8,192 GPUs exceeds what a two-tier Clos of 64-port switches supports. Compare adding a tier against using higher-radix switches.

Past the two-tier limit there are two ways forward and they are not close. One triples the switch count and adds fifty percent more optics for two extra hops of latency; the other is the same shape with bigger switches. The counts for both, and the reason the industry follows switch radix.

Updated Sep 2026 · Grounded in real AI infrastructure interview loops and written to a senior-engineer editorial bar, with every number worked and every diagram hand-built.

20 answers per topic instead of 10, plus saved progress and bookmarks · no cardor unlock all 283 remaining answers · ₹2,000 / $25

The concepts behind this question

Ranked by how closely each one overlaps this question's topic, so the first card is the thing to read if the answer above moved too fast.

Advanced

🔌 Networking & Storage🔒 Premium

Rail-Optimized and Fat-Tree FabricsA GPU cluster's network is built from two ideas: a fat tree (Clos) that gives every node a path to every other node with a chosen amount of oversubscription, and rail optimization, which wires GPU i of every node to the same leaf switch so the collectives that dominate training stay one hop away. Sizing one is arithmetic on port counts, and the interview question is usually that arithmetic: how many switches, what oversubscription, and where the NVLink domain ends and the fabric begins.

Foundational

🖧 Hardware & Cluster Build-Out

Cables, Transceivers and the Optics Power BudgetCable choice is set by distance and it is the most common ordering mistake in a GPU cluster build. Passive copper reached 3 m at 400G and tops out near 2 m at 800G, so a bill of materials copied from the previous generation produces links that will not come up. Beyond copper come active copper, then active optical cables, then transceivers and fiber. Each step adds reach and adds power, and at cluster scale the transceivers alone draw tens of kilowatts that nobody budgeted.

Foundational

📐 AI Systems Design

Multi-Region Serving and FailoverRunning inference in more than one region buys latency for distant users and survival when a region fails, and it costs a second fleet that must be capable of absorbing the first one's traffic. The design turns on three decisions: whether regions are active-active or active-passive, what state has to cross regions and what deliberately does not, and how much headroom each region carries so a failover does not simply move the outage.

Advanced

🕸️ Distributed Training🔒 Premium

Ring vs Tree All-ReduceA ring all-reduce moves the minimum possible bytes per rank but takes 2(N-1) steps, so its latency grows with the number of GPUs; a tree finishes in a logarithmic number of steps but is harder to keep bandwidth-optimal. NCCL keeps both, chooses per message size and rank count, and reading its choice is how you diagnose a collective that is slower than the fabric allows.

Explore the full curriculum →

UP NEXT ON YOUR JOURNEY

Next in this trackHow would you know GPUDirect RDMA is not being used, given that nothing reports an error when it is disabled?Next in this trackDesign storage for a multi-tenant GPU cloud. How do you keep one customer's checkpoint burst from slowing another's training run?Next in this trackAfter a switch firmware update, all-reduce bandwidth across the cluster fell 20 percent. Every link is up and no errors are logged. Find it.

FEDITOR'S NOTE

Scored on deriving the endpoint limit from radix, on counting switches and optics for both designs, and on the latency and failure-domain consequences of the extra tier.

DISCUSSION · 0

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