GPU Computing with CUDA Introduction | Lecture Note

University:
Boston University
Course:
GRS CS 910 | Directed Study: Systems
Academic year:

2020
Views:

418

Pages:

19
Author:

shakira4p0mo

GPU Computing with CUDA Lecture 1 - Introduction General Ideas ‣ Objectives - Learn CUDA - Recognize CUDA friendly algorithms and practices ‣ Requirements - C/C++ (hopefully) ‣ Lab oriented course! Outline ‣ Understanding the need of multicore architectures ‣ Overview of the GPU hardware ‣ Introduction to CUDA - Main features - Thread hierarchy - Simple example - Concepts behind a CUDA friendly code Moore’s Law ‣ Transistor count of integrated circuits doubles every two years Getting performance ‣ We have more and more transistors, then? ‣ How can we get more performance? - Increase processor speed ‣ Have gone from MHz to GHz in the last 30 years ‣ Power scales as frequency3! ‣ Memory needs to catch up - Parallel executing ‣ Concurrency, multithreading Getting performance ‣ We have more and more transistors, then? ‣ How can we get more performance? - Increase processor speed ‣ Have gone from MHz to GHz in the last 30 years ‣ Power scales as frequency3! ‣ Memory needs to catch up - Parallel executing ‣ Concurrency, multithreading Getting performance ‣ We have more and more transistors, then? ‣ How can we get more performance? - Increase processor speed ‣ Have gone from MHz to GHz in the last 30 years ‣ Power scales as frequency3! ‣ Memory needs to catch up - Parallel executing ‣ Concurrency, multithreading Getting performance Moore’s Law ‣ Serial performance scaling reached its peak ‣ Processors are not getting faster, but wider ‣ Challenge: parallel thinking Graphic Processing Units (GPU) Graphic Processing Units (GPU) ‣ GPU is a hardware specially designed for highly parallel applications (graphics) !"#$%&'()*(+,%'-./0%1-,! ! Graphic Processing Units (GPU) ‣ Fast processing must come with high bandwidth! ! >17/'&()?)*( >@-#%1,7?6-1,%(A$&'#%1-,;($&'(B&0-,.(#,.( Graphic Processing Units (GPU) ‣ Graphics market is hungry for better and faster rendering ‣ Development is pushed by this HUGE industry - High quality product - Cheap! ‣ Proven to be a real alternative for scientific applications Top 500 list. June 2011 +'!-#&-!-#$!978!+'!'/$.+&2+:$,!1(%!.(;/<-$4+)-$)'+=$>!#+3#20!/&%&22$2! -#$!978!+'!-#&-!-#$!978!+'!'/$.+&2+:$,!1(%!.(;/<-$4+)-$)'+=$>!#+3#20!/ +()!!!$?&.-20!5#&-!3%&/#+.'!%$),$%+)3!+'!&*(<-!!!&),!-#$%$1(%$!,$'+3)$, .(;/<-&-+()!!!$?&.-20!5#&-!3%&/#+.'!%$),$%+)3!+'!&*(<-!!!&),!-#$%$1(% ;(%$!-%&)'+'-(%'!&%$!,$=(-$,!-(!,&-&!/%(.$''+)3!%&-#$%!-#&)!,&-&!.&.#+) '<.#!-#&-!;(%$!-%&)'+'-(%'!&%$!,$=(-$,!-(!,&-&!/%(.$''+)3!%&-#$%!-#&)!, GPU chip design .()-%(2>!&'!'.#$;&-+.&220!+22<'-%&-$,!*0!@+3<%$!A4BC! &),!12(5!.()-%(2>!&'!'.#$;&-+.&220!+22<'-%&-$,!*0!@+3<%$!A4BC! ! ‣ CPU vs GPU "*+,-*.!'()! '()! '()! '()! '()! '()! '()! '()! !! !! !! !! !! !! !! !! "#$%&! /0'1! /0'1! "2)! "2)! CPU ! ! ! ! ! ! ! ! !! !! !! !! !! !! !! !! /0'1! 32)! GPU 32)! GPU devotes more transistors to data processing ! )?@*( A"&(852(3&B-%&:(C-'&(A'#,:1:%-':(%-(3#%#( >16/'&()?@*( A"&(852(3&B-%&:(C-'&(A'#,:1:%-':(%-(3 GPU chip design GPU chip design ‣ A glance at the GeForce 8800 Host Streaming multiprocessor Input Assembler Streaming processor Thread Execution Manager Parallel Data Cache Parallel Data Cache Parallel Data Cache Parallel Data Cache Parallel Data Cache Parallel Data Cache Parallel Data Cache Parallel Data Cache Texture Texture Texture Texture Texture Texture Texture Texture Texture Load/store Load/store Load/store Load/store Global Memory Load/store Load/store GPU chip design 3 billion transistors 512 CUDA cores (32 in 16 SMs) 64kB of on chip RAM High bandwidth ‣ Cores? SP SM GPU chip design ‣ The GPU core is the stream processor ‣ Stream processors are grouped in Stream Multiprocessors - SM is basically a SIMD processor (Single Instruction Multiple Data) GPU chip design ‣ Core ideas - GPUs consist of many “simple” cores ‣ Designed for many simpler tasks: high throughput ‣ Fewer control units: latency Maximize throughput of all threads Minimize latency of a thread