Nvidia Fermi Architecture (pdf)

CSC596/687-D: Accelerate — GPU

by: burt rosenberg
at: university of miami
semester: spring 2019 (192)
time: W 11:15 AM – 12:05 PM
location: Ungar 330 G

News:

• Sign up on slack to csc-minicourses
• coming soon! templates for assignment in [repo]/class/proj1.
• The class advertisement.
• The class github

Assignments:

• Assignment 0
  • ssh to sickles
  • subversion
  • study the example programs and makefiles in [repo]/class
• Assignment 1
  • Implement vector addition and inner product
• Assignment 2
  • Implement: Batcher Even-Odd transposition network for sorting
  • Extra Credit: Implement Bitonic sort (see wikipedia )
• Assignment 3
  • Implement Prefix sums
• Assignment 4
  • Implement Fast Fourier Transform, Butterfly network.
• Assignment 5
  • Choose and study an algorithm or process and implement.

Syllabus:

• Wednesday only course.
• Seminar style — student involvement expected.
• An experiment in opening up the curriculum as a community by the UM community.
• Project based, with introduction to the architecture of GPU's, and algorithms for concurrent computation.
• Sign up on slack to csc-minicourses

Class Notes:

• Lecture 1
  • What is GPGPU and CUDA
  • Basics of the NVIDA architecture
  • Resources for the class
  • Demonstration and assignment
• Lecture 2, Cumulative sum
• Lecture 3, Bitonic Sort
• Lecture 4, Butterfly Networks

Textbook like stuff:

• (perhaps) Professional CUDA C Programming by Chen, Grossman and KcKerncher. There might be other books you would prefer.
• CUDA Toolkit Documentation
  • Programming Guide ← start here

References like stuff:

• NVIDIA Microarchtectures
  • Tesla Architecture
  • Fermi Architecture Whitepaper
  • Kepler Architecture Whitepaper
  • Maxwell Architecture
  • Pascal Architecture Whitepaper
  • Volta Architecture
  • Turing Architecture

This work is licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.

author: burton rosenberg
created: 15 jan 2019
update: 6 mar 2019