What is a vCPU? Cloud cores, threads and memory ratios

Von Rohit Athaide · Veröffentlicht am 4. Nov. 2026

Short answer: a vCPU is a unit of processing time on a physical machine, and what it maps to depends on the cloud and the processor. On most x86 machines it is one hardware thread, so two vCPUs are one physical core. On Arm machines it is usually a full core. That is why you cannot compare vCPU counts across machines without checking.

Key points

• vCPU is a billing unit, not a physical measurement.
• x86: two vCPUs usually equal one core. Arm: one vCPU is usually one core.
• Oracle sells by OCPU, one physical core, which equals two x86 vCPUs.
• Memory per vCPU is set by the machine family.
• Burstable machines share CPU and can slow under sustained load.

Is a vCPU a thread or a core?

Most x86 processors run two threads per physical core. On many x86 machines, a vCPU is one thread, so two vCPUs equal one core. On Arm machines, a vCPU is usually a full core. Oracle prices by OCPU, which is one physical core, and Oracle’s own pricing pages explain that one OCPU gives two vCPUs of compute on x86, so its per-OCPU price is about twice a per-vCPU price.

That is why a four-vCPU Arm machine and a four-vCPU x86 machine do not perform the same, and why a core count from one cloud cannot be set against another cloud’s vCPU count without checking the vendor’s definition. Vendor pages: EC2 instance types, Azure VM sizes, Compute Engine machine families.

How much memory comes with each vCPU?

Each machine family sets a ratio of memory to vCPUs:

• Compute-heavy families give less memory per vCPU, around 2 GB.
• General purpose families sit in the middle, around 4 GB.
• Memory-heavy families give the most, around 8 GB or more.

Pick the ratio your workload needs. Paying for memory you do not use is a common reason a bill is higher than needed.

What does a worked example look like?

Made-up round numbers to show the method, not real prices. You need the equivalent of 8 x86 threads of compute.

The Oracle-style row looks twice as expensive per unit and costs the same, because the unit is two threads. The Arm row costs less and gives full cores, but the software must run on Arm. Compare cost per equal piece of work, which means cost per thread of the same kind, not cost per unit.

What is the difference between shared and dedicated CPU?

Some small sizes share a CPU and earn credits when idle, which they spend during short bursts. If you run at high load for long, credits run out and speed drops. See burstable vs fixed-performance machines.

What do I do when comparing?

1. Compare the same architecture, so Arm with Arm. See Arm or x86.
2. Compare the same memory per vCPU.
3. Check what the vendor means by a vCPU on the page for that family.
4. Test your real workload if the choice matters.

Which mistakes are common?

• Treating an OCPU as a vCPU.
• Buying vCPUs for a single-threaded program.
• Choosing memory-heavy sizes for workloads that need CPU.
• Comparing a burstable size with a fixed one.

Decision checklist

• Does my program use more than one thread?
• Is it CPU-bound or memory-bound?
• Is the comparison Arm with Arm and x86 with x86?
• Is the size burstable or fixed?

Next steps

Our web app estimator fixes the vCPU and memory for each cloud’s server and shows the sizes it picked. For sizing from your own load, read choosing a VM size and instance equivalents across clouds.