NVIDIA·Software Engineer·Technical Phone Screen·Intermediate

Intermediate

May 2026

Summary

NVIDIA coding round, one design question the whole time. Pretty focused on getting the implementation right rather than just talking through ideas.

Questions Asked (1)

Design and implement a disk space manager with a fixed total capacity. It should support storing datasets by ID and size, retrieving a dataset's size, and evicting existing datasets when space is needed. The capacity constraint is on total bytes used, not on the number of items stored.

System DesignAlgorithms & Data StructuresTechnical Trade-offs

Author's notes

The tricky part I kept second-guessing was the capacity semantics.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying requirements and constraints, then propose a design using a hash map for O(1) lookups and a doubly linked list for O(1) eviction (LRU). Discuss trade-offs and potential optimizations for large-scale systems.

Pro tip: Emphasize that the eviction policy should be pluggable (e.g., LRU, LFU, FIFO) and discuss how to handle concurrent access and persistence, showing awareness of real-world system design.

1. Clarify Requirements

Ask about expected dataset sizes, access patterns, eviction policy preferences, concurrency needs, and persistence requirements.

2. Design Data Structures

Propose a hash map for O(1) dataset lookup and a doubly linked list to track usage order for O(1) eviction. Explain how they interact.

3. Implement Core Operations

Detail the algorithms for store (add/update), retrieve (get size), and evict (remove least recently used) while maintaining capacity.

4. Handle Edge Cases and Concurrency

Discuss handling datasets larger than capacity, concurrent access (locks or lock-free), and thread safety.

5. Analyze Trade-offs and Optimizations

Compare eviction policies, consider memory overhead, and suggest improvements like sharding or persistence.

Key Points to Mention

Use a hash map for O(1) lookup and a doubly linked list for O(1) eviction (LRU).
Eviction policy should be configurable (LRU, LFU, FIFO) based on use case.
Handle datasets larger than total capacity by rejecting or splitting.
Ensure thread safety with locks or concurrent data structures.
Consider memory overhead of metadata and potential for sharding.
Discuss persistence and recovery options for durability.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.