Jenkins' hardware recommendations page, checked on 1 November 2026, refuses to give a single number, and it is right to: a Jenkins controller that coordinates a handful of pipelines and one that runs a company's builds are different machines. What the page does give is a model, and this guide applies it to a team-sized Jenkins on MassiveGRID PaaS, built on Virtuozzo Application Platform (formerly Jelastic), where the Jenkins package installs a controller with auto-scaling agents.
What Jenkins says about the controller
The hardware recommendations page states the range plainly: the RAM for a controller "can range from 200 MB for a small installation to 70+ GB for a single and massive Jenkins controller". It gives one per-unit cost, each build node connection takes 2 to 3 threads, about 2 MB or more of memory, and notes CPU overhead when many users hit the web interface. The main rule is architectural: "it is generally a bad practice to allocate executors on a controller", because builds can overload its CPU and memory and crash it; set up agents the controller delegates to, keeping the bulk of the work off the controller.
Agents, the page says, are typically generic machines "with enough memory to run specific build types". That is the honest answer: the agent's size is the size of your build, a Maven compile and a browser test suite being different animals. The installation guide adds only that Jenkins runs on any platform with a supported Java.
Jenkins on MassiveGRID PaaS: one-click install, about $59.04 a month for the Starter configuration (Jenkins controller, 2 GiB; Agent, 1 GiB) at $0.003372 per cloudlet-hour, before volume discounts, in New York, London, Frankfurt or Singapore. 14-day free trial, no credit card. See the sizing table.
Sizing a controller for a team
A controller for a team of ten to thirty developers, with a few dozen pipelines and plugins for Git, Docker and the usual notifications, sits comfortably in 2 GiB of heap once the JVM is told so, and wants 4 GiB when the plugin count grows or the build history is kept for months. The things that push it toward the page's upper range are long build logs kept forever, hundreds of jobs with heavy plugins, and executors on the controller, which the page rules out. Disk is the build history and artifacts: set a discarder on every job and archive artifacts to Nexus or object storage rather than the controller.
The three configurations on PaaS and their cost
The Jenkins package installs the controller with pre-installed plugins and auto-scaling agents, plus a platform subdomain and certificate. Cloudlets are 128 MiB RAM plus 400 MHz CPU, billed at $0.003372 per cloudlet-hour, $2.46 per cloudlet-month, no licence fee.
| Configuration | Controller | Agents | Cloudlets | Per month (730 h) | Working hours (176 h) |
|---|---|---|---|---|---|
| Starter | 2 GiB | 1 × 1 GiB | 24 | $59.04 | $14.24 |
| Standard | 4 GiB | 2 × 1 GiB | 48 | $118.08 | $28.47 |
| Production | 4 GiB | 4 × 2 GiB | 96 | $236.16 | $56.94 |
The right-hand column matters more for Jenkins than for most applications: agents are billed for the hours they run, and the package's auto-scaling adds and removes them with the queue, so a team that builds in office hours pays close to the working-hours figure for the agent layer even when the controller stays up. The PaaS pricing calculator prices other agent counts.
Agents are where the money and the speed are
Every minute a build waits in the queue is a developer waiting. On a fixed build farm the answer is more permanent servers; on the platform the answer is a trigger that starts another agent when the queue is non-empty and stops it ten minutes after it empties. The auto-scaling agents post shows the configuration and the graph of a release day: agents from one to five and back, billed for the hours between. Size the agent to the build, as the page says: 1 GiB for linting and unit tests, 2 to 4 GiB for Docker image builds and browser tests, and a separate larger agent label for the one job that needs it.
Upgrades and the plugin problem
Jenkins' real operational cost is the plugin ecosystem, where an upgrade of one plugin can break another and the LTS line moves every few months. Clone the environment, upgrade the clone, run the pipelines that matter against it, then upgrade production; the clone exists for an afternoon and costs cents. The migration post moves an old Jenkins onto the platform that way, and the Database Backup/Restore and Backup Storage packages keep JENKINS_HOME safe between upgrades.
Frequently Asked Questions
How much RAM does a Jenkins controller need?
Jenkins' hardware recommendations, checked on 1 November 2026, give a range from 200 MB for a small installation to 70 GB or more for a massive one, with each agent connection costing about 2 MB. A team-sized controller runs well in 2 GiB and comfortably in 4 GiB, which are the Starter and Standard configurations on MassiveGRID PaaS.
Should builds run on the Jenkins controller?
No. Jenkins calls executors on the controller a bad practice, because builds can overload and crash it. The package installs separate auto-scaling agents, and the controller is left to coordinate.
What does a Jenkins setup cost per month?
About $59.04 for a 2 GiB controller with one agent, $118.08 with a 4 GiB controller and two agents, and $236.16 with four 2 GiB agents on MassiveGRID PaaS, at $0.003372 per cloudlet-hour before volume discounts. Agents are billed only for the hours they run, so a team that builds in office hours pays much less than the always-on figure for them.
A controller that coordinates and agents that come and go
The Jenkins package on MassiveGRID PaaS installs the controller with pre-installed plugins and auto-scaling agents in one click, from about $59.04 a month always on and far less when the agents run only in office hours. 14-day free trial, no credit card.
Jenkins on PaaS