Our enhanced runners are executed on bespoke hardware that sports the following.
- a sustained CPU frequency of 5.1Ghz
- burst up to 4X the requested core count at no additional charge
- an Intelligent Cache designed from the ground up for CI workloads
- high speed local storage
- pre-scaled nodes for quick job start
- flexible sizing instead of fixed cpu counts (ex. 0.5 core, or 12 cores)
- log noise reduction
The features combine to provide the following over Gitlab.com/similar runners.
- generally a 2-10X speed up* without modification
- 2-3X hardware performance over Gitlab.com
- Intelligent Cache overhead reduction and cache expansion
- increased reliability
- improved engineer efficiency
- cost savings
The improved performance and reliability result in a shorter execution time which drastically reduces the effective cost in terms of dollars and engineer time. When combined with the 4X core burst and non-expiring credit, in contrast to Gitlab.com quota, the enhanced runner provides an incredible value.
* Speed-up is not guaranteed. Extensive network requests and poor resource utilization can negate enhanced runner optimization.
Efficiency
Given intermittent spikes of CPU are common of CI workloads, CPU burst can efficiently share the underlying compute resource to provide 4X the power at no additional cost. Instead of reserving CPU even when not in use, unused CPU can be shared, which benefits all customers, while still guaranteeing access to the requested power.
Noise reduction
Instead of showing everything one could possibly ever want to know, and then some, we have pruned the information emitted by the runner and default to collapsing all but the most relevant information. Instead of squinting and scrolling to figure out what went wrong, we remove everything else so you only see what went wrong.
Consider the following job that only prints 7 lines of output. The example does not contain caches, artifacts, nor even git clone output which all add noise, but still the results are evident.