Lect02-2010-GPUs - ,Knoxville StanTomov CS594LectureNotes...

Info iconThis preview shows pages 1–7. Sign up to view the full content.

View Full Document Right Arrow Icon
    1/18 HPC with Multicore and GPUs Stan Tomov Electrical Engineering and Computer Science Department University of Tennessee, Knoxville CS 594 Lecture Notes January 20, 2010
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
    2/18 Outline Introduction Hardware trends Challenges of using multicore+GPUs How to code for GPUs and multicore An approach that we will study Introduction to CUDA and the cs954  project/library Conclusions
Background image of page 2
    3/18 Speeding up Computer Simulations Better numerical  methods Exploit advances  in hardware http://www.cs.utk.edu/~tomov/cflow/  Manage to use hardware     efficiently for real-world      HPC  applications   Match LU benchmark in     performance ! e.g.  a posteriori error analysis :   solving for much less DOF but    achieving the same accuracy
Background image of page 3

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
    4/18 Why multicore and GPUs? Hardware trends                                       (Source: slide from Kathy Yelick)     (Source: “ NVIDIA's GT200: Inside a Parallel Processor ”)   Power  is the root cause of all this  Multicore  GPU Accelerators
Background image of page 4
    5/18 Main Issues Increase in parallelism * 1 How to code (programming model,                        language,                        productivity, etc.)?   Increase in commun.    * 2 cost (vs computation) How to redesign algorithms? Hybrid Computing         * 3 How to split and schedule the computation between hybrid hardware components? Despite issues,  high speedups  on HPC applications are reported using GPUs   (from NVIDIA CUDA Zone homepage) :   * 1 - A data-parallel approach that scales - Similar amount of efforts on using   CPUs  vs  GPUs by domain scientists   demonstrate the GPUs' potential Processor speed                   * 2 improves 59% / year but  memory bandwidth by 23% latency by 5.5% e.g., schedule small          * 3 non-parallelizable tasks on the CPU, and large and  paralelizable on the GPU
Background image of page 5

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
    6/18 Scene model Graphics pipelined      computation Final image streams   of data Repeated fast over and over:  e.g. TV refresh rate is 30 fps; limit is 60 fps   GPUs : excelling in graphics rendering   This type of computation:  Requires  enormous computational power  Allows for  high parallelism
Background image of page 6
Image of page 7
This is the end of the preview. Sign up to access the rest of the document.

This note was uploaded on 04/01/2010 for the course COMPUTER S cs202 taught by Professor Jiuhui during the Spring '08 term at 東京国際大学.

Page1 / 18

Lect02-2010-GPUs - ,Knoxville StanTomov CS594LectureNotes...

This preview shows document pages 1 - 7. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online