Disclosure: we may earn a commission or referral reward from CloudCone, Contabo, DigitalOcean (account credit, not cash), Hostwinds, InterServer, RackNerd, VPSDime and Vultr if you sign up through our links. It does not change the price you pay. Every other provider pays us nothing. Prices in this guide are as of the date shown and may have changed; see the comparison pages for the latest checked prices.
Guides · BLOG
vCPU vs CPU: what you're actually buying in a VPS (2026)
Some prices quoted in this guide are out of date. It is being rebuilt on our price index; for current prices use the comparison pages, which show when each price was checked.
Every cloud VPS spec sheet lists "vCPU" and every buyer treats it as if it means a whole CPU core. It rarely does. Understanding what vCPU actually maps to determines whether you get more compute than you paid for, exactly what you paid for, or considerably less — and whether the $48/mo DigitalOcean box is genuinely 6x the compute of an $8 Hetzner box or whether they're closer than they look.
What "CPU" and "vCPU" actually mean
A CPU is a physical processor. Modern server CPUs (Intel Xeon, AMD EPYC) have many cores per chip — 16, 32, 64, sometimes 96 or more. Each core can typically run 2 hardware threads simultaneously via SMT (Intel Hyper-Threading, AMD SMT).
A vCPU is a virtualised presentation of one of those hardware threads to a guest operating system. In cloud parlance:
- 1 vCPU = 1 hardware thread, typically — but not always.
- On dedicated instances (which cost more), 1 vCPU maps to 1 hardware thread that is fully yours.
- On shared instances (the default), 1 vCPU maps to a fraction of a hardware thread you share with other tenants.
Every provider does this differently. What "1 vCPU" is worth on a $6 DigitalOcean Basic Droplet is not what "1 vCPU" is worth on a $12.32 OVHcloud VPS-3.
The oversubscription reality
Providers put more virtual cores on a physical host than there are physical cores available. This works because most workloads are idle most of the time. When you request "1 vCPU", you get scheduled fractional time on a real hardware thread. If your neighbours are quiet, you get most of a thread. If they're busy, you compete.
The oversubscription ratio varies:
- Dedicated / bare-metal cloud (e.g., Hetzner CCX line, DigitalOcean Premium/CPU-Optimized): 1:1. What you pay for is what you get. Higher price.
- Shared "burstable" tier (DigitalOcean Basic, Vultr Regular, Linode Nanode, AWS Lightsail): ~1.5-3:1. Fine for bursty workloads, less predictable under sustained load.
- Aggressive shared tier (Contabo Cloud VPS, OVHcloud VPS-1): ~3-5:1. Excellent price per rated vCPU, real trade-off if you run sustained CPU.
How this shows up in real benchmarks
For a rough calibration, single-thread Geekbench 6 scores on similar-generation hardware:
- Dedicated 3.5 GHz Intel Xeon Ice Lake core (bare metal): ~2500
- Vultr High Frequency 1 vCPU (3.7 GHz, low oversubscription): ~2400
- Hetzner CX line (Intel Xeon, moderate oversubscription): ~2000
- Hetzner CPX line (AMD EPYC, moderate oversubscription): ~2100
- DigitalOcean Basic Droplet 1 vCPU: ~1500
- Vultr Regular 1 vCPU: ~1400
- Contabo Cloud VPS 1 vCPU: ~1200-1600 (variable)
- AWS Lightsail 1 vCPU: ~1300
The gap between "premium" vCPU (Vultr HF at $6/mo) and "budget" vCPU (Contabo at ~$2/mo per vCPU) is roughly 2x on single-thread. Sometimes bigger, sometimes smaller.
What this means for buying
If your workload is CPU-bound (video encoding, ML inference, unoptimised sync web frameworks like Django without gunicorn, PHP without OPcache):
- Buy premium vCPU (Hetzner CPX / Vultr HF / DO Premium tier).
- Do not buy shared tier just because it's cheaper.
If your workload is I/O-bound (typical web app, database with occasional heavy queries, file server, Postgres, most Node.js APIs):
- Buy the cheapest shared tier that has enough RAM.
- Hetzner CX / DigitalOcean Basic / Contabo are all fine.
- Save the money for RAM and storage, which usually bottleneck you first.
If your workload spikes CPU occasionally (WordPress with occasional cache misses, small SaaS backends):
- Buy shared burstable (DO Basic, Vultr Regular). You get bursty performance when you need it and pay burst prices, not premium prices.
The multi-thread question
Almost all workloads scale poorly beyond the first few cores. Adding a 6th vCPU rarely doubles the throughput of your 3-core setup. If your Node app is single-threaded, more vCPUs help only if you cluster (PM2 cluster mode). If your Rails app uses Puma with multiple workers, each worker takes a core, but the memory footprint scales linearly.
Before buying an 8 vCPU plan because "more is better", profile your workload. On most web apps, the 4 vCPU tier is the last one where you get proportional performance gains from adding cores.
Cost per vCPU across our 10 providers
Cost per vCPU at each provider's cheapest 4-core-or-more plan (as of 2026-09-22):
| Provider | Plan | $ per vCPU |
|---|---|---|
| Contabo | Cloud VPS 6 | $1.77 (6 vCPU / $10.62) |
| Hetzner | CX32 | $1.90 (4 vCPU / $7.60) |
| Contabo | Cloud VPS 4 | $1.95 (4 vCPU / $7.79) |
| OVHcloud | VPS-3 | $2.05 (6 vCPU / $12.32) |
| OVHcloud | VPS-2 | $2.13 (4 vCPU / $8.50) |
| Contabo | Cloud VPS 8 | $2.48 (8 vCPU / $19.82) |
| Hetzner | CPX31 (AMD) | $3.90 (4 vCPU / $15.60) |
| UpCloud | Starter 8 GB / 4 vCPU | $5.40 (4 vCPU / $21.60) |
| Vultr | Regular 8 GB | $10.00 (4 vCPU / $40) |
| DigitalOcean | Basic 8 GB | $12.00 (4 vCPU / $48) |
| Akamai/Linode | Linode 8 GB | $12.00 (4 vCPU / $48) |
But cost per vCPU is not cost per unit of throughput. A Contabo vCPU at $1.77 and a DigitalOcean vCPU at $12 do not do the same amount of work. They do maybe 60% of the same work — meaning DigitalOcean's vCPU costs approximately 4-5x per unit of actual computed output at this shape, still, but not the 7x the sticker suggests.
The one thing to remember
Cheap vCPU is real value for I/O-bound workloads and stops being value when your workload is actually compute-bound. Nine out of ten small web apps are I/O bound. If yours is the tenth, spend the extra $30/mo on the premium tier — the throughput difference will be visible.
Related guides
- How to right-size your VPS — the workload-first framework.
- DigitalOcean vs Vultr 2026 — Vultr HF is the practical answer for CPU-bound work.
- Cheapest 4 vCPU / 8 GB VPS — the full price comparison at this spec.
- Hetzner Cloud pricing analysis — deeper look at CX vs CPX (Intel vs AMD).
Prices as of 2026-09-22.