How can I limit the CPU an application uses?

What this is#

One process, a backup compressing, a video encode, an overeager script, pins the CPU and everything else suffers. Two different goals hide in "limit its CPU", and they want different tools:

• "It can use idle CPU, just never at others' expense" → priority (nice). Usually what you actually want: the job still finishes as fast as conditions allow.
• "It must never exceed X% no matter what" → a hard cap (CPUQuota, cpulimit, --cpus). Right for untrusted or misbehaving things; understand it leaves CPU idle on purpose.

Priority: nice and renice (the polite fix)#

nice -n 19 tar czf backup.tar.gz /var/www     # start at lowest priority
renice 19 -p 12345                            # demote an already-running PID

A nice-19 process uses all the idle CPU but yields instantly when anything else wants a turn, the batch job barely slows down, and the site stops noticing it. For disk-heavy jobs, the I/O sibling: ionice -c3 <command>. This should be the default posture of every scheduled heavy job you own.

Hard caps: systemd CPUQuota (the right tool for services)#

For anything running as a service (or a timer job), cgroup-backed limits are precise and permanent, in the unit's [Service] section:

CPUQuota=50%

(50% of one core; 200% means two cores.) One-liner for an existing service, no file editing:

systemctl set-property myapp.service CPUQuota=50%

MemoryMax= lives in the same family, it's the same mechanism the timers guide recommends for heavy backup jobs.

cpulimit: a cap for a one-off process#

For a process that isn't a service and needs restraining now:

apt install cpulimit
cpulimit -p 12345 -l 50       # hold PID 12345 at ~50% of one core

It works by rapidly pausing/resuming the process, crude but effective, and nothing to clean up afterwards.

Docker: --cpus#

Containers get the cgroup treatment natively: docker run --cpus="1.5" ..., or in compose, deploy.resources.limits.cpus: "1.5", the same knob the container-sprawl section recommends so one stack can't starve the VPS.

The honest footnote#

A cap protects neighbors inside your VPS; it doesn't create CPU. If the workload legitimately needs more than the plan has, caps just make it slower, extra vCPU or a bigger plan is the real fix, and if the CPU is being eaten by something you don't recognize, that's a different article entirely.

Still need help?#

You can open a support ticket. So we can help on the first reply, it's worth mentioning:

• the VPS hostname or IP,
• the process you're capping, and whether you want priority (nice) or a hard cap.

Related questions#

• "How do I stop a backup job from slowing my website?"
• "How do I cap a service at 50% CPU?"
• "What's the difference between nice and CPUQuota?"
• "How do I limit CPU for a Docker container?"