Nvidia Interview Process - Intern Software Engineer

GPU Performance Engineering internship on NVIDIA's Deep Learning and HPC team in Santa Clara, California. Two rounds in December 2024. I applied cold, directly. To prepare I read Glassdoor interview experiences and practiced Python, object-oriented programming and systems fundamentals.

Round 1: Python, HPC, Linux and Docker

• Python: counting occurrences of elements in a list. I used a dictionary to track frequencies, explained the time and space complexity, and we discussed handling an empty list.

• HPC: why doubling the number of GPUs might not double performance. I talked about communication overhead, synchronization, and the parts of the workload that can't run in parallel.

• Linux: how to find a running process and stop it.

• Docker: the difference between an image and a container.

Round 2: Object-oriented programming and Python

Design a Python class that stores benchmark results and calculates average runtime. I explained how I'd initialize the class, add measurements, and handle a request for an average before any measurements were recorded. The discussion was about organizing data and behavior, clear method responsibilities, and explaining my design decisions.

Tips

• Practice writing Python while explaining your reasoning. Review dictionaries, lists, classes and edge cases.

• For systems topics, understand how compute, memory bandwidth, temperature, batching and communication can limit performance.

• Hands-on Linux and Docker experience helps.

• Be ready to discuss your projects: what you contributed, why you chose your approach, and how you verified your results.

Outcome

Received an offer.