GPU for Beginners
Learn how GPUs work and why they are used in AI workloads, starting from the fundamentals.
📋 Recommended background: Basic understanding of computers, processors, and memory
🗒️ Learning path: GPU fundamentals, memory, and AI workload concepts
🎯 Focus: How GPUs work and support AI training and inference
Welcome to GPU for Beginners!
Learn how graphics processing units work, how they differ from CPUs, and why they are useful for many AI workloads.
This course introduces GPU architecture, memory, CUDA, and performance concepts. You’ll then explore how these ideas connect to model training, inference, and language-model workloads.
The lessons cover general GPU concepts and introduce NVIDIA-specific terminology where relevant.
What You’ll Learn
- GPU fundamentals: Why GPUs exist, how they differ from CPUs, and which workloads suit parallel processing.
- Architecture and execution: Streaming multiprocessors, threads, warps, CUDA, and matrix multiplication.
- Memory and connectivity: VRAM, memory bandwidth, PCIe, and NVLink.
- Numerical precision: FP32, FP16, BF16, and lower-precision formats used in AI.
- AI workloads: Training, inference, model size, tokens, context windows, and KV cache.
- Performance: Batch size, throughput, latency, and GPU utilization.
Who Is This Course For?
This course is for developers, DevOps engineers, students, and anyone starting to explore AI infrastructure.
If terms such as VRAM, CUDA, or TFLOPS feel unfamiliar, this learning path will help you build a foundation before exploring more advanced topics.
How to Get Started
Begin with Module 1: GPU Fundamentals, then continue to the memory, performance, and AI terminology lessons.
Focus on understanding how the concepts connect.
For example, memory capacity determines how much data fits, while bandwidth describes how quickly data can move.