Performance
Fast by default, not by accident.
The stack, the defaults, and the numbers — published in full. If your site loads slower on CubitRace than your current host, we will refund you.
Numbers we publish
The 90-day median across our fleet.
64ms
TTFB · median
EU origin, cached
1.1s
LCP · median
real users · 28d
94%
edge cache hit
across all sites
0.003%
5xx rate
edge + origin combined
The stack
What we actually run, and why each piece is there.
Caddy at the edge
HTTP/3, Brotli, on-demand TLS. Connection reuse for upstreams. Cert issuance handled in seconds.
NGINX + PHP-FPM 8.3
JIT enabled. opcache.preload pre-tuned per stack. Static assets served direct, never via PHP.
Per-site Redis
Object cache with eviction policy tuned per workload. Sessions in a separate Redis to avoid lock waits.
MySQL 8 / Postgres 16
innodb_buffer_pool sized to plan. Query cache disabled (because it should be). Slow query log on by default.
NVMe storage
Local NVMe with predictable latency. Backups stream to off-site object storage out-of-band.
Real-user monitoring
Beacon-based RUM, sampled by region. p50/p95/p99 per route. Built-in, no third party required.
Methodology
How we benchmark, and how you can reproduce it.
Same site, same probe, every host.
- • Identical site export (full DB + uploads + plugins) deployed to each host.
- • Probes from k6 in the same DC as the origin — no CDN tricks in our favor.
- • 30-minute warmup, then 1-hour load at 50 / 200 / 500 concurrent users.
- • We publish raw k6 JSON, not screenshots.
Want the raw runs? Email [email protected]. We will send them.
Get started
Try it on a real site, against your real traffic.
7-day free trial, free migration.