What vCPU Counts Do and Do Not Tell You
A vCPU is usually a hardware thread, not a core, and counts are only comparable within one provider and family. Here is how to read them.
vCPU is the headline number on nearly every compute listing, and it is one of the least standardised figures in the industry. Two machines advertising the same count can differ substantially in what they actually deliver.
What a vCPU usually is
On most providers a vCPU corresponds to a hardware thread rather than a full physical core. Modern processors run two threads per core, so an eight vCPU machine is commonly four physical cores presented as eight logical processors.
Threads on the same core share execution resources. Two threads do not deliver twice the throughput of one; the uplift depends heavily on the workload, and for some compute-bound work it is modest.
Why counts do not compare across providers
Three variables sit behind the number:
- Processor generation and clock speed. A newer core at a higher clock does more per second than an older one.
- Threads or cores. A minority of providers allocate full cores, which changes the meaning entirely.
- Contention. Whether other tenants share the underlying physical core.
That last point is the biggest, and it is covered in shared versus dedicated vCPU.
When the count is the wrong number to look at
Most workloads are not limited by processor count. Databases are frequently limited by storage latency; applications with large working sets are limited by memory; proxies are limited by network throughput.
Adding vCPUs to a workload waiting on disk changes nothing except the invoice. Identifying the real constraint first is the subject of how to size your first cloud server.
Where the count genuinely matters
- Parallel compilation and build pipelines.
- Media encoding and transcoding.
- Scientific and numerical simulation.
- Serving many concurrent requests that each do real computation.
vCPU as an account-level figure
In account listings, a vCPU number often describes quota rather than a single machine: how much compute the account is allowed to run in total. That is a different claim from the size of one server, and it is the sense used in the AWS and OVHcloud configurations.
The practical test
Benchmark with your own workload for an hour. A representative run tells you more than any specification sheet, because it measures the combination of processor, memory bandwidth and storage that your application actually depends on.
Cloud accounts on OldCloudShop
Ten providers, every configuration and the starting price for each.
Continue in the library
Other entries covering account structure, billing, security and provider infrastructure.