What is a vCPU?

A vCPU (virtual CPU) is a unit of CPU compute that cloud providers allocate to your instance. It's not a physical chip — it's a virtualized slice of a physical CPU core.

On most cloud providers, 1 vCPU = 1 hardware thread. Modern CPUs use hyperthreading (Intel) or SMT (AMD), which makes each physical core appear as 2 threads. So 1 physical core = 2 vCPUs.

When you see a GPU instance with "26 vCPUs," that means you get 26 hardware threads — roughly 13 physical cores — dedicated to your instance for data loading, preprocessing, and running your training framework.

In a GPU cloud instance, the GPU does the heavy math, but the CPU does everything else:

• Data loading — reading training data from storage, decompressing, decoding images
• Preprocessing — tokenization, augmentation, normalization
• Data pipeline — feeding batches to the GPU fast enough to keep it busy
• Framework overhead — PyTorch/TensorFlow runtime, gradient synchronization
• Monitoring — logging, checkpointing, metrics collection

If you don't have enough vCPUs, the GPU sits idle waiting for data. This is called being CPU-bound — your expensive GPU is underutilized because the CPU can't keep up.

vCPUs are in the compute section:

{
  "compute": {
    "vcpus": 26,
    "ram_gb": 200,
    "storage_gb": 512,
    "storage_type": "nvme-ssd",
    "network_gbps": 25
  }
}

compute.vcpus is the total number of virtual CPUs allocated to the instance. For multi-GPU instances, this is the total — not per GPU.