Benchmarks

The method, published before the results

Olympia has not run a benchmark yet. Rather than say “coming soon” on every board, we have published the thing we can publish honestly today: the exact commands, the spec cohort, the rule for picking which plan to compare against, and the conditions under which we will refuse to print a ranking at all. It is dated. When results arrive you can check whether the method moved to suit them.

No results published

Three boards

vs hyperscalers & VPS hosts → AWS, Azure and Google Cloud, plus DigitalOcean and Hetzner — measured on a matched spec cohort and normalized by price. The board names which of the five are hyperscalers and which are not.
vs developer platforms → Vercel and Railway. A narrower board, because a platform sells an execution model rather than a machine — and its price column is a model, not a quote.
vs per-seat SaaS → Cost per seat against cost per server. Not a performance comparison, and not presented as one — it measures monthly cost at a stated seat count.

What the two hardware boards measure

Open-source tools only, run with published command lines, on three separately provisioned instances per plan. The full command list is on each board’s page.

Compute: sysbench cpu, 7z b, openssl speed
Storage: fio, random and sequential, --direct=1. Hyperscale board only — the developer-platforms board sells no raw block device, so the number would be fiction.
Network: iperf3, both directions, to a fixed independent endpoint. Hyperscale board only, for the same reason.
Application: pgbench, valkey-benchmark, oha / wrk

Three rules that apply to every board

Method before measurement

Nobody in this market publishes their methodology before their results, because publishing it afterwards lets you shape it around the outcome. Ours is dated and versioned. If it changes after a run, the change is visible.

Matched cohort, price-normalized

We compare the cheapest plan at each provider that meets or exceeds a fixed spec cohort. Not our tier against someone’s premium tier, and not against their entry tier either — that mismatch is how these tables get rigged. Then we divide by the price.

No hardware claim

Nothing on these pages asserts that Olympia designs, owns or operates the machines behind any plan. The comparison is between purchasable plans at their prices, which is a claim that stays true whoever owns the metal.

Comparisons that do exist today

Performance boards are empty, but the sourced price comparisons against Hetzner, AWS and others are real and live now.

Go to Compare →