The agreed performance goals for the accelerator on its target workloads: throughput, latency, efficiency and scaling, with the conditions under which each is measured.
Requirements Work product in the AI accelerator lifecycle and the Fabless startup chip lifecycle
They tell architecture, design, software and benchmarking teams exactly what the product must achieve, and they become the basis of customer claims. If targets are not agreed, each team optimises its own metric and performance validated has no clear pass criteria.
The full lifecycle detail, with its charts, flows, tables and every section, is for registered members. Registration is free and takes a minute. You come back to this page afterwards.