Upload
raania-naeem-khan
View
220
Download
0
Embed Size (px)
Citation preview
7/29/2019 D&I of GPU based Image Processing on CASE cluster
1/28
7/29/2019 D&I of GPU based Image Processing on CASE cluster
2/28
2
Presented By:
1. Usama Aziz E09-108
2. Raania Naeem Khan E09-083
Supervised by:
Dr. M.Shamim Baig
7/29/2019 D&I of GPU based Image Processing on CASE cluster
3/28
Introduction
Computing at a Glance
Need for GPU computing
History of GPU computing
Transition to GP-GPUs Utilizing GPUs for general purpose computing
Clusters
Our Cluster.
Cluster Specifications. Our Graphic Card
Cluster Working
Applications
Project Timeline
Content
7/29/2019 D&I of GPU based Image Processing on CASE cluster
4/28
4
Introduction
7/29/2019 D&I of GPU based Image Processing on CASE cluster
5/28
objectives
Designing & Implementing a General Purpose GPU based clustercomputing system comprising of 1 master & 8 computer nodes
capable of executing an image processing system in a cost effective
and power efficient manner.
7/29/2019 D&I of GPU based Image Processing on CASE cluster
6/28
1980-1999 2000-2005
Computing at a Glance
Processing powerincreased byincreasing clockrate
Desktop1MHz
Desktop1GHz
x 1000
Limitations (processing power):-IC fabrication-power-heat-rapidly approaching limit totransistor size
Era of Multicorecomputing
MulticoreCPUsavailable tocustomers
2005- onwards
Era of GPUComputing
7/29/2019 D&I of GPU based Image Processing on CASE cluster
7/28
Low power consumption
Considerable Low cost
Greater computational power
Why GPUs?
7/29/2019 D&I of GPU based Image Processing on CASE cluster
8/28
Motivation: Computational Power
7/29/2019 D&I of GPU based Image Processing on CASE cluster
9/28
GPU-Computing: Concepts and Constraints
Conventionally GPUs produced a color against every pixelusing input colors, spatial and texture coordinates
Input colors could represent any data hence GPUs
could be tricked to perform multiple tasks.
GPUs usage as general purpose processor requiredgraphics-only languages
Nov 2006 NVIDIA first GPU GeForce 8800 GTX onCUDA Architecture was launched to addressconstraints.
7/29/2019 D&I of GPU based Image Processing on CASE cluster
10/28
Compute Unified DeviceArchitecture :CUDA
Allows every ALU to be accessed while performinggeneral purpose task
Capable of handling floating point operations
Introduction of CUDA C (2007)harnessed features of
the CUDA architecture
7/29/2019 D&I of GPU based Image Processing on CASE cluster
11/28
Overview of GPU computing
3D Graphic
concepts in
cinematography
by Silicon
Graphics
2D Graphics
Rise in Poplularityamong
consumers due to
incorporation in
Windows
1992- OpenGL
libraries to assist 3D graphicapplications
1995 onwards
-rise in demand of consumer
graphics games (Doom/Duke
Nukem)
-NVIDIA graphic accelerators
GeForce 3 series
Direct X 8.0standard (1st time
user controlled
GPU computations)
NVIDIA- Fermi
Architecture forGPUs 2DP
GFLOPs/Watt
Graphic Accelerators
Hardware assisted
bitmap operations to
assist display & usability
sold as part of CPUs and
speeding up graphics
NVIDIA GeForce
256
Transform & lighting
computations
directly on Graphics
Processor (part Open
GL graphics pipeline)
NVIDIA- Tesla
(2007)
Architecture for
GPUs < 2DP
GFLOPs/Watt
200920011991-20001980-1990
NVIDIA Kepler
(2011)
>5 DPGFLOPs/
Watt
7/29/2019 D&I of GPU based Image Processing on CASE cluster
12/28
12
Content Introduction
Computing at a Glance
Need for GPU computing
History of GPU computing
Transition to GP-GPUs
Utilizing GPUs for general purpose computing
Clusters.
Our Cluster.
Cluster Specifications.
Our Graphic Card
Cluster Working
Applications
Project Timeline
7/29/2019 D&I of GPU based Image Processing on CASE cluster
13/28
13
Group of computers and servers connectedtogether that act like a single system
Each system called a Node
Each Node contains one or more processor, RAM,HD, LAN card
Nodes operate in parallel. Performance can beincreased by adding more Nodes
Clusters
7/29/2019 D&I of GPU based Image Processing on CASE cluster
14/28
14
Our Cluster
7/29/2019 D&I of GPU based Image Processing on CASE cluster
15/28
15
Communication : Ethernet Switch 1GBSpecs:
-Master Node
Quad core i5 Processor 2.8 GHz
Ram 4GB
Hard disk 250GBGraphics Card NVIDIA GeForce GT210
-Node (1-8)
Quad core i5 Processor 2.8 GHz
Ram 4GB
Hard disk 250GB
Graphics Card NVIDIA GeForce GT210
-OS
Linux Cent-OS
-Cluster Management tool
Rocks(6.0) www.rocksclusters.org
Cluster specifications
7/29/2019 D&I of GPU based Image Processing on CASE cluster
16/28
GeForce GT 210 Specifications
Manufacturer: NVIDIASeries: GeForce GT 200GPU: GT218Core Clk: 489 MhzMemory Clk: 500 MhzFLOPS: 44.864 GFLOPSMax Power Draw: 30.5WattNoise Level: SilentMemory Type: DDR2Frame Buffer: 512, 1024 MBMemory Bus Type: 64bit
7/29/2019 D&I of GPU based Image Processing on CASE cluster
17/28
17
Working
Running Program(sequential)
7/29/2019 D&I of GPU based Image Processing on CASE cluster
18/28
18
Working
Running Program(sequential)
7/29/2019 D&I of GPU based Image Processing on CASE cluster
19/28
19
Working
Running Program(sequential)
7/29/2019 D&I of GPU based Image Processing on CASE cluster
20/28
20
Running Program(sequential)
7/29/2019 D&I of GPU based Image Processing on CASE cluster
21/28
21
Data sent
Data sent
Data sent
Running Program(Parallel)
7/29/2019 D&I of GPU based Image Processing on CASE cluster
22/28
22
Working
Working
Working
Working
Running Program(Parallel)
7/29/2019 D&I of GPU based Image Processing on CASE cluster
23/28
23
Finished
Finished
Finished
Results
Results
Results
Get results
Running Program(Parallel)
7/29/2019 D&I of GPU based Image Processing on CASE cluster
24/28
Applications of GPU computing
Digital Biology Molecular
Dynamics
Video Imaging Weather & OceanModelling
Forensic s
ScientificComputation
7/29/2019 D&I of GPU based Image Processing on CASE cluster
25/28
Introduction Computing at a Glance
Need for GPU computing
History of GPU computing
Transition to GP-GPUs Utilizing GPUs for general purpose computing
Clusters
Our Cluster.
Cluster Specifications.
Our Graphic Card
Cluster Working
Applications
Project Timeline
Content
7/29/2019 D&I of GPU based Image Processing on CASE cluster
26/28
Project Timeline
Equipment Acquisition
Cluster Establishment
Running Test App Parallelizing Test App
Implementing Algorithm.
7/29/2019 D&I of GPU based Image Processing on CASE cluster
27/28
References CUDA by example by Jason Sanders, Edward Kandrot
Programming Massively Parallel Processors by Kirk, Hwu
CUDA application design and development by Farber
GPU computing Gems by Hwu
Rocks user guide & Rolls base user guide
CUDA C getting started guide for Windows
Image browsing, processing and clustering for participatory sensing: Lessons froma diet sense prototype Reddy, Parker, et al.
Parallel Image Processing System on a cluster of personal computers Barbosa,Travares et al.
NVIDIA Fermi Whitepaper
www.rocksclusters.org
www.wikipedia.org
www.nvidia.com
http://www.rocksclusters.org/http://www.wikipedia.org/http://www.nvidia.com/http://www.nvidia.com/http://www.wikipedia.org/http://www.rocksclusters.org/7/29/2019 D&I of GPU based Image Processing on CASE cluster
28/28
28
Thank you for your time.