lab
Inference infrastructure, hands on
Small simulations of the trade-offs behind serving models on GPUs: how slicing changes the cost of throughput, how a fleet autoscales through a traffic day, and what a gateway should do when a region fails.
demos
01building
GPU slicing calculator
How many model replicas fit on a GPU, and when MIG slices beat whole GPUs on cost per unit of throughput.
02building
Inference fleet simulator
Autoscale a GPU inference fleet through a traffic day, then survive an on-demand capacity outage.
03building
Gateway failover
Retries, failover and data residency across three regions behind an inference gateway.
ground rules
- Everything runs in your browser. No backend, no GPU, no API keys.
- Numbers are illustrative or taken from public cloud pricing, with the date they were taken.
- These are teaching models, not benchmarks. Each demo says what it leaves out.