AI Infra Interviews logo
← Distributed Training & Parallelism / 37
expertNewGoogleNVIDIA

Halve the experts and top-K, then double expert width. What actually gets cheaper?

Compare two synthetic MoE designs with the same routed parameter budget. Derive assignment counts, dispatch bytes and expected weight coverage before predicting a throughput gain.

Updated Nov 2026 · Learn how AI infrastructure works through explanations, worked examples and diagrams, then practise applying it to interview questions.

Compare two synthetic MoE designs with the same routed parameter budget. Derive assignment counts, dispatch bytes and expected weight coverage before predicting a throughput gain.

The full answer is part of Premium. A free account includes more answers in each topic, but does not unlock this one.

20 answers per topic instead of 10, plus saved progress and bookmarks · no cardor unlock all 327 remaining answers · ₹2,000 / $25

Work through the numbers

Change the assumptions in a free calculator, then explain what the result establishes and what still needs measurement.

UP NEXT ON YOUR JOURNEY
FEDITOR'S NOTE

Original architecture-design exercise, not a reported company question. The important distinction is between wider expert matrices, more rows per expert and fewer total assignments. Only the first and third follow from the proposed change.

DISCUSSION · 0

No comments yet — be the first to share your approach.