Shared VPS plans with the most CPU steal
Shared-CPU plans ranked by how much of the CPU time they asked for went to someone else across every qualifying run.
| # | Plan | Median steal | Runs over 10% | Median worst minute | Runs | Dates |
|---|---|---|---|---|---|---|
| 1 | Amazon Lightsail Micro, 2 TB transfer | 32.0% | 7 of 7 | 88.9% | 7 runs, 4 locations | Oct 7 to Oct 8 |
| 2 | OneProvider General Purpose GP1.1-1-10 | 23.0% | 3 of 3 | 24.1% | 3 runs, 3 locations | Oct 9 to Oct 12 |
| 3 | DigitalOcean Basic 1 vCPU / 1 GB | 0.98% | 0 of 4 | 6.4% | 4 runs, 3 locations | Oct 7 to Oct 8 |
| 4 | Linode Nanode 1GB | 0.93% | 1 of 6 | 3.5% | 6 runs, 4 locations | Oct 7 to Oct 8 |
| 5 | UpCloud Premium 1 vCPU / 1 GB | 0.06% | 0 of 3 | 0.51% | 3 runs, 3 locations | Oct 7 to Oct 10 |
| 6 | DigitalOcean Basic AMD 1 vCPU / 1 GB | 0.03% | 0 of 4 | 0.18% | 4 runs, 4 locations | Oct 9 to Oct 12 |
| 7 | UpCloud Starter 1 vCPU / 1 GB | 0.02% | 0 of 6 | 0.12% | 6 runs, 4 locations | Oct 7 to Oct 8 |
| 8 | Vultr High Performance AMD 1 vCPU / 1 GB | 0.02% | 0 of 3 | 0.03% | 3 runs, 3 locations | Oct 8 to Oct 10 |
| 9 | Scaleway PRO2-XXS | 0.02% | 0 of 3 | 0.09% | 3 runs, 3 locations | Oct 8 |
| 10 | Linode Linode 4GB | 0.01% | 0 of 3 | 0.04% | 3 runs, 3 locations | Oct 10 to Oct 12 |
| 11 | Vultr Regular Performance Intel 1 vCPU / 1 GB | 0.0% | 0 of 6 | 0.04% | 6 runs, 4 locations | Oct 7 to Oct 8 |
Left out because the host never reported any steal, which usually means it isn't exposed to the server rather than that none happened: Hetzner CPX22.
How plans get on this list
- A plan needs at least 3 completed runs from the last 90 days, spread across at least 2 locations or 2 different days, so one bad host or one bad afternoon can't rank it.
- Plans are only compared with plans of the same CPU class. Shared CPU plans are never ranked against dedicated ones.
- Every completed run counts. We don't drop bad runs or rerun a plan until it looks better.
- Steal is measured once a second for the whole run, counting only the seconds where the server was asking for at least half its CPU. Steal on an idle server doesn't tell you much.
- A run only counts if it has at least 5 minutes of that kind of load.
- The worst minute is the worst 60-second average in a run. Single-second spikes are mostly noise, so they aren't used.