Which characteristic of a GPU makes it better than a typical CPU for training deep neural networks?
Nca Aiio Essential Ai Knowledge
Question · Easy
Spot the AI hallucination
Short
Which NVIDIA hardware unit is purpose-built to accelerate mixed-precision matrix-multiply-accumulate operations used in deep learning?
NVIDIA's software layer that lets workloads use GPU throughput without going through graphics APIs.
Essential AI Knowledge
Guess the Term
Chapter 1: Exam Overview, AI Fundamentals & Adoption Drivers
1.1 Exam Structure, Logistics & Certification Roadmap
Study guide
What is the primary distinction between training and inference workloads for an AI model?
Read the clue and guess the term in 6 tries.
Game
What is CUDA in the NVIDIA software stack?
NVIDIA's GPU-accelerated set of primitives for routines such as convolution, attention, pooling and normalization.
Which NVIDIA library provides primitives for collective communication (all-reduce, all-gather, broadcast) across multiple GPUs?
Question · Medium
Which NVIDIA framework is purpose-built to train, customize, and deploy generative AI and large language models with techniques like RLHF and PEFT?
NVIDIA's collection of open, modular libraries for specializing, optimizing and governing AI agents.
1.2 AI, Machine Learning & Deep Learning Workloads
Which NVIDIA software product packages enterprise-supported AI frameworks, libraries, and tools as a single licensed software stack?
Swap tiles to line up three of the same term.
Term Match
What problem does NVIDIA NIM (NVIDIA Inference Microservices) primarily solve?
NVIDIA's open-source software collection for running entire data science pipelines on GPUs.
Which NVIDIA suite is designed for end-to-end GPU-accelerated data science (DataFrame, ML, graph) using a familiar pandas/scikit-learn-like API?
Which NVIDIA inference compiler and runtime is specifically optimized for transformer-based large language models?
___ Core: specialized GPU core that accelerates mixed-precision AI and HPC workloads.
1.3 Training vs. Inference Computational Profiles
In data parallel training, what is replicated across each GPU and what is split?
Flip two cards at a time to find matching terms.
Term Pairs
What is the primary motivation for using model parallelism (tensor or pipeline) instead of pure data parallelism?
Abbreviation for the class of deep-learning systems, mostly transformer networks, that summarize, translate, predict and generate content from huge datasets.
ZeRO (Zero Redundancy Optimizer) is associated most directly with which optimization?
Question · Hard
Which precision is most commonly used for the activations and gradients during mixed-precision training on Hopper GPUs to balance accuracy and throughput?
Open-source serving software, now part of NVIDIA's Dynamo family, that deploys models built with PyTorch, TensorRT or ONNX.
1.4 Why AI Adoption Accelerated: Data, Compute & Algorithms
Why does inference frequently use INT8 or FP8 quantization on NVIDIA GPUs?
Find five hidden terms in a grid of letters.
Term Search
What is supervised fine-tuning (SFT) of a foundation model?
Direct GPU-to-GPU interconnect inside a server that scales multi-GPU input and output.
AI Infrastructure
What is the goal of retrieval-augmented generation (RAG)?
Which of the following is a core principle of Responsible AI as articulated by NVIDIA?
NVIDIA's complete server systems, purpose-built for AI workloads from analytics to training and inference.
1.5 Key AI Use Cases Across Industries
What is the main role of cuDNN in a typical PyTorch or TensorFlow training pipeline on NVIDIA GPUs?
Tech Exam Prep Podcast
Listen on Spotify
Podcast
Which NVIDIA component is described as "a high-throughput, low-latency runtime for serving multiple AI models from any framework on a single GPU server"?
Chip class, such as NVIDIA BlueField, that offloads networking, storage and security services from a server's host CPU.
Which of the following best describes the role of the NVIDIA Container Toolkit on a GPU host?
What is NGC (NVIDIA GPU Cloud)?
NVIDIA's Arm-based server processor whose two-chip Superchip module carries 144 Neoverse V2 cores.
Chapter 2: GPU Architecture & CUDA Fundamentals
2.1 GPU vs. CPU Architecture & Parallel Execution
What does NVIDIA AI Workbench enable a developer to do?
Which of the following best describes a foundation model?
NVIDIA baseboard that holds eight Rubin, Blackwell or Blackwell Ultra SXM modules.
Loading more…