NVLink Congestion Routing

What this hard GPU network routing problem tests

This is a hard coding problem that combines graph algorithms with practical systems thinking—the kind Nvidia and other GPU cluster operators ask to assess how you optimize communication in high-performance computing environments. It probes whether you can model a constrained pathfinding problem and implement an efficient solution under time pressure.

The core challenge is finding the path that maximizes the minimum link capacity along the route (known as the widest path or bottleneck shortest path problem). Unlike standard shortest-path algorithms that minimize total cost, here you must reason about the constraint that the slowest link determines overall throughput. Strong solutions recognize this as a variant of Dijkstra's algorithm adapted to work with a maximization objective, handle effective bandwidth as a derived quantity, and manage edge cases like disconnected nodes or fully congested links.

• Greedy graph traversal and priority queues
• Bottleneck path optimization (widest-path semantics)
• Effective capacity and congestion modeling
• Floating-point precision and rounding in physics-adjacent calculations