CI Bench

Head to head · CI Bench v1

Ubicloud vs Self-hosted runners on AWS

no clear difference in time between Ubicloud and Self-hosted runners on AWS (one VM per job) (the 95% interval, 0.94 to 1.05, includes 1.00); Ubicloud cost 55% less per job (at 100,000 jobs a month).

Self-hosted options spread fixed monthly costs over the monthly volume, so their cost per job changes with it; see the volume table.

Speed

× faster than GitHub-hosted (ubuntu-latest) · higher is better
Ubicloud
1.1×
Self-hosted runners on AWS (one VM per job)
1.2×
GitHub-hosted (ubuntu-latest)
1.0×

Cost per job

USD per $1.00 spent on GitHub-hosted (ubuntu-latest), at 100,000 jobs a month · lower is better
Ubicloud
$0.17
Self-hosted runners on AWS (one VM per job)
$0.38
GitHub-hosted (ubuntu-latest)
$1.00

Workload by workload

Ubicloud had the lowest median time of the two on 5 of 12 workloads, Self-hosted runners on AWS on 7.
WorkloadUbicloudSelf-hosted runners on AWS (one VM per job)Lowest median
Burst: 20 jobs at once203 s161 sSelf-hosted runners on AWS (one VM per job)
cli359 s357 sSelf-hosted runners on AWS (one VM per job)
codex-lint682 s512 sSelf-hosted runners on AWS (one VM per job)
container213 s247 sUbicloud
fd87 s114 sUbicloud
flask36 s70 sUbicloud
guava881 s717 sSelf-hosted runners on AWS (one VM per job)
monorepo276 s281 sUbicloud
pydantic159 s154 sSelf-hosted runners on AWS (one VM per job)
ts560 s433 sSelf-hosted runners on AWS (one VM per job)
vite502 s403 sSelf-hosted runners on AWS (one VM per job)
vue114 s117 sUbicloud

Cost by volume

Mean cost per job in USD, from the benchmark’s sensitivity results
VolumeUbicloudSelf-hosted runners on AWS (one VM per job)
100,000 jobs a month$0.173$0.383
10,000 jobs a month$0.173$0.395
1,000 jobs a month$0.173$0.508

Failed runs

Counted, never replaced
Ubicloud
1 of 360 measured runs (0.3%)
Self-hosted runners on AWS (one VM per job)
0 of 360 measured runs (0.0%)

Each configuration page lists where its failures happened. All losses.

All results for this pair

One line per configuration pairing, generated from the benchmark data