
Two candidates describe Nvidia AI interviews: an intern faced transformer coding and math rounds, while an AI Developer Technology Engineer faced GPU architecture and performance questions.
$200K
Avg. Base Comp
$460K
Avg. Total Comp
Not reported
Typical Rounds
Not reported
Process Length
The two candidate reports behind this guide describe different roles, so they are best read as two views of Nvidia's AI hiring rather than one fixed loop. One is an AI internship and the other an AI Developer Technology Engineer role. Both candidates were not offered the job.
The intern describes a process built on fundamentals. After a relaxed phone screen about past projects, there was a PyTorch GAN take-home and two back-to-back 90-minute interviews, one on probability and linear algebra (including solving Ax = B) and one on transformers and CNNs. The most memorable part was a live coding task: build a causal mask for a decoder transformer from scratch, then reimplement np.zeros_like in native Python. A behavioral round on projects and optimizing LLM usage closed the loop, and feedback took several weeks.
The AI Developer Technology Engineer describes a longer, hardware-oriented path: a two-hour call with the hiring manager, two calls with team members, a take-home presented to a panel, and a virtual final round. Questions centered on optimizing common deep learning operations, CPU and GPU architecture, latency and bandwidth, and how past work relates to parallel computing. No CUDA was written, but each first answer was pushed with follow-ups.
For preparation, rehearse transformer internals in PyTorch, refresh linear algebra and probability, and practice explaining GPU memory and parallel execution aloud, defending your reasoning step by step.
Synthesized from 3 candidate reports by our editorial team.
Had an interview recently?
Share your experience. Unlock the full guide.
Real interview reports from people who went through the Nvidia process.
I was in the final panel round for a role at NVIDIA, and it was done onsite, so I got to meet the hiring manager in person. The panel wrapped up last Friday, and I'm still waiting on a decision. Then a manager from my panel, who wasn't the hiring manager for the original role, asked whether I'd interview for a different role on his own team while the first decision was still pending.
That put me in an awkward spot. I didn't want to hurt my chances with the first team, and I wasn't sure whether I was allowed to run two loops at once or whether the first team would find out and change its mind. I posted the question and got conflicting answers. Some people said parallel interviews are normal and that the recruiter is the one who tracks them. Others said NVIDIA doesn't allow them and that it can be treated as a lack of integrity. I also wondered whether the panel manager reaching out meant anything about how the first team saw me, but nobody could say that for sure.
The outcome is still pending for me. I'm planning to ask my recruiter directly whether parallel interviews are okay and whether they can speed up the first decision.
Prep tip from this candidate
Before you schedule an interview with a second team, ask your recruiter whether parallel loops are allowed, since commenters on this thread disagreed. Tell them where your current panel stands so they can coordinate the timing.
Share your own interview experience to unlock all reports, or subscribe for full access.
Sourced from candidate reports and verified by our team.
Topics based on recent interview experiences.
| Question | |
|---|---|
| Merge Sorted Lists | |
| Closed Accounts | |
| LRU Cache 1 | |
| Employee Project Budgets | |
| Shortest Transformation | |
| Reservoir Sampling Stream | |
| Integer to Roman | |
| Weighted Average Campaigns | |
| Using R Squared | |
| Sample Time Series | |
| Bias vs. Variance Tradeoff | |
| Justify a Neural Network | |
| The Brackets Problem | |
| Type I and II Errors | |
| Skewed Pricing | |
| Percentage of Revenue by Year | |
| Unbiased Estimator | |
| Singly Linked List | |
| Concentric Circles | |
| Overfit Avoidance | |
| Target Indices | |
| Minimum Directional Path | |
| 5th Largest Number | |
| NxN Grid Traversal | |
| Matrix Rotation | |
| Bias Variance Tradeoff | |
| FAQ Matching | |
| Merge N Sorted Lists | |
| Matrix Multiplication |
Synthesized from candidate reports. Individual experiences may vary.
Both candidates started with a conversation about their work. The intern describes a relaxed phone screen about prior projects and how to train an LLM to handle a specific set of tasks. The AI Developer Technology Engineer describes a two-hour technical and background call with the hiring manager.
The AI Developer Technology Engineer reports two phone calls with team members, followed by a take-home assignment presented to a panel. The intern reports a separate take-home in which they worked with GANs in PyTorch. Each take-home was described by only one candidate, so the format may differ by role.
The intern describes two 90-minute interviews back-to-back: first probability and linear algebra, including Ax = B, then transformers and CNNs. The AI Developer Technology Engineer describes questions on optimizing deep learning operations, CPU and GPU architecture, latency and bandwidth, a general ML question and a brain teaser, with probing follow-ups.
The intern wrote a causal mask for a decoder transformer from scratch, then np.zeros_like in native Python, and finished with a behavioral round on projects and optimizing LLM usage. They report that feedback took several weeks. The AI Developer Technology Engineer describes a final round that was virtual, though it used to be onsite.