GitHub Actions runners with the best disk value
Blacksmith Linux x64 leads the measured results at 665.8K per USD. 26 of 26 configurations have a score. All value scores use USD compute rates.
A fast disk can help a file-heavy workflow, but its compute rate also affects the choice. Disk value divides our four-component disk score by the USD cost per vCPU per minute. The ranking highlights measured storage performance at a normalized compute rate. It calculates leaders from the same dataset, pricing, and dated exchange rate as the main tool, with every provider in one ranking.
By TechCompare · Updated
Current leaders
Results imported 2026-10-10. Pricing checked 2026-10-10. Higher is better. All operating systems and architectures are included. Prices and value use USD. RunJob's EUR rate is converted at €1 = $1.1206 (ECB, 2026-10-09).
Calculator
Ranked runner results
| RunnerRunner | DiskDisk score | Disk valueDisk value | Cost/vCPU/minCost/vCPU/minUSD | Max vCPUMax runner vCPU |
|---|---|---|---|---|
| Blacksmith Linux x64 AMD EPYC (model hidden) | 1,331.6 | 665.8K/USD | 0.002 | 32 |
| Avrea Linux x64 AMD EPYC 4585PX 16-Core | 864.5 | 432.3K/USD | 0.002 | 32 |
| RunJob Linux x64 AMD EPYC 7502 32-Core | 492.9 | 366.5K/USD | 0.0013 | 48 |
| StarSling Linux x64 AMD EPYC (model hidden) | 559.1 | 279.5K/USD | 0.002 | 64 |
| Blacksmith Linux arm64 ARM64 (model undisclosed) | 320.8 | 256.7K/USD | 0.0013 | 32 |
| Ubicloud Linux x64 AMD Ryzen 9 7950X3D 16-Core | 243.7 | 243.7K/USD | 0.001 | 30 |
| Avrea Linux arm64 Apple M5 Max | 1,239.3 | 206.5K/USD | 0.006 | 16 |
| Namespace Linux x64 AMD EPYC (model hidden) | 297.3 | 198.2K/USD | 0.0015 | 64 |
| Blacksmith Windows x64 AMD EPYC 4565P 16-Core | 708.9 | 177.2K/USD | 0.004 | 32 |
| Namespace Linux arm64 AmpereOne family (SKU undisclosed) | 197.8 | 131.9K/USD | 0.0015 | 32 |
| Ubicloud Linux arm64 Ampere Altra Q80-30 | 80.6 | 129K/USD | 0.0006 | 30 |
| Avrea macOS arm64 Apple M5 Max | 1,022 | 102.2K/USD | 0.01 | 16 |
| Avrea Windows x64 AMD EPYC 4585PX 16-Core | 335.3 | 83.8K/USD | 0.004 | 16 |
| Namespace macOS arm64 Apple M4 Pro | 948.7 | 63.2K/USD | 0.015 | 16 |
| GitHub macOS arm64 Apple M2 Pro | 656.7 | 31.8K/USD | 0.0207 | 5 |
| GitHub Linux arm64 Microsoft Cobalt 100 | 73.1 | 29.2K/USD | 0.0025 | 64 |
| GitHub Linux x86 AMD EPYC 7763 64-Core | 81.2 | 27.1K/USD | 0.003 | 96 |
| Latchkey Linux x64 AMD EPYC 7R13 | 28.3 | 22.7K/USD | 0.0013 | 16 |
| GitHub Linux x64 AMD EPYC 9V74 80-Core | 67 | 22.3K/USD | 0.003 | 96 |
| Blacksmith macOS arm64 Apple M4 Pro | 262.4 | 19.7K/USD | 0.0133 | 12 |
| BuildPulse Linux arm64 AWS Graviton (gen undisclosed) | 20.1 | 16.1K/USD | 0.0013 | 64 |
| GitHub Windows x64 AMD EPYC 7763 64-Core | 76.7 | 15.3K/USD | 0.005 | 96 |
| BuildPulse Linux x64 AMD EPYC 9R45 | 28.2 | 14.1K/USD | 0.002 | 64 |
| GitHub Windows x86 AMD EPYC 7763 64-Core | 68.6 | 13.7K/USD | 0.005 | 96 |
| GitHub Windows arm64 Cobalt 100 | 35 | 7K/USD | 0.005 | 64 |
| GitHub macOS x64 Intel Core i7-8700B | 91.9 | 5.9K/USD | 0.0155 | 12 |
Single thread CPU: CoreMark iterations/s. Disk and cache: composite MiB/s. Internet: composite Mbps. Value: score ÷ USD cost/vCPU/min. K means thousand, M means million. Higher is better. Cache uses each runner's best tested cache or artifact backend. Limits: maximum runner vCPU. Prices and value use USD. RunJob's EUR rate is converted at €1 = $1.1206 (ECB, 2026-10-09).
Open the full GitHub Actions runner benchmark toolCalculate disk value from the tested configuration
Disk value equals composite disk MiB/s divided by quoted cost per vCPU-minute. A hypothetical disk score of 500 at $0.001 per vCPU-minute gives 500,000 per USD. The rate belongs to the tested runner tier and its researched billing assumptions. It is not the price of the largest available shape. More vCPUs do not multiply this disk score or establish how storage scales at a larger size.
Disk sample coverage and missing evidence
Disk value uses the same accepted sample medians as the performance ranking. The updated BuildPulse ARM result is included using three completed samples, with its original counts in the full tool. A disk report without accepted verified measurements, or a missing or nonpositive price, leaves value unscored. Every scored configuration is ranked by value using a USD compute rate, followed by missing results. Reimporting disk measurements, correcting a price, or refreshing the EUR conversion recalculates the next built page and its leader summary.
Include the rest of the job bill
A file-heavy job can reserve an entire runner while it waits on reads or writes. Per-vCPU normalization makes rates easier to inspect, but the provider may bill by a fixed machine minute, round job durations, or require a plan. Consumed CPU billing has a different relationship to disk waits and can include separate memory charges. The ratio does not model those costs. Use actual job records to estimate the savings from a provider change.
Measurement context
The disk numerator is a fixed geometric mean of sequential read and write bandwidth plus random read and write throughput. Its components do not change when a cheaper or faster runner is added. The denominator is a compute rate, not a storage-only price. A runner with a low rate can rank well here without leading the raw disk table.
Verdict
Compare raw disk performance alongside value before choosing a runner. A high ratio is helpful when price matters, but slow absolute throughput can still hold up a time-sensitive job. Confirm complete workflow duration and billed usage at the requested size.
More runner benchmark scenarios
Frequently asked questions
Which runner currently leads for disk value?
Do these runner rankings update automatically?
Are all operating systems and architectures included?
Does a missing result mean a runner scored zero?
Related tools
Data Transfer Calculator
Estimate transfer times for files over USB, WiFi, Ethernet, and more.
Use tool ➜Memory and Storage Throughput Visualizer
Visualize the massive speed difference between CPU cache, RAM, and storage.
Use tool ➜RAM Latency Calculator
Convert DDR3/DDR4/DDR5 timings (CL, tRCD, tRP, tRAS) into true latency in nanoseconds.
Use tool ➜