BuyAWSAccount
FAQ Hub/What is a vCPU in AWS?

What is a vCPU in AWS?

A vCPU (virtual CPU) in AWS is a virtualized processing core assigned to an EC2 instance. Each physical CPU core typically provides 2 vCPUs through hyper-threading. AWS accounts have default vCPU limits per region that can be increased.

A vCPU (virtual Central Processing Unit) in Amazon Web Services represents a virtualized processing core that is assigned to an EC2 instance. Understanding vCPUs is fundamental to selecting the right instance types and managing your AWS compute capacity effectively.

In AWS's infrastructure, each physical server CPU uses Intel Hyper-Threading or AMD SMT technology to present two logical processor threads per physical core. These two threads are what AWS exposes as vCPUs. So a physical server with a 48-core CPU would present 96 vCPUs to the virtualization layer. Different EC2 instance types are allocated specific numbers of vCPUs: a t3.micro has 2 vCPUs, a c5.2xlarge has 8 vCPUs, a c5.18xlarge has 72 vCPUs, and the largest instances like u-24tb1.metal have 448 vCPUs.

AWS enforces vCPU limits (called service quotas) at the account and region level to prevent overprovisioning and to manage capacity. New AWS accounts by default are limited to relatively low vCPU counts — typically 32 vCPUs for standard instance families (running-on-demand-standard-instances). This means you cannot launch more EC2 instances than your total vCPU quota allows in that region. If you try to launch instances that would exceed your limit, AWS will return an error.

This quota system is why many users purchase AWS accounts with elevated vCPU limits. Instead of submitting a quota increase request through AWS Support (which requires justification and can take several days to approve), our accounts come pre-approved for specific vCPU counts: 8 ($35), 32 ($80), 64 ($120), 128 ($150), 256 ($190), or 512 ($299) vCPUs. These limits are for the running-on-demand instances quota, which covers the standard EC2 instance families most commonly used.

For workloads that require massive parallelism — such as distributed machine learning training, large-scale data processing, genomics research, or video rendering — having 128 to 512 vCPUs available immediately (rather than waiting for quota approval) can save days of productive work time. Each vCPU roughly corresponds to one thread of a modern server-grade processor, so 512 vCPUs is equivalent to running approximately 256 physical processor cores simultaneously.