDR5 (Nuremberg, DE): 4 vCPU on an AMD EPYC 9645 (96-core Genoa-based platform), 7.7 GB RAM, 251 GB storage, Ubuntu 24.04. The DDR5 designation here refers to the host platform’s memory architecture, which matters for memory bandwidth.
Oracle Cloud Free A1 Flex (Chicago, US): 4 vCPU ARM Neoverse-N1, 23.4 GB RAM, 45 GB storage, Ubuntu 24.04. The crown jewel of Oracle’s Always Free tier. ARM architecture, noted as experimental in YABS.
CPU Performance: Intel vs ARM, or Rather AMD vs ARM
Geekbench 6 gives us a clean apples-to-apples view here:
| Server | Single Core | Multi Core |
|---|---|---|
| Netcup DE (DDR5) | 1,988 | 6,196 |
| Netcup US (Virginia) | 1,608 | 5,058 |
| Oracle A1 Free | 1,088 | 3,631 |
The Netcup DDR5 machine pulls ahead by a meaningful margin. The EPYC 9645 is a newer silicon generation and it shows. The Oracle ARM instance comes in last — not terrible in absolute terms for a free server, but for PHP-heavy WordPress workloads (WooCommerce checkout, LearnDash quiz processing, BuddyBoss activity queries), single-core performance is what actually matters. And at 1,088 points, it’s 46% slower per core than the DDR5 Netcup.
For WordPress specifically, higher single-core scores mean faster PHP execution, faster WP-Cron jobs, and snappier admin panel responses.
Disk I/O: The Number That Will Make or Break Your WordPress Site
This is where things got shocking.
| Server | 4K Random Read | 4K Random Write | 1M Sequential Read |
|---|---|---|---|
| Netcup DE (DDR5) | 201 MB/s (50.3K IOPS) | 201 MB/s (50.4K IOPS) | 1,590 MB/s |
| Netcup US (Virginia) | 189 MB/s (47.4K IOPS) | 190 MB/s (47.5K IOPS) | 541 MB/s |
| Oracle A1 Free | 6.26 MB/s (1.5K IOPS) | 6.26 MB/s (1.5K IOPS) | 24 MB/s |
The Oracle numbers are not a typo. At 1,500 IOPS for 4K random reads, Oracle’s free tier disk is roughly 33x slower than Netcup. For WordPress, this translates directly to:
- Slow database queries. Every MySQL lookup, every
wp_postsjoin, every options table read hits this ceiling. - Painful plugin activation/updates. File operations crawl.
- Object cache misses hurt more. When Redis or Memcached can’t serve a request, the fallback disk read is painfully slow.
- Backup jobs drag. Running UpdraftPlus or similar on this disk is going to be painful.
The Netcup DDR5 node, on the other hand, is genuinely exceptional. Sequential reads above 1.5 GB/s put it in enterprise NVMe territory. For production membership sites running MemberPress, LearnDash, and BuddyBoss, this makes a tangible difference.
Network Performance: Oracle’s Surprise Party
Here’s the twist. Oracle’s networking turned out to be legitimately competitive, especially for US-based traffic:
| Route | Netcup US | Netcup DE | Oracle Free |
|---|---|---|---|
| NYC send | 2.73 Gbps | 1.94 Gbps | 3.95 Gbps |
| LA send | 2.61 Gbps | 1.12 Gbps | 3.84 Gbps |
| London send | 710 Mbps | 2.70 Gbps | 724 Mbps |
| Amsterdam send | 38 Mbps | 2.67 Gbps | 757 Mbps |
Oracle absolutely dominates US domestic routes, and Netcup DE is the clear winner for European connectivity (as you’d expect from a Nuremberg data center with sub-11ms ping to Amsterdam). The Netcup Virginia node had notably inconsistent European sends, which is worth flagging if you’re hosting European users from a US server.
For most WordPress sites, network bandwidth isn’t the bottleneck — but for sites serving large media assets, video-based LMS platforms, or running CDN origin pulls, these numbers matter.
What Does This Actually Mean for WordPress Hosting?
Here’s my honest take, server by server:
Netcup DDR5 Germany: The clear winner if your audience is in Europe. Best CPU, best disk, solid network. At roughly €5-10/month (depending on configuration), this is extraordinary value for production membership and LMS sites. This is what I now use as the primary host for several client sites.
Netcup Virginia: A solid, reliable performer for US-targeted sites. The EPYC-Genoa CPU is a generation behind the 9645, which shows in benchmarks, but it’s still miles ahead of Oracle for day-to-day WordPress performance. The larger disk (503 GB vs 251 GB) is a real advantage for media-heavy sites.
Oracle A1 Free (ARM): Here’s the nuanced answer. The 24 GB of RAM is genuinely exciting, and the network bandwidth is impressive. But the disk I/O makes it unsuitable as a primary production WordPress host. Where it shines:
- As a staging environment with a RAM-based tmpfs for MySQL (if you’re running tests, you can mount /var/lib/mysql to RAM and sidestep the disk entirely)
- As a Redis/Valkey node — caching is memory-bound, not disk-bound
- As a build server for CI/CD pipelines
- As an additional compute node in a multi-server stack where your primary database lives elsewhere
The ARM architecture is also worth noting. Most WordPress plugins work fine on ARM, but if you’re running custom native extensions or anything that compiles to x86-specific binaries, test carefully.
The Free Tier Catch Nobody Talks About
Oracle’s Always Free tier is genuinely generous on paper. But there are a few things worth knowing beyond the benchmark numbers:
- Getting an instance is hard. Oracle regularly shows “out of capacity” for A1 instances in popular regions. It took me significant time and effort to finally provision one. There are scripts people use to retry automatically, but this is a well-known pain point.
- Reliability questions. The Always Free tier is “best effort” from a support standpoint. There’s no uptime SLA. For production client sites, that’s a real consideration.
- Egress costs on paid tiers. If you ever move off the free tier, Oracle’s egress pricing is notoriously aggressive. Plan your exit strategy.
- Boot volume = block storage. That painful 1,500 IOPS figure comes from Oracle’s remote block storage architecture. It’s not local NVMe, and it shows.
My Recommendation
If you’re running WordPress, WooCommerce, LearnDash, MemberPress, or BuddyBoss in a VPS environment and your users are in Europe, Netcup’s DDR5 servers in Nuremberg are currently my top recommendation. The price-to-performance ratio is the best I’ve tested, and I’ve tried quite a few.
For US-focused sites, the Netcup Virginia VPS holds up well. I switched from Hetzner to Netcup a while back after benchmarking showed Netcup coming out ahead, and these results continue to justify that choice.
The Oracle Free tier is a brilliant complement to a paid VPS setup — use its RAM generously for caching layers, staging environments, or low-write workloads. Just don’t put your production MySQL database on it and expect good query times.
A Note on YABS
All tests were run using Yet Another Bench Script v2025-04-20. The fio disk tests use mixed 50/50 read/write with a partition-level test (not a file-level test), which gives a more realistic picture than sequential-only benchmarks. Geekbench 6 scores reflect the actual CPU resources allocated to the VM, not the underlying host’s theoretical peak.
Benchmarks are a snapshot, not a guarantee. Your workload may differ. But for PHP-heavy, database-driven WordPress sites, disk IOPS and single-core CPU score are the two numbers I watch most closely. On both metrics, the paid Netcup options win decisively.
If you’re running benchmarks on other VPS providers — Hetzner, Contabo, Vultr, DigitalOcean — I’d love to see the numbers in the comments.
Update: July 2026
Oracle A1 Free no longer offers 4vCPU/24 GB Ram. The new FREE limit is 2 vCPU/12GB RAM on ARM.
Running a Membership or LMS Site? We Have the Right Hosting for You.
If this benchmark taught you anything, it’s that not all servers are equal — and WordPress membership sites running MemberPress, LearnDash, or BuddyBoss need hosting that’s actually built for them. Our managed plans run on the same high-performance DDR5 infrastructure we tested above, with Redis object caching, KVM-isolated resources, and zero overselling. Starting at just $25/month, every plan is fully managed — server config, caching, security, and optimizations are all handled by our team.
Already on another host? We offer free site migrations — no downtime, no technical headaches. You focus on your community, we handle the move.
Frequently Asked Questions
Is Oracle Always Free good for WordPress?
Not as a primary host. Oracle Free’s ARM A1 instance offers impressive RAM (24 GB) and solid network bandwidth, but its block storage delivers only around 1,500 IOPS on 4K random reads — about 33x slower than a comparable paid VPS.
For WordPress sites running MySQL, WooCommerce, LearnDash, or MemberPress, that disk bottleneck shows up as slow database queries, sluggish WP-Admin, and poor performance under any real load. It works well as a staging server, Redis node, or dev box — but production membership and LMS sites need dedicated NVMe storage.
Which VPS is best for WooCommerce?
Look for: high single-core CPU score, 50,000+ IOPS NVMe disk, at least 4 GB dedicated RAM, and Redis object caching. In our testing, Netcup’s DDR5 VPS in Nuremberg (AMD EPYC 9645, 50,300 IOPS, Geekbench single-core 1,988) is the standout for EU-based stores. For US-based stores, the Netcup Virginia VPS performs reliably.
Avoid shared hosting — it cannot handle checkout spikes. Avoid Oracle Free tier due to its poor disk I/O. A managed VPS with KVM virtualization and dedicated resources is the minimum recommended for any WooCommerce store with real traffic.
What is a good IOPS score for WordPress hosting?
10,000+ IOPS is acceptable. 30,000–50,000+ is excellent for high-traffic sites. IOPS at the 4K block size is the most WordPress-relevant disk metric — it reflects how fast the server handles the small random reads and writes that MySQL and PHP generate constantly.
Oracle Free at 1,500 IOPS is the cautionary tale. Netcup DDR5 at 50,300 IOPS is the benchmark to aim for. For membership and LMS sites with active users, anything under 10,000 IOPS will create noticeable slowdowns during peak activity.
What is YABS and how does it benchmark VPS performance?
YABS (Yet Another Bench Script) is an open-source Linux benchmarking tool that runs three tests: fio for disk I/O (IOPS and throughput at multiple block sizes, 50/50 mixed read/write), iperf3 for network throughput to global endpoints, and Geekbench 6 for CPU performance. For WordPress hosting decisions, the most useful numbers are 4K random read IOPS (database speed), single-core Geekbench score (PHP execution), and network latency to your audience’s region.
Is Netcup good for WordPress hosting?
Yes — particularly for European audiences. Netcup’s DDR5 server in Nuremberg delivered 50,300 IOPS, a single-core Geekbench score of 1,988, and network speeds above 2.5 Gbps to London and Amsterdam in our tests. The price-to-performance ratio beats Hetzner and DigitalOcean at comparable tiers, and KVM virtualization guarantees your resources are genuinely dedicated with zero overselling. Their Virginia, US location is solid for North American audiences, though disk performance is slightly behind the German DDR5 nodes.
How much RAM does a WordPress membership site need?
Minimum 4 GB dedicated RAM, with 8 GB recommended for sites with 500+ active members. With Redis object caching, 8 GB lets WordPress serve most requests from memory without hitting the database on every page load. Sites running BuddyBoss, LearnDash, MemberPress, and WooCommerce simultaneously benefit from 16 GB or more — especially during course launches or membership sale traffic spikes.
Opulent Cloud — Miami, FL
Intel Xeon E5-2680 v3 · 4 vCPU · 15.3 GB RAM · 1 TB disk · XEN · AlmaLinux 9.7
This one came up while auditing a client’s existing environment. The E5-2680 v3 is a 2014 CPU and it benchmarks exactly like one — single-core of 615 is the lowest in the entire comparison, trailing even Oracle’s ARM chip by 43%. For WordPress that means slow PHP execution on every uncached page load. The 7,400 IOPS disk is borderline — usable for a low-traffic site, but well under the 10,000 minimum you’d want for anything with active members or WooCommerce. XEN paravirtualization without hardware passthrough adds overhead on top. The only bright spot is the large 1 TB disk and solid NYC latency at 6.36ms. If this is a client’s current host, it’s a migration conversation waiting to happen.
| Metric | Result | vs Netcup DDR5 |
|---|---|---|
| Geekbench 6 single-core | 615 | 3.2× slower |
| Geekbench 6 multi-core | 1,704 | 3.6× slower |
| 4K random read IOPS | 7,400 | 6.8× slower |
| 1M sequential read | 294 MB/s | 5.4× slower |
| NYC network send | 939 Mbps | 6.36ms ping |
OVH — Hillsboro, OR
Intel Xeon E-2288G · 4 vCPU @ 3.70 GHz · 4 GB RAM · 187 GB disk · OpenVZ · AlmaLinux 9.4
Another client audit. The E-2288G is a 2019 Coffee Lake chip that punches above its weight on single-core — 1,570 lands it right behind Netcup Virginia and well ahead of Oracle ARM. The 3.70 GHz base clock is the reason: for PHP execution, clock speed matters more than core count. The 23,500 IOPS disk is comfortably workable for a production WordPress site. The caveats are real though: OpenVZ container virtualization means shared kernel, no true hardware isolation, and resource limits the host can quietly impose. The ploop disk layer also creates overhead on large sequential reads (127 MB/s at 1M blocks). Only 4 GB RAM is tight for BuddyBoss or WooCommerce under load. Oregon location means 207ms to London — not great for European audiences. If a client is already on it and it’s working, no urgent reason to move. Choosing fresh, Netcup KVM at a similar price is a better foundation.
| Metric | Result | vs Netcup DDR5 |
|---|---|---|
| Geekbench 6 single-core | 1,570 | 1.3× slower |
| Geekbench 6 multi-core | 3,654 | 1.7× slower |
| 4K random read IOPS | 23,500 | 2.1× slower |
| 1M sequential read | 127 MB/s | 12.5× slower |
| London ping | 207 ms | Oregon → EU |
Kinsta — Google Cloud, Iowa (US)
AMD EPYC 9B14 · LXC · Ubuntu 24.04 · Google Cloud us-central1 · Staging (1 vCPU) vs Production (12 vCPU)
I ran YABS on both a Kinsta staging environment and a live production site — and the difference between the two tells the real story. The staging node had 1 vCPU allocated, scoring 1,022 single-core. The production node had 12 vCPUs, scoring 1,456 single-core and 5,509 multi-core. The EPYC 9B14 is 2024 Turin architecture — newer than anything else in this comparison — but LXC container scheduling and shared host contention pull the scores down from what the underlying hardware could deliver. Both nodes are on Google Cloud Iowa with no IPv6, no swap, and ZFS storage that made fio disk results unreliable (YABS threw an explicit warning). Google’s network backbone is the highlight: 3.78–3.95 Gbps to NYC and a surprisingly solid 1.6–1.85 Gbps to London for a US server.
The honest Kinsta verdict: production CPU is competitive (better single-core than Oracle, better multi-core than OVH), the network is excellent, but you’re paying managed hosting prices for infrastructure that raw benchmarks don’t fully justify. What you’re actually paying for is the managed stack — Nginx, Redis, Cloudflare Enterprise CDN, automated backups, and a dashboard clients can use without calling you. That’s a real value. The raw VPS comparison is just useful context.
| Metric | Staging (1 vCPU) | Production (12 vCPU) |
|---|---|---|
| Geekbench 6 single-core | 1,022 | 1,456 |
| Geekbench 6 multi-core | 1,478 | 5,509 |
| 4K IOPS | ⚠ ZFS — unreliable | ⚠ ZFS — unreliable |
| NYC network send | 3,780 Mbps | 1,600 Mbps |
| London send | 1,850 Mbps | 1,720 Mbps |
Vultr HFC — Piscataway, NJ
Intel Skylake IBRS · 4 vCPU @ 3.79 GHz · 15.6 GB RAM · 360 GB disk · Hyper-V · Debian 12
Vultr’s HFC (High Frequency Compute) tier is a server of extremes. The CPU is Intel Skylake — a 2015-era microarchitecture with Spectre IBRS mitigations adding overhead — scoring 1,363 single-core, firmly mid-pack. The disk is the best in the entire comparison at 97,800 IOPS on 4K random reads, nearly double the Netcup DDR5. The network is extraordinary: 8.78 Gbps send to NYC with a 2.40ms ping, and a remarkable 8.41 Gbps to Amsterdam for a US server, thanks to Vultr’s NJ location sitting adjacent to major internet exchange points. The “HFC” branding promises high frequency but Skylake’s IPC can’t keep pace with modern AMD EPYC, and Hyper-V virtualization adds more overhead than KVM for Linux workloads.
For a WordPress site with Redis fully warmed — where most requests are served from cache without touching PHP or MySQL — Vultr’s disk and network advantage would shine. For a cache-cold site doing heavy PHP execution and database queries, the aging CPU becomes the bottleneck. Between Vultr HFC and Netcup DDR5: Netcup wins CPU by 45%, Vultr wins disk by 2x and network by 3–4x. With Redis in place, Vultr is genuinely competitive. Without it, Netcup is the better day-to-day performer.
| Metric | Result | vs Netcup DDR5 |
|---|---|---|
| Geekbench 6 single-core | 1,363 | 1.5× slower |
| Geekbench 6 multi-core | 2,967 | 2.1× slower |
| 4K random read IOPS | 97,800 | 1.9× faster ★ |
| 1M sequential read | 750 MB/s | 2.1× slower |
| NYC network send | 8,780 Mbps | 4.5× faster ★ |
| Amsterdam send | 8,410 Mbps | 3.1× faster ★ |
The Full Picture — All Seven Servers Ranked for WordPress
| Server | Single-Core | 4K IOPS | Virt | RAM | WP Verdict |
|---|---|---|---|---|---|
| 🥇 Netcup DDR5 DE | 1,988 | 50,300 | KVM | 7.7 GB | ✅ Best overall |
| 🥈 Netcup VPS US | 1,608 | 47,400 | KVM | 7.7 GB | ✅ Strong US pick |
| 🥉 OVH Oregon | 1,570 | 23,500 | OpenVZ | 4 GB | ⚠ Mid-tier, US only |
| 4. Kinsta Production | 1,456 | ⚠ ZFS invalid | LXC | 7.5 GB | ✅ Managed value |
| 5. Vultr HFC NJ | 1,363 | 97,800 ★ | Hyper-V | 15.6 GB | ⚠ Old CPU, great disk |
| 6. Oracle A1 Free | 1,088 | 1,500 | KVM | 23.4 GB* | ❌ Disk unusable for WP |
| 7. Opulent Cloud FL | 615 | 7,400 | XEN | 15.3 GB | ❌ Aging hardware |
* Oracle Always Free RAM reduced to 12 GB (2 vCPU) for new accounts as of July 2026. Kinsta disk results excluded — YABS flagged ZFS + low disk space as unreliable. All benchmarks run using YABS v2025-04-20.
