NVIDIA GPU Computing - A Journey from PC Gaming to Deep Learning

Explore the evolution of NVIDIA GPU computing from PC gaming to deep learning in this Stanford University seminar. Discover how GPUs have transformed from powering gaming platforms to driving cutting-edge applications in data centers and supercomputers. Gain insights into the architectural advancements of NVIDIA GPUs, including the development of unified shaders, CUDA, and Tensor Cores. Learn about the impact of GPUs on deep neural networks, image processing, and autonomous vehicles. Understand the journey through various GPU architectures, from classic GPUs to the Volta GV100, and their applications in gaming, cloud computing, and supercomputing. Delve into the complexities of training deep neural networks and the role of GPUs in accelerating these processes. Examine the development of specialized hardware like the Drive PX2 for automotive applications, showcasing the versatility of GPU technology across multiple industries.

Introduction
Background about Nvidia
Gaming
Console Gaming
Cloud Gaming
Supercomputing
Tesla V100
How do we get here
The intent
Classic GPUs
Rendering
Numeric representations
Vertex fetch engine
Unified shaders
G80
throughput vs latency
CUDA
Fermi Architecture
Kepler Architecture
Pascal Architecture
GTX 1080 TI
Volta GV100
Tensor Core
Interconnect
Titan
Deep Neural Network
ImageNet
Models are Complex
Training
Tensor RT
Image Per Second
Automotive
SOCs
Drive PX2