Products

Seven resource networks, one meter

A machine is not one resource. It is compute, memory, an accelerator, a disk, a network link and a power draw — and every one of those is spent separately, so every one of them should be measured separately. Each network below is metered in its own unit and signed by the node that provided it.

Why two of these say “partially measured”

Because they are. Each node benchmarks itself and signs the result; where a benchmark does not exist yet, the node records configured, not-yet-benchmarked and emits no number in its place. We could have written a throughput figure on the GPU page and nobody would have checked. That is precisely why we did not.