GPU Compute
Also written as GPU, CUDA, TPU, Accelerator
The specialised chips model training and inference run on. They perform many small calculations at once, which is what makes training practical, and they are the dominant cost line in most AI work.
Think of it like
A thousand cashiers, rather than one very fast one.
Junior or senior?
High signal, because real capacity is scarce and shared.
Senior sounds like
Talks about utilisation, queueing and cost per run, not just which card they used.
Ask them
“How busy were your GPUs actually kept, and what did a training run cost?”
Sounds like real experience
Talks about utilisation, waiting for capacity, spot instances or cost per run — the constraints of someone who had to share a real cluster.
Probe further if
Names hardware but has no sense of utilisation or cost, which usually means a managed notebook rather than owning any infrastructure.