Skip to content

Public benchmarks

The measurements behind the product.

Build time, cache reuse, and storage answer different questions. Explore the published comparisons for the workloads and configurations that matter to you.

  • PostHog

  • Mastodon

  • gRPC

  • Immich

  • n8n

  • Hugo

Build time

Compare the same workload on the same runner class with different cache setups.

Docker without layer hits

2.9×

Faster Docker builds without layer hits

Across three Mastodon source changes on AMD64 and ARM64, ccache and persistent mounts averaged 2m 39s, compared with 7m 34s for cache:false.

Every Docker instruction ran · six paired builds.

Docker

11–60%

Faster Docker builds

BoringCache cut measured build time by 11–60% across 6 public Docker comparisons with actions/cache.

Bazel

28%

Faster Bazel builds than BuildBuddy

BoringCache finished the same-source Bazel benchmark in 260s, compared with 360s for BuildBuddy.

Cache reads and compiler reuse

Cache hits and read times reported by the build tools.

Cargo + sccache

95.8%

Compiler results reused

1,333 of 1,391 compiler requests hit the shared cache in a Deno build.

Cargo target cache also restored.

ccache

77%

Compiler results reused

447 of 578 compiler requests hit the shared cache in an OBS Studio build.

sccache · macOS ARM64

5 ms

Average sccache read hit

sccache recorded 1,287 cache hits through BoringCache in this macOS ARM64 build.

September 22, 2026 · sccache 0.17.0 · excludes proxy startup and cache misses.

Cache storage

70–94% less cache data stored across six projects compared with actions/cache.

Cache storage used by each project with BoringCache and actions/cache
Project BoringCache actions/cache Difference
Hugo Go 0.279 GiB 1.457 GiB
81% less
Storybook Nx 0.841 GiB 6.045 GiB
86% less
OpenTelemetry Java Gradle 0.758 GiB 3.511 GiB
78% less
Spring AI Maven 1.356 GiB 4.574 GiB
70% less
n8n-io/n8n Turbo 0.775 GiB 12.773 GiB
94% less
Zed sccache 0.882 GiB 11.46 GiB
92% less

Find out what your build can reuse.

Start with a relevant tool integration, then compare the complete job on your own workload.