Skip to content

Public projects · recorded builds

Build caches tested as projects change.

Shared build cache on the runners you use. See BoringCache across projects and runner platforms, with measured restore, build and save times.

Projects
21
Build tools
15
Builds in sample
3,376

PostHog

Docker

Average restore, build and save · lower is better

GitHub runner

4 vCPU · 16 GB · Ubuntu 24.04 x86-64

Base · Image layers

BoringCache
15 min 2 s 30 builds
GHA Cache
28 min 22 s 30 builds
GHCR
20 min 58 s 30 builds

Plus · Image layers, tool caches and cache mounts

BoringCache +
18 min 1 s 30 builds
GHA Cache Dance
41 min 8 s 30 builds
  • PostHog
  • Mastodon
  • n8n
  • gRPC
  • Zed
  • Deno
  • OBS Studio
  • Immich

Time saved per build

The difference between BoringCache and another cache on the same commits and runner class. Runner cost applies the provider's listed price per minute to that difference.

Project Compared with Same CPU model Saved per build Per 1,000 builds Runner cost saved per 1,000 builds
PostHog Docker · GitHub runner, 4 vCPU GHA Cache 1.89× · faster on 30 of 30 builds 1.96× · 10 of 30 13 min 20 s 222 hours $160 at $0.012/min
Zed Rust · GitHub runner, 4 vCPU GHA Cache + target 1.28× · faster on 29 of 30 builds 1.21× · 6 of 30 8 min 33 s 142 hours $103 at $0.012/min
gRPC Bazel · GitHub runner, 4 vCPU BuildBuddy 1.49× · faster on 8 of 9 builds 1.33× · 1 of 9 83.2 s 23 hours $17 at $0.012/min
Hugo Go · GitHub runner, 4 vCPU Depot Cache 2.19× · faster on 18 of 18 builds 2.26× · 7 of 18 49.7 s 14 hours $9.94 at $0.012/min
Lichess sbt · GitHub runner, 4 vCPU GHA Cache 2.52× · faster on 9 of 9 builds 1.75× · 3 of 9 2 min 59 s 50 hours $36 at $0.012/min
Spring AI Maven · GitHub runner, 4 vCPU Depot Cache 1.40× · faster on 16 of 17 builds 1.30× · 7 of 17 95.3 s 26 hours $19 at $0.012/min
OBS Studio ccache · GitHub runner, 4 vCPU GHA Cache 1.11× · faster on 22 of 28 builds 1.28× · 11 of 28 11.8 s 3 hours $2.36 at $0.012/min

Smaller runner, faster image build

BoringCache on a runner with fewer vCPUs, compared with the provider's remote builder on the same commits.

PostHog · Depot

BoringCache on Depot runner8 vCPU · 32 GB
5 min 58 s
Depot remote builder16 vCPU · 32 GB
6 min 34 s

1.10× faster on 8 vCPU than the 16-vCPU remote builder · faster on 24 of 30 builds. Per 1,000 builds: $143 of runner minutes against $263 of remote build minutes.

PostHog · Namespace

BoringCache on Namespace runner4 vCPU · 16 GB
9 min 16 s
Namespace remote builder8 vCPU · 16 GiB
10 min 3 s

1.08× faster on 4 vCPU than the 8-vCPU remote builder · faster on 10 of 18 builds.

Runner cost uses each provider's listed price for the runner size measured. GitHub runs public repositories on the 4-vCPU size free; private repositories pay the larger-runner price for it. Prices from GitHub and Depot, checked 14 October 2026. Costs cover the measured restore, build and save time only. Plan fees, included minutes, storage, the BoringCache subscription and the rest of the job are not included. A remote builder job also occupies a CI runner while it waits.

Build times

Selected averages of restore, build and save time, using up to 30 matched commits per runner. Cold and warm builds are shown separately.

1.89×

Faster Docker builds

Immich vs GitHub Actions cache · 30 paired builds

See the project →

1.11×

Faster ccache builds

OBS Studio vs GitHub Actions cache · 28 paired builds

See the project →

97.0%

Median cache hit rate on new commits

Across 28 workloads. Each project's hit rate appears with its build times.

See the project →

2.7 ms

Median cache read time

Reported by sccache across 3 projects. Project averages: 2.0–3.4 ms. Hits can be served locally after prefetch.

See the project →

Docker

BoringCache caches: Image layers · Image layers, tool caches and cache mounts

Chroma

Docker

Embedding database. Rust and Python image.

Average restore, build and save · lower is better

GitHub runner

4 vCPU · 16 GB · Ubuntu 24.04 x86-64

Base · Image layers

BoringCache
45 min 28 s 5 builds
GHA Cache
50 min 5 s 5 builds
GHCR
47 min 21 s 5 builds

Plus · Image layers, tool caches and cache mounts

BoringCache +
31 min 22 s 5 builds
Same CPU model

Builds of one commit can land on different CPU models. These ratios use only the build pairs where both caches ran on the same model; above 1.00× means BoringCache's total time over those pairs was lower.

GHA Cache vs BoringCache
1.14× · 3 of 5
GHCR vs BoringCache
1.04× · 3 of 5
4 CPU models recorded
BoringCache · Image layers
AMD EPYC 9V74 (3 builds) · AMD EPYC 7763 (2 builds)
BoringCache + · Image layers, tool caches and cache mounts
AMD EPYC 7763 (2 builds) · Intel Xeon Platinum 8573C (1 build) · AMD EPYC 9V74 (1 build) · Intel Xeon Platinum 8370C (1 build)
GHA Cache · Buildx type=gha, mode=max
AMD EPYC 9V74 (3 builds) · AMD EPYC 7763 (2 builds)
GHCR · Buildx type=registry, mode=max
AMD EPYC 7763 (4 builds) · AMD EPYC 9V74 (1 build)
Cold and warm builds

Each table is one workflow run with one build per cache. A row marked Different CPU ran on another CPU model than BoringCache at the same level, so part of its time difference can come from the hardware.

One cold and warm build per cache
Cache Cold Warm CPU
Base · Image layers
BoringCache Image layers 1 h 19 min 23.9 s AMD EPYC 7763
Plus · Image layers, tool caches and cache mounts
BoringCache + Image layers, tool caches and cache mounts 1 h 13 min 30.9 s Cold: AMD EPYC 9V74 · Warm: AMD EPYC 7763
One cold and warm build per cache
Cache Cold Warm CPU
GHA Cache Buildx type=gha, mode=max 1 h 5 min 14.5 s Cold: AMD EPYC 9V45 · Warm: AMD EPYC 7763
GHCR Buildx type=registry, mode=max 1 h 29 min 13.4 s Cold: AMD EPYC 7763 · Warm: AMD EPYC 9V74
GitHub Arm runner

4 vCPU · 16 GB · Ubuntu 24.04 Arm64

Base · Image layers

BoringCache
33 min 7 s 5 builds
GHA Cache
41 min 37 s 5 builds
GHCR
34 min 36 s 5 builds

Plus · Image layers, tool caches and cache mounts

BoringCache +
23 min 32 s 5 builds

Neoverse-N2

Cold and warm builds

Each table is one workflow run with one build per cache. A row marked Different CPU ran on another CPU model than BoringCache at the same level, so part of its time difference can come from the hardware.

One cold and warm build per cache
Cache Cold Warm CPU
Base · Image layers
BoringCache Image layers 56 min 19 s 17.8 s Neoverse-N2
Plus · Image layers, tool caches and cache mounts
BoringCache + Image layers, tool caches and cache mounts 55 min 36 s 38.2 s Neoverse-N2
One cold and warm build per cache
Cache Cold Warm CPU
GHA Cache Buildx type=gha, mode=max 1 h 20 min 12.4 s Neoverse-N2
GHCR Buildx type=registry, mode=max 59 min 49 s 13.0 s Neoverse-N2
Cache hits

BoringCache · GitHub runner · 4 vCPU

95.2% of 187 lookups over 5 builds

Stored cache
BoringCacheImage-layer cache tag; bytes referenced by 1 current tag
6.7 GB
BoringCacheImage-layer and cache-mount tags; bytes referenced by 4 current tags
11.3 GB

Discourse

Docker

Discussion platform. Base images built with docker buildx bake.

Average restore, build and save · lower is better

GitHub runner

4 vCPU · 16 GB · Ubuntu 24.04 x86-64

BoringCache
11 min 36 s 1 build
GHA Cache
15 min 48 s 1 build
GHCR
15 min 20 s 1 build

AMD EPYC 9V74

Cold and warm builds

Each table is one workflow run with one build per cache. A row marked Different CPU ran on another CPU model than BoringCache at the same level, so part of its time difference can come from the hardware.

One cold and warm build per cache
Cache Cold Warm CPU
BoringCache Image layers 15 min 49 s 11 min 4 s AMD EPYC 7763
GHA Cache Bake type=gha, mode=max 23 min 26 s 13 min 1 s AMD EPYC 7763
GHCR Bake type=registry, mode=max 17 min 37 s 13 min 19 s AMD EPYC 9V74 Different CPU
GitHub Arm runner

4 vCPU · 16 GB · Ubuntu 24.04 Arm64

BoringCache
8 min 57 s 1 build
GHA Cache
11 min 43 s 1 build
GHCR
10 min 19 s 1 build

Neoverse-N2

Cold and warm builds

Each table is one workflow run with one build per cache. A row marked Different CPU ran on another CPU model than BoringCache at the same level, so part of its time difference can come from the hardware.

One cold and warm build per cache
Cache Cold Warm CPU
BoringCache Image layers 13 min 16 s 9 min 27 s Neoverse-N2
GHA Cache Bake type=gha, mode=max 15 min 50 s 9 min 39 s Neoverse-N2
GHCR Bake type=registry, mode=max 14 min 36 s 10 min 35 s Neoverse-N2
Cache hits

BoringCache · GitHub runner · 4 vCPU

94.4% of 108 lookups over 1 build

Hugo

Docker

Static site generator written in Go.

Average restore, build and save · lower is better

GitHub runner

4 vCPU · 16 GB · Ubuntu 24.04 x86-64

Base · Image layers

BoringCache
3 min 2 s 15 builds
GHA Cache
3 min 17 s 15 builds
GHCR
2 min 59 s 15 builds

Plus · Image layers, tool caches and cache mounts

BoringCache +
1 min 41 s 15 builds
GHA Cache Dance
5 min 30 s 15 builds
Same CPU model

Builds of one commit can land on different CPU models. These ratios use only the build pairs where both caches ran on the same model; above 1.00× means BoringCache's total time over those pairs was lower.

GHA Cache vs BoringCache
1.13× · 3 of 15
GHCR vs BoringCache
0.99× · 4 of 15
GHA Cache Dance vs BoringCache +
3.55× · 6 of 15
6 CPU models recorded
BoringCache · Image layers
AMD EPYC 7763 (5 builds) · AMD EPYC 9V74 (4 builds) · AMD EPYC 9V45 (4 builds) · Intel Xeon 6973P-C (1 build) · Intel Xeon Platinum 8573C (1 build)
BoringCache + · Image layers, tool caches and cache mounts
AMD EPYC 7763 (7 builds) · AMD EPYC 9V74 (5 builds) · AMD EPYC 9V45 (1 build) · Intel Xeon Platinum 8370C (1 build) · Intel Xeon Platinum 8573C (1 build)
GHA Cache · Buildx type=gha, mode=max
AMD EPYC 9V74 (7 builds) · Intel Xeon Platinum 8573C (3 builds) · AMD EPYC 9V45 (2 builds) · AMD EPYC 7763 (2 builds) · Intel Xeon Platinum 8370C (1 build)
GHCR · Buildx type=registry, mode=max
AMD EPYC 7763 (5 builds) · AMD EPYC 9V45 (4 builds) · AMD EPYC 9V74 (4 builds) · Intel Xeon Platinum 8573C (1 build) · Intel Xeon Platinum 8370C (1 build)
GHA Cache Dance · type=gha, mode=max; cache mounts saved by actions/cache
AMD EPYC 9V74 (6 builds) · AMD EPYC 7763 (6 builds) · AMD EPYC 9V45 (3 builds)
Cold and warm builds

Each table is one workflow run with one build per cache. A row marked Different CPU ran on another CPU model than BoringCache at the same level, so part of its time difference can come from the hardware.

One cold and warm build per cache
Cache Cold Warm CPU
Base · Image layers
BoringCache Image layers 3 min 43 s 18.9 s AMD EPYC 7763
GHA Cache Buildx type=gha, mode=max 6 min 28 s 20.1 s AMD EPYC 7763
GHCR Buildx type=registry, mode=max 4 min 2 s 11.9 s AMD EPYC 7763
Plus · Image layers, tool caches and cache mounts
BoringCache + Image layers, tool caches and cache mounts 4 min 2 s 22.0 s AMD EPYC 7763
GitHub Arm runner

4 vCPU · 16 GB · Ubuntu 24.04 Arm64

Base · Image layers

BoringCache
2 min 19 s 15 builds
GHA Cache
2 min 32 s 15 builds
GHCR
2 min 23 s 15 builds

Plus · Image layers, tool caches and cache mounts

BoringCache +
86.1 s 15 builds
GHA Cache Dance
4 min 38 s 15 builds

Neoverse-N2

Cold and warm builds

Each table is one workflow run with one build per cache. A row marked Different CPU ran on another CPU model than BoringCache at the same level, so part of its time difference can come from the hardware.

One cold and warm build per cache
Cache Cold Warm CPU
Base · Image layers
BoringCache Image layers 2 min 38 s 19.7 s Neoverse-N2
GHA Cache Buildx type=gha, mode=max 3 min 29 s 14.4 s Neoverse-N2
GHCR Buildx type=registry, mode=max 2 min 31 s 9.7 s Neoverse-N2
Plus · Image layers, tool caches and cache mounts
BoringCache + Image layers, tool caches and cache mounts 3 min 14 s 17.8 s Neoverse-N2
Cache hits

BoringCache · GitHub runner · 4 vCPU

96.6% of 436 lookups over 15 builds

Stored cache
BoringCacheImage-layer cache tag; bytes referenced by 1 current tag
393 MB
BoringCacheImage-layer and cache-mount tags; bytes referenced by 3 current tags
1.79 GB

Immich

Docker

Photo and video management server. NestJS server image.

Average restore, build and save · lower is better

GitHub runner

4 vCPU · 16 GB · Ubuntu 24.04 x86-64

Base · Image layers

BoringCache
2 min 9 s 30 builds
GHA Cache
4 min 4 s 30 builds
GHCR
2 min 46 s 30 builds

Plus · Image layers, tool caches and cache mounts

BoringCache +
2 min 16 s 30 builds
GHA Cache Dance
7 min 0 s 30 builds
Same CPU model

Builds of one commit can land on different CPU models. These ratios use only the build pairs where both caches ran on the same model; above 1.00× means BoringCache's total time over those pairs was lower.

GHA Cache vs BoringCache
1.96× · 8 of 30
GHCR vs BoringCache
1.36× · 10 of 30
GHA Cache Dance vs BoringCache +
3.06× · 14 of 30
6 CPU models recorded
BoringCache · Image layers
AMD EPYC 7763 (12 builds) · AMD EPYC 9V45 (8 builds) · AMD EPYC 9V74 (7 builds) · Intel Xeon Platinum 8370C (1 build) · Intel Xeon Platinum 8573C (1 build) · Intel Xeon 6973P-C (1 build)
BoringCache + · Image layers, tool caches and cache mounts
AMD EPYC 7763 (16 builds) · AMD EPYC 9V45 (6 builds) · AMD EPYC 9V74 (5 builds) · Intel Xeon Platinum 8573C (3 builds)
GHA Cache · Buildx type=gha, mode=max
AMD EPYC 7763 (14 builds) · AMD EPYC 9V45 (7 builds) · AMD EPYC 9V74 (4 builds) · Intel Xeon Platinum 8573C (3 builds) · Intel Xeon 6973P-C (1 build) · Intel Xeon Platinum 8370C (1 build)
GHCR · Buildx type=registry, mode=max
AMD EPYC 7763 (14 builds) · AMD EPYC 9V45 (7 builds) · AMD EPYC 9V74 (4 builds) · Intel Xeon Platinum 8370C (2 builds) · Intel Xeon Platinum 8573C (2 builds) · Intel Xeon 6973P-C (1 build)
GHA Cache Dance · type=gha, mode=max; cache mounts saved by actions/cache
AMD EPYC 7763 (21 builds) · AMD EPYC 9V45 (5 builds) · AMD EPYC 9V74 (4 builds)
Cold and warm builds

Each table is one workflow run with one build per cache. A row marked Different CPU ran on another CPU model than BoringCache at the same level, so part of its time difference can come from the hardware.

One cold and warm build per cache
Cache Cold Warm CPU
Base · Image layers
BoringCache Image layers 3 min 33 s 34.6 s Cold: Intel Xeon Platinum 8573C · Warm: Intel Xeon Platinum 8370C
GHA Cache Buildx type=gha, mode=max 10 min 2 s 56.6 s Cold: AMD EPYC 9V45 · Warm: Intel Xeon Platinum 8573C Different CPU
GHCR Buildx type=registry, mode=max 4 min 24 s 40.8 s Intel Xeon Platinum 8573C Different CPU
Plus · Image layers, tool caches and cache mounts
BoringCache + Image layers, tool caches and cache mounts 3 min 54 s 37.0 s Cold: AMD EPYC 9V74 · Warm: AMD EPYC 7763
GitHub Arm runner

4 vCPU · 16 GB · Ubuntu 24.04 Arm64

Base · Image layers

BoringCache
1 min 53 s 30 builds
GHA Cache
3 min 9 s 30 builds
GHCR
2 min 13 s 30 builds

Plus · Image layers, tool caches and cache mounts

BoringCache +
1 min 57 s 30 builds
GHA Cache Dance
6 min 39 s 30 builds

Neoverse-N2

Cold and warm builds

Each table is one workflow run with one build per cache. A row marked Different CPU ran on another CPU model than BoringCache at the same level, so part of its time difference can come from the hardware.

One cold and warm build per cache
Cache Cold Warm CPU
Base · Image layers
BoringCache Image layers 2 min 41 s 30.8 s Neoverse-N2
GHA Cache Buildx type=gha, mode=max 8 min 57 s 26.6 s Neoverse-N2
GHCR Buildx type=registry, mode=max 3 min 16 s 28.1 s Neoverse-N2
Plus · Image layers, tool caches and cache mounts
BoringCache + Image layers, tool caches and cache mounts 3 min 11 s 29.3 s Neoverse-N2
Cache hits

BoringCache · GitHub runner · 4 vCPU

98.9% of 2,760 lookups over 30 builds

Stored cache
BoringCacheImage-layer cache tag; bytes referenced by 1 current tag
1.64 GB
BoringCacheImage-layer and cache-mount tags; bytes referenced by 3 current tags
2.39 GB

llama.cpp

Docker

LLM inference in C and C++. Server image compiled from source.

Average restore, build and save · lower is better

GitHub runner

4 vCPU · 16 GB · Ubuntu 24.04 x86-64

Base · Image layers

BoringCache
1 h 20 min 14 builds
GHA Cache
1 h 23 min 14 builds
GHCR
1 h 24 min 14 builds

Plus · Image layers, tool caches and cache mounts

BoringCache +
12 min 56 s 14 builds
GHA Cache Dance
20 min 12 s 14 builds
Same CPU model

Builds of one commit can land on different CPU models. These ratios use only the build pairs where both caches ran on the same model; above 1.00× means BoringCache's total time over those pairs was lower.

GHA Cache vs BoringCache
1.04× · 2 of 14
GHCR vs BoringCache
0.92× · 4 of 14
GHA Cache Dance vs BoringCache +
1.71× · 2 of 14
6 CPU models recorded
BoringCache · Image layers
AMD EPYC 7763 (4 builds) · AMD EPYC 9V74 (3 builds) · AMD EPYC 9V45 (3 builds) · Intel Xeon Platinum 8573C (2 builds) · Intel Xeon Platinum 8370C (2 builds)
BoringCache + · Image layers, tool caches and cache mounts
AMD EPYC 9V45 (5 builds) · AMD EPYC 7763 (5 builds) · AMD EPYC 9V74 (3 builds) · Intel Xeon Platinum 8573C (1 build)
GHA Cache · Buildx type=gha, mode=max
AMD EPYC 7763 (7 builds) · Intel Xeon 6973P-C (3 builds) · AMD EPYC 9V74 (3 builds) · Intel Xeon Platinum 8370C (1 build)
GHCR · Buildx type=registry, mode=max
AMD EPYC 7763 (8 builds) · AMD EPYC 9V74 (3 builds) · Intel Xeon Platinum 8573C (2 builds) · AMD EPYC 9V45 (1 build)
GHA Cache Dance · type=gha, mode=max; cache mounts saved by actions/cache
AMD EPYC 7763 (5 builds) · AMD EPYC 9V45 (3 builds) · AMD EPYC 9V74 (3 builds) · Intel Xeon 6973P-C (2 builds) · Intel Xeon Platinum 8573C (1 build)
Cache hits

BoringCache · GitHub runner · 4 vCPU

97.7% of 768 lookups over 14 builds

Stored cache
BoringCacheImage-layer cache tag; bytes referenced by 1 current tag
6.84 GB
BoringCacheImage-layer and cache-mount tags; bytes referenced by 2 current tags
6.84 GB

Mastodon

Docker

Social network server. Ruby on Rails image with compiled assets.

Average restore, build and save · lower is better

GitHub runner

4 vCPU · 16 GB · Ubuntu 24.04 x86-64

Base · Image layers

BoringCache
2 min 2 s 30 builds
GHA Cache
3 min 8 s 30 builds
GHCR
2 min 27 s 30 builds

Plus · Image layers, tool caches and cache mounts

BoringCache +
2 min 6 s 30 builds
GHA Cache Dance
4 min 0 s 30 builds
Same CPU model

Builds of one commit can land on different CPU models. These ratios use only the build pairs where both caches ran on the same model; above 1.00× means BoringCache's total time over those pairs was lower.

GHA Cache vs BoringCache
1.48× · 8 of 30
GHCR vs BoringCache
1.22× · 7 of 30
GHA Cache Dance vs BoringCache +
2.04× · 10 of 30
6 CPU models recorded
BoringCache · Image layers
AMD EPYC 7763 (12 builds) · AMD EPYC 9V74 (5 builds) · AMD EPYC 9V45 (5 builds) · Intel Xeon 6973P-C (4 builds) · Intel Xeon Platinum 8573C (4 builds)
BoringCache + · Image layers, tool caches and cache mounts
AMD EPYC 7763 (17 builds) · AMD EPYC 9V74 (4 builds) · Intel Xeon Platinum 8573C (3 builds) · Intel Xeon Platinum 8370C (2 builds) · Intel Xeon 6973P-C (2 builds) · AMD EPYC 9V45 (2 builds)
GHA Cache · Buildx type=gha, mode=max
AMD EPYC 7763 (14 builds) · AMD EPYC 9V74 (7 builds) · AMD EPYC 9V45 (5 builds) · Intel Xeon Platinum 8573C (3 builds) · Intel Xeon Platinum 8370C (1 build)
GHCR · Buildx type=registry, mode=max
AMD EPYC 7763 (15 builds) · AMD EPYC 9V74 (6 builds) · AMD EPYC 9V45 (4 builds) · Intel Xeon Platinum 8573C (3 builds) · Intel Xeon Platinum 8370C (1 build) · Intel Xeon 6973P-C (1 build)
GHA Cache Dance · type=gha, mode=max; cache mounts saved by actions/cache
AMD EPYC 7763 (17 builds) · AMD EPYC 9V45 (6 builds) · AMD EPYC 9V74 (5 builds) · Intel Xeon Platinum 8573C (2 builds)
Cold and warm builds

Each table is one workflow run with one build per cache. A row marked Different CPU ran on another CPU model than BoringCache at the same level, so part of its time difference can come from the hardware.

One cold and warm build per cache
Cache Cold Warm CPU
Base · Image layers
BoringCache Image layers 9 min 0 s 45.5 s Cold: AMD EPYC 7763 · Warm: AMD EPYC 9V74
GHA Cache Buildx type=gha, mode=max 8 min 52 s 40.0 s Cold: AMD EPYC 9V74 · Warm: AMD EPYC 7763 Different CPU
GHCR Buildx type=registry, mode=max 9 min 7 s 28.5 s Cold: Intel Xeon Platinum 8573C · Warm: AMD EPYC 7763 Different CPU
Plus · Image layers, tool caches and cache mounts
BoringCache + Image layers, tool caches and cache mounts 9 min 57 s 45.6 s Cold: AMD EPYC 7763 · Warm: AMD EPYC 9V74
GitHub Arm runner

4 vCPU · 16 GB · Ubuntu 24.04 Arm64

Base · Image layers

BoringCache
1 min 49 s 30 builds
GHA Cache
2 min 42 s 30 builds
GHCR
2 min 7 s 30 builds

Plus · Image layers, tool caches and cache mounts

BoringCache +
1 min 53 s 30 builds
GHA Cache Dance
3 min 36 s 30 builds

Neoverse-N2

Cold and warm builds

Each table is one workflow run with one build per cache. A row marked Different CPU ran on another CPU model than BoringCache at the same level, so part of its time difference can come from the hardware.

One cold and warm build per cache
Cache Cold Warm CPU
Base · Image layers
BoringCache Image layers 6 min 19 s 32.7 s Neoverse-N2
GHA Cache Buildx type=gha, mode=max 8 min 5 s 34.9 s Neoverse-N2
GHCR Buildx type=registry, mode=max 6 min 45 s 28.9 s Neoverse-N2
Plus · Image layers, tool caches and cache mounts
BoringCache + Image layers, tool caches and cache mounts 6 min 56 s 39.0 s Neoverse-N2
Cache hits

BoringCache · GitHub runner · 4 vCPU

98.2% of 1,677 lookups over 30 builds

Stored cache
BoringCacheImage-layer cache tag; bytes referenced by 1 current tag
1.05 GB
BoringCacheImage-layer and cache-mount tags; bytes referenced by 7 current tags
1.45 GB

n8n

Docker

Workflow automation platform. Node.js monorepo with a pnpm workspace.

Average restore, build and save · lower is better

GitHub runner

4 vCPU · 16 GB · Ubuntu 24.04 x86-64

BoringCache
2 min 11 s 30 builds
GHA Cache
2 min 41 s 30 builds
GHCR
2 min 37 s 30 builds
Same CPU model

Builds of one commit can land on different CPU models. These ratios use only the build pairs where both caches ran on the same model; above 1.00× means BoringCache's total time over those pairs was lower.

GHA Cache vs BoringCache
1.19× · 12 of 30
GHCR vs BoringCache
1.21× · 10 of 30
5 CPU models recorded
BoringCache · Image layers
AMD EPYC 7763 (15 builds) · AMD EPYC 9V45 (5 builds) · AMD EPYC 9V74 (5 builds) · Intel Xeon Platinum 8573C (3 builds) · Intel Xeon 6973P-C (2 builds)
GHA Cache · Buildx type=gha, mode=max
AMD EPYC 7763 (19 builds) · AMD EPYC 9V45 (6 builds) · AMD EPYC 9V74 (3 builds) · Intel Xeon Platinum 8573C (2 builds)
GHCR · Buildx type=registry, mode=max
AMD EPYC 7763 (18 builds) · AMD EPYC 9V74 (7 builds) · AMD EPYC 9V45 (3 builds) · Intel Xeon Platinum 8573C (2 builds)
Cold and warm builds

Each table is one workflow run with one build per cache. A row marked Different CPU ran on another CPU model than BoringCache at the same level, so part of its time difference can come from the hardware.

One cold and warm build per cache
Cache Cold Warm CPU
BoringCache Image layers 5 min 23 s 2 min 9 s Cold: AMD EPYC 7763 · Warm: AMD EPYC 9V45
GHA Cache Buildx type=gha, mode=max 6 min 54 s 2 min 17 s Cold: AMD EPYC 7763 · Warm: AMD EPYC 9V74 Different CPU
GHCR Buildx type=registry, mode=max 5 min 43 s 2 min 34 s AMD EPYC 7763 Different CPU
GitHub Arm runner

4 vCPU · 16 GB · Ubuntu 24.04 Arm64

BoringCache
1 min 48 s 30 builds
GHA Cache
2 min 18 s 30 builds
GHCR
2 min 7 s 30 builds

Neoverse-N2

Cold and warm builds

Each table is one workflow run with one build per cache. A row marked Different CPU ran on another CPU model than BoringCache at the same level, so part of its time difference can come from the hardware.

One cold and warm build per cache
Cache Cold Warm CPU
BoringCache Image layers 4 min 5 s 2 min 2 s Neoverse-N2
GHA Cache Buildx type=gha, mode=max 8 min 31 s 1 min 53 s Neoverse-N2
GHCR Buildx type=registry, mode=max 4 min 49 s 1 min 55 s Neoverse-N2
Cache hits

BoringCache · GitHub runner · 4 vCPU

98.4% of 1,830 lookups over 30 builds

Stored cache
BoringCacheImage-layer cache tag; bytes referenced by 1 current tag
470 MB

n8n distroless runners

Docker

n8n's task runner image on a distroless base.

Average restore, build and save · lower is better

GitHub runner

4 vCPU · 16 GB · Ubuntu 24.04 x86-64

BoringCache
91.0 s 30 builds
GHA Cache
2 min 20 s 30 builds
GHCR
1 min 56 s 30 builds
Same CPU model

Builds of one commit can land on different CPU models. These ratios use only the build pairs where both caches ran on the same model; above 1.00× means BoringCache's total time over those pairs was lower.

GHA Cache vs BoringCache
1.58× · 12 of 30
GHCR vs BoringCache
1.32× · 7 of 30
6 CPU models recorded
BoringCache · Image layers
AMD EPYC 7763 (11 builds) · AMD EPYC 9V74 (11 builds) · AMD EPYC 9V45 (5 builds) · Intel Xeon Platinum 8573C (3 builds)
GHA Cache · Buildx type=gha, mode=max
AMD EPYC 7763 (19 builds) · AMD EPYC 9V45 (5 builds) · AMD EPYC 9V74 (3 builds) · Intel Xeon Platinum 8573C (2 builds) · Intel Xeon Platinum 8370C (1 build)
GHCR · Buildx type=registry, mode=max
AMD EPYC 7763 (15 builds) · AMD EPYC 9V45 (6 builds) · AMD EPYC 9V74 (3 builds) · Intel Xeon 6973P-C (2 builds) · Intel Xeon Platinum 8370C (2 builds) · Intel Xeon Platinum 8573C (2 builds)
Cold and warm builds

Each table is one workflow run with one build per cache. A row marked Different CPU ran on another CPU model than BoringCache at the same level, so part of its time difference can come from the hardware.

One cold and warm build per cache
Cache Cold Warm CPU
BoringCache Image layers 79.5 s 97.4 s AMD EPYC 7763
GHA Cache Buildx type=gha, mode=max 5 min 32 s 93.2 s AMD EPYC 7763
GHCR Buildx type=registry, mode=max 2 min 6 s 91.2 s Cold: AMD EPYC 9V74 · Warm: AMD EPYC 7763 Different CPU
GitHub Arm runner

4 vCPU · 16 GB · Ubuntu 24.04 Arm64

BoringCache
77.2 s 30 builds
GHA Cache
2 min 3 s 30 builds
GHCR
1 min 41 s 30 builds

Neoverse-N2

Cold and warm builds

Each table is one workflow run with one build per cache. A row marked Different CPU ran on another CPU model than BoringCache at the same level, so part of its time difference can come from the hardware.

One cold and warm build per cache
Cache Cold Warm CPU
BoringCache Image layers 70.0 s 83.7 s Neoverse-N2
GHA Cache Buildx type=gha, mode=max 5 min 18 s 79.1 s Neoverse-N2
GHCR Buildx type=registry, mode=max 98.4 s 75.2 s Neoverse-N2
Cache hits

BoringCache · GitHub runner · 4 vCPU

98.6% of 2,160 lookups over 30 builds

Stored cache
BoringCacheImage-layer cache tag; bytes referenced by 1 current tag
552 MB

n8n task runners

Docker

n8n's JavaScript and Python task runner image.

Average restore, build and save · lower is better

GitHub runner

4 vCPU · 16 GB · Ubuntu 24.04 x86-64

BoringCache
59.8 s 30 builds
GHA Cache
88.2 s 30 builds
GHCR
73.2 s 30 builds
Same CPU model

Builds of one commit can land on different CPU models. These ratios use only the build pairs where both caches ran on the same model; above 1.00× means BoringCache's total time over those pairs was lower.

GHA Cache vs BoringCache
1.48× · 5 of 30
GHCR vs BoringCache
1.25× · 14 of 30
6 CPU models recorded
BoringCache · Image layers
AMD EPYC 7763 (17 builds) · AMD EPYC 9V74 (5 builds) · AMD EPYC 9V45 (3 builds) · Intel Xeon 6973P-C (2 builds) · Intel Xeon Platinum 8573C (2 builds) · Intel Xeon Platinum 8370C (1 build)
GHA Cache · Buildx type=gha, mode=max
AMD EPYC 7763 (12 builds) · AMD EPYC 9V74 (9 builds) · AMD EPYC 9V45 (5 builds) · Intel Xeon Platinum 8370C (2 builds) · Intel Xeon Platinum 8573C (2 builds)
GHCR · Buildx type=registry, mode=max
AMD EPYC 7763 (20 builds) · AMD EPYC 9V74 (4 builds) · Intel Xeon Platinum 8573C (3 builds) · Intel Xeon Platinum 8370C (2 builds) · AMD EPYC 9V45 (1 build)
Cold and warm builds

Each table is one workflow run with one build per cache. A row marked Different CPU ran on another CPU model than BoringCache at the same level, so part of its time difference can come from the hardware.

One cold and warm build per cache
Cache Cold Warm CPU
BoringCache Image layers 62.3 s 62.4 s Cold: Intel Xeon Platinum 8573C · Warm: AMD EPYC 7763
GHA Cache Buildx type=gha, mode=max 5 min 45 s 51.8 s Cold: AMD EPYC 7763 · Warm: AMD EPYC 9V45 Different CPU
GHCR Buildx type=registry, mode=max 90.5 s 50.9 s Cold: AMD EPYC 9V74 · Warm: AMD EPYC 9V45 Different CPU
GitHub Arm runner

4 vCPU · 16 GB · Ubuntu 24.04 Arm64

BoringCache
49.8 s 30 builds
GHA Cache
73.5 s 30 builds
GHCR
63.1 s 30 builds

Neoverse-N2

Cold and warm builds

Each table is one workflow run with one build per cache. A row marked Different CPU ran on another CPU model than BoringCache at the same level, so part of its time difference can come from the hardware.

One cold and warm build per cache
Cache Cold Warm CPU
BoringCache Image layers 48.7 s 49.6 s Neoverse-N2
GHA Cache Buildx type=gha, mode=max 1 min 43 s 47.2 s Neoverse-N2
GHCR Buildx type=registry, mode=max 63.9 s 51.8 s Neoverse-N2
Cache hits

BoringCache · GitHub runner · 4 vCPU

97.9% of 1,410 lookups over 30 builds

Stored cache
BoringCacheImage-layer cache tag; bytes referenced by 1 current tag
395 MB

PostHog

Docker

Product analytics platform. One Docker image with the Django backend and Node.js frontend.

Average restore, build and save · lower is better

GitHub runner

4 vCPU · 16 GB · Ubuntu 24.04 x86-64

Base · Image layers

BoringCache
15 min 2 s 30 builds
GHA Cache
28 min 22 s 30 builds
GHCR
20 min 58 s 30 builds

Plus · Image layers, tool caches and cache mounts

BoringCache +
18 min 1 s 30 builds
GHA Cache Dance
41 min 8 s 30 builds
Same CPU model

Builds of one commit can land on different CPU models. These ratios use only the build pairs where both caches ran on the same model; above 1.00× means BoringCache's total time over those pairs was lower.

GHA Cache vs BoringCache
1.96× · 10 of 30
GHCR vs BoringCache
1.39× · 7 of 30
GHA Cache Dance vs BoringCache +
2.32× · 11 of 30
6 CPU models recorded
BoringCache · Image layers
AMD EPYC 7763 (16 builds) · AMD EPYC 9V74 (10 builds) · AMD EPYC 9V45 (3 builds) · Intel Xeon Platinum 8370C (1 build)
BoringCache + · Image layers, tool caches and cache mounts
AMD EPYC 7763 (12 builds) · AMD EPYC 9V45 (8 builds) · AMD EPYC 9V74 (7 builds) · Intel Xeon Platinum 8573C (2 builds) · Intel Xeon Platinum 8370C (1 build)
GHA Cache · Buildx type=gha, mode=max
AMD EPYC 7763 (15 builds) · AMD EPYC 9V74 (8 builds) · AMD EPYC 9V45 (4 builds) · Intel Xeon Platinum 8573C (2 builds) · Intel Xeon 6973P-C (1 build)
GHCR · Buildx type=registry, mode=max
AMD EPYC 9V74 (13 builds) · AMD EPYC 7763 (9 builds) · AMD EPYC 9V45 (5 builds) · Intel Xeon Platinum 8370C (2 builds) · Intel Xeon Platinum 8573C (1 build)
GHA Cache Dance · type=gha, mode=max; cache mounts saved by actions/cache
AMD EPYC 7763 (18 builds) · AMD EPYC 9V74 (5 builds) · AMD EPYC 9V45 (2 builds) · Intel Xeon Platinum 8370C (2 builds) · Intel Xeon Platinum 8573C (2 builds) · Intel Xeon 6973P-C (1 build)
Cold and warm builds

Each table is one workflow run with one build per cache. A row marked Different CPU ran on another CPU model than BoringCache at the same level, so part of its time difference can come from the hardware.

One cold and warm build per cache
Cache Cold Warm CPU
Base · Image layers
BoringCache Image layers 15 min 28 s 2 min 13 s Cold: AMD EPYC 7763 · Warm: AMD EPYC 9V74
Plus · Image layers, tool caches and cache mounts
BoringCache + Image layers, tool caches and cache mounts 16 min 29 s 3 min 4 s Cold: AMD EPYC 7763 · Warm: Intel Xeon Platinum 8573C
One cold and warm build per cache
Cache Cold Warm CPU
GHA Cache Buildx type=gha, mode=max 39 min 59 s 3 min 6 s Cold: AMD EPYC 7763 · Warm: AMD EPYC 9V74
GHCR Buildx type=registry, mode=max 23 min 24 s 2 min 44 s AMD EPYC 7763
GitHub Arm runner

4 vCPU · 16 GB · Ubuntu 24.04 Arm64

Base · Image layers

BoringCache
11 min 57 s 30 builds
GHA Cache
21 min 19 s 30 builds
GHCR
16 min 10 s 30 builds

Plus · Image layers, tool caches and cache mounts

BoringCache +
15 min 22 s 30 builds
GHA Cache Dance
34 min 13 s 30 builds

Neoverse-N2

Cold and warm builds

Each table is one workflow run with one build per cache. A row marked Different CPU ran on another CPU model than BoringCache at the same level, so part of its time difference can come from the hardware.

One cold and warm build per cache
Cache Cold Warm CPU
Base · Image layers
BoringCache Image layers 12 min 51 s 2 min 20 s Neoverse-N2
Plus · Image layers, tool caches and cache mounts
BoringCache + Image layers, tool caches and cache mounts 12 min 32 s 2 min 18 s Neoverse-N2
One cold and warm build per cache
Cache Cold Warm CPU
GHA Cache Buildx type=gha, mode=max 26 min 41 s 2 min 39 s Neoverse-N2
GHCR Buildx type=registry, mode=max 16 min 37 s 2 min 15 s Neoverse-N2
Cache hits

BoringCache · GitHub runner · 4 vCPU

98.5% of 2,486 lookups over 30 builds

Stored cache
BoringCacheImage-layer cache tag; bytes referenced by 1 current tag
5.97 GB
BoringCacheImage-layer and cache-mount tags; bytes referenced by 5 current tags
10.8 GB

Qdrant

Docker

Vector database written in Rust. Release image built from source.

Average restore, build and save · lower is better

GitHub runner

4 vCPU · 16 GB · Ubuntu 24.04 x86-64

BoringCache
4 min 28 s 30 builds
GHA Cache
6 min 23 s 30 builds
GHCR
5 min 59 s 30 builds
Same CPU model

Builds of one commit can land on different CPU models. These ratios use only the build pairs where both caches ran on the same model; above 1.00× means BoringCache's total time over those pairs was lower.

GHA Cache vs BoringCache
1.41× · 7 of 30
GHCR vs BoringCache
1.20× · 9 of 30
6 CPU models recorded
BoringCache · Image layers
AMD EPYC 7763 (14 builds) · AMD EPYC 9V45 (5 builds) · AMD EPYC 9V74 (5 builds) · Intel Xeon Platinum 8573C (4 builds) · Intel Xeon 6973P-C (2 builds)
GHA Cache · Buildx type=gha, mode=max
AMD EPYC 7763 (10 builds) · AMD EPYC 9V74 (8 builds) · AMD EPYC 9V45 (5 builds) · Intel Xeon Platinum 8573C (3 builds) · Intel Xeon Platinum 8370C (3 builds) · Intel Xeon 6973P-C (1 build)
GHCR · Buildx type=registry, mode=max
AMD EPYC 7763 (13 builds) · AMD EPYC 9V74 (12 builds) · Intel Xeon Platinum 8573C (4 builds) · AMD EPYC 9V45 (1 build)
Cold and warm builds

Each table is one workflow run with one build per cache. A row marked Different CPU ran on another CPU model than BoringCache at the same level, so part of its time difference can come from the hardware.

One cold and warm build per cache
Cache Cold Warm CPU
BoringCache Image layers 9 min 39 s 17.7 s Cold: Intel Xeon Platinum 8573C · Warm: Intel Xeon Platinum 8370C
One cold and warm build per cache
Cache Cold Warm CPU
GHA Cache Buildx type=gha, mode=max 14 min 57 s 14.4 s Cold: AMD EPYC 7763 · Warm: AMD EPYC 9V74
GHCR Buildx type=registry, mode=max 9 min 26 s 8.0 s AMD EPYC 9V74
Cache hits

BoringCache · GitHub runner · 4 vCPU

97.2% of 1,061 lookups over 30 builds

Stored cache
BoringCacheImage-layer cache tag; bytes referenced by 1 current tag
1.97 GB

How the benchmark is measured

Projects and commits
Each project is pinned to an upstream repository. Rolling builds advance along its main branch by a fixed number of commits per step, and every cache builds the same commit.
Runner class
Caches are compared on the same runner class: provider, vCPU count, memory and platform. Remote builders are shown on their own compute.
CPU models
One runner size can run on several CPU models: 6 for GitHub's 4-vCPU x64 runners and 1 for its Arm runners in this report. A newer model can change a long build by several minutes. Each comparison also shows its ratio on the build pairs where both caches ran on the same CPU model. That subset is a separate measurement, not an adjustment to the full sample.
Cold and warm
Cold and warm timings are grouped by their source run, separately from the changed-build averages. Cold starts with an empty cache; warm rebuilds the same source on a new runner. Tables from different runs are separate observations.
Average build time
Each bar averages up to 30 commits that passed in every published lane on that runner, within one rolling series and BoringCache CLI version. We select complete matched samples across releases; we do not combine individual lanes from different samples. Counts appear beside the times. Bars start at zero and share one scale per project. Headline speed ratios compare the same runner class and cache capabilities.
Timing
Times cover validated builds and include cache restore, the build and cache save, including post-job saves. Checkout, preparation, CLI download and output checking are outside the timer. Changed builds start with cache from an earlier commit.
Time saved and runner cost
Time saved is the difference between average times on the same commits. Runner cost multiplies it by the provider's listed price per minute for that runner size, checked 14 October 2026. Remote builders are compared with BoringCache on a smaller runner over the commits both built. A cache added to a project joins its comparison after five counted builds, or after its first when BoringCache has fewer than five.
Stored cache
Stored cache was measured once for the selected samples, separately from the build times. GitHub Actions cache sizes include the latest archive and retained archives. BoringCache sizes add the bytes referenced by each current image-layer and Cargo cache tag, so content shared between tags counts once per tag.

These averages compare successful builds, not availability. The report excludes the following periods when cache uploads were rejected:

  • October 8, 2026 12:55 to October 8, 2026 16:20 UTC: Cache uploads rejected at the workspace storage limit and during a deploy (1,026 records).
  • October 8, 2026 21:35 to October 9, 2026 07:25 UTC: Cache uploads rejected at the workspace storage limit (963 records).

3,376 builds in the selected samples. Every build is recorded in the benchmark records.

Find out what your build can reuse.

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