Nvidia Interview Questions

Louie Z. - "DNNs can learn hierarchical features, with each layer learning progressively more abstract features, and generalizes better. SNNs are better for simplier problems involving smaller datasets and if low latency is required."See full answer

Srini K. - "performance issues and sudden spikes on input requests by scaling techniques and optimization."See full answer

DM - "Clarifying my assumptions first, i.e. "Technical contributions are not just writing or reviewing code". Answer : My contributions to any program I lead as a TPM are as follows : Inputs towards design, architecture review, mapping requirements to the proposed design, reviewing the implementation strategy w.r.t scalable solution, Writing core user-centric test scenarios, Validating proposed design vs implementation estimates vs initial planning. etc... Example : Situation: We are a"See full answer

Praveen D. - "Let me try to explain it with simple life analogy You're cooking dinner in the kitchen. Multithreading is when you've got a bunch of friends helping out. Each friend does a different job—like one chops veggies while another stirs a sauce. Everyone focuses on their task, and together, you all make the meal faster. In a computer, it's like different jobs happening all at once, making stuff happen quicker, just like having lots of friends helping makes dinner ready faster."See full answer

Curly T. - "One thing is not clear to me, We encoded the length of the word to a character, but the max number which can be converted to char ascii is 255. How will it work for length till 65535?"See full answer

Scott S. - " Project Overview: Real-Time Risk Management System Objective The goal was to develop a real-time risk management system capable of processing and analyzing large volumes of trading data to provide near-instantaneous risk assessments. This system was crucial for enabling traders to make informed decisions while managing their exposure to various market risks in real-time. Complexity Factors 1. \\Data Volume and Velocity\\ \\High Throughput:\\ The system needed to ha"See full answer

Rick E. - " from typing import List def getnumberof_islands(binaryMatrix: List[List[int]]) -> int: if not binaryMatrix: return 0 rows = len(binaryMatrix) cols = len(binaryMatrix[0]) islands = 0 for r in range(rows): for c in range(cols): if binaryMatrixr == 1: islands += 1 dfs(binaryMatrix, r, c) return islands def dfs(grid, r, c): if ( r = len(grid) "See full answer

Tracy M. - "Clarify LLM purpose; research, extraction, reasoning, planning, assessing? Assuming document extraction skill only, large db, multiple doc types, unstructured data; assuming Haiku-2.5 (less complex tasks) and LLM parsing tool exists pre-extraction call. Assuming golden dataset for ref is accessible, experiments and observability set in place. Likewise prompts. Eval set:document extraction coverage: % of accurately extracted (doc type, structured outputs) document field/value/format coverage"See full answer

Rohith K. - "Stakeholder management is kinda a critical aspect of product management. Because you would be the glue between the leadership team, design, GTM and engineering team. My simple approach to handle different stakeholders is - Deeply grounding myself into the product vision and business value that we are envisioning to create. Any decision or initiative needs to directly or indirectly benefit that path Open communication - any new feature, solution or product decision taken needs to be communi"See full answer

Jyoti V. - "Over-fitting of a model occurs when model fails to generalize to any new data and has high variance withing training data whereas in under fitting model isn't able to uncover the underlying pattern in the training data and high bias. Tree based model like decision tree and random forest are likely to overfit whereas linear models like linear regression and logistic regression tends to under fit. There are many reasons why a Random forest can overfits easily 1. Model has grown to its full depth a"See full answer

Surbhi G. - "Deep Learning is a part of Artificial Intelligence, it's like teaching the machine to think and make decisions on its own. It's like how we teach a child the concept of an apple - it's round, red, has a stem on top. We show them multiple pictures of apples and then they understand and can recognize an apple in future. Similarly, we feed lots of data to the machine, and slowly, it starts learning from that data, and can then make relevant predictions or decisions based on what it has learnt. A co"See full answer

Jet 1. - "`#include using namespace std; void printNumbersTillN(int n){ if(n_==0){ return; } printNumbersTillN(n-1); // go to the end -> reach 1 cout>_n; printNumbersTillN(n); }`"See full answer