May 8, 2007Farid Harhad and Alaa Shams CS7080 Overview of the GPU Architecture CS7080 Final Class Project Supervised by: Dr. Elias Khalaf By: Farid Harhad.

May 8, 2007 Farid Harhad and Alaa ShamsCS7080

Overview of the GPU Architecture

CS7080 Final Class ProjectSupervised by: Dr. Elias Khalaf

By: Farid Harhad & Alaa Shams

May 8, 2007 Farid Harhad and Alaa ShamsCS7080

Outline

• Introduction

• GPU Architecture

• GPU programming– GPU programming model– Toolkit and language

• Sample Code

• Conclusion

May 8, 2007 Farid Harhad and Alaa ShamsCS7080

Introduction

• The GPU on commodity video cards has evolved into extremely flexible and powerful processor.

• GPUs are fast :– 3.0 GHz Pentium 4: 6 GFLOPs, 6 GB/Sec peak– 3.0 GHz dual-core Pentium 4: 24.6 GFLOPs– GeoForceFX 6800: 53 GFLOPs, 34 GB/Sec Peak – GeoForceFX 7800: 165 GFLOPs– 1066 MHz FSB Pentium Extreme Edition: 8.5 GB/s– ATI Radeo X850 XT Platinum Edition: 37.8 GB/s

• GPUS are getting faster and faster– CPUs: ~1.5x annual growth ~60x per decade– GPUs: ~2.3x annual growth ~1000x per decade

May 8, 2007 Farid Harhad and Alaa ShamsCS7080

Computational power

May 8, 2007 Farid Harhad and Alaa ShamsCS7080

Cont.

May 8, 2007 Farid Harhad and Alaa ShamsCS7080

Why are GPUsgetting faster so fast?

• Arithmetic intensity: the specialized nature of GPUs makes it easier to use additional transistors for computation not cache

• Economics: multi-billion dollar video game market is a pressure cooker that drives innovation

May 8, 2007 Farid Harhad and Alaa ShamsCS7080

Flexible and Precise

• Modern GPUs are deeply Programmable– Programmable pixel, vertex, video engines

– Solidifying high-level language support

• Modern GPUs support high precision– 32 bit floating point throughout the pipeline

– High enough for many (not all) applications

May 8, 2007 Farid Harhad and Alaa ShamsCS7080

The Potential of GPU

• In short:– The power and flexibility of GPUs makes

them an attractive platform for general- purpose computation

– Example applications range from in-game physics simulation to conventional Computational science

– Goal: make the inexpensive power of the GPU available to developers as a sort of computational coprocessor

May 8, 2007 Farid Harhad and Alaa ShamsCS7080

The Problem: Difficult To Use

• GPUs designed for & driven by video games– Programming model unusual– Programming idioms tied to computer graphics– Programming environment tightly constrained

• Underlying architectures are:– Inherently parallel– Rapidly evolving (even in basic feature set!)– Largely secret

• Can’t simply “port” CPU code!

May 8, 2007 Farid Harhad and Alaa ShamsCS7080

GPU ArchitectureGraphic PL

May 8, 2007 Farid Harhad and Alaa ShamsCS7080

Modern Graphic PL

May 8, 2007 Farid Harhad and Alaa ShamsCS7080

Transform

• Vertex processor

(multiple in parallel)– Transform from “world space” to “image space”

– Compute per-vertex lighting

May 8, 2007 Farid Harhad and Alaa ShamsCS7080

Rasterizer

– Convert geometric rep. (vertex) to image rep. (fragment)• Fragment = image fragment

– Pixel + associated data: color, depth, stencil, etc.

– Interpolate per-vertex quantities across pixels

May 8, 2007 Farid Harhad and Alaa ShamsCS7080

Shade

• Fragment processors

(multiple in parallel)– Compute a color for each pixel

– Optionally read colors from textures (images)

May 8, 2007 Farid Harhad and Alaa ShamsCS7080

GPU programming

May 8, 2007 Farid Harhad and Alaa ShamsCS7080

GPU Programming Model

• Useful analogies:– Rasterization = Kernel Invocation– Texture coordinates = Computation domain– Vertex coordinates = computational range

• Invoking computation amounts to drawing pixels:– GPGPU invocation is commonly a full-screen quad

GPU CPU

Stream / Data array:• Memory read

Texture:• Texture sampling

Loop body / Kernel / Algorithm Fragment program

Feedback: Array write Feedback: render a texture

May 8, 2007 Farid Harhad and Alaa ShamsCS7080

GPU Programming Model

May 8, 2007 Farid Harhad and Alaa ShamsCS7080

Toolkits and Language

• High level shading languages– Cg: C for Graphics– HLSL: The D3D Shading Language– The OpenGL Shading Language

• GPGPU Languages– Sh - University of Waterloo– Brook - Stanford University

• CUDA SDK– Includes a C compiler and many libraries

May 8, 2007 Farid Harhad and Alaa ShamsCS7080

Sample Code

May 8, 2007 Farid Harhad and Alaa ShamsCS7080

Conclusion

• GPU provide the programmer with unparalleled flexibility and performance

in a product line that spans the entire PC market.

• Utilizing the capabilities of the GPU allow the programmers to develop newer applications-either graphical or general purpose-in more efficient way.

References

• GPU Gem2 (Chapters 29 & 30)

• SIGGRAPH 2005 GPGPU Course

• http://www.gpgpu.org/

May 8, 2007 Farid Harhad and Alaa ShamsCS7080

May 8, 2007 Farid Harhad and Alaa ShamsCS7080

Questions?

Thanks