Transcript Diapositivo 1

Tools For Intensive Numerical Computation
GPU – Graphic Processing Unit
Jogo, Pedro; Oliveira, Ana; Silva, Rafaela;
Departamento de Engenharia Electrotécnica e Computadores/Instituto Superior Técnico
Introduction
Materials and Method
In recent decades the incidence of some diseases, such as cancer, where
In order to accelerate image registration, normally done with a CPU, the same was
diagnosis has a crucial role, has been increasing at an alarming rate. As a
tried with a GPU. The 3D data sets are “sliced” and then those obtained images suffer
consequence, the development of quick and accurate imaging techniques is not
registration. The algorithm was, obviously, translated to GPU “language”.
only an interest of the producers but also a society need. Advances have been
A
made in this field and now it is known that the use of a GPU to help the image
processing can not only make it faster but also more accurate.
B
Figure 8 – “Brain extraction” from a MRI
from two different subjects
Figure 7 – NVIDIA 9600 GX
Figure 1 – GeForce 6600 GT (NV43) GPU
Algorithm that seeks to find an
affine transformation that maps a
“source” volume onto a “target”
volume
Figure 2 – Incidence of cancer in UK
How does GPU work?
Figure 9 – Parameter affine
registration of A to B
Figure 3 – Compared to the CPU, the GPU Devotes More Transistors to Data Processing. The
GPU is especially well-suited to address problems that can be expressed as data-parallel
computations – the same program is executed on many data elements in parallel – with high
arithmetic intensity – the ratio of arithmetic operations to memory operations.
Figure 5 – FloatingPoint Operations
per Second for the
CPU and GPU
Results
The results are illustrated in the following table:
Experiment
12 Parameter
Affine
Registration
6 Parameter
Registration
CPU
8.5 minutes
GPU
6 seconds
Speed Up
98%
270 seconds
2.39 seconds
99%
With GPU, a 12 parameter affine registration took 6 seconds. Compared to the 8.5
Figure 6 – General
structure of a GPU
minutes normally taken with a CPU, it’s an amazing speed up.
It was also noticed that the quality of the image obtained with the GPU was better
than the one with the CPU.
Figure 4 – Serial code executes
on the host while parallel code
executes on the device.
Conclusion
It is clear that if this technique was applied to each diagnosis method where image
registration has a significant role, it would become not only faster but also better. This
Literature Cited
Ansorge, Richard E.: AIRWC: Accelerated Image Registration With CUDA. Cavendish Laboratory of
University of Cambridge. August 2008
could, consequently, lead to a previous diagnosis and reduce the increase of diseases
in which diagnosis has a crucial role.
Blas, Andrea Di; Lakdewey, Tim: Data Monster. September 2009
Acknowledgements
Luebke, David; Humphreys, Greg: How GPUs Work. NVIDIA Research and University of Virginia, 2007
We would like to thank to Rodrigo Ventura for all the provided information.
http://developer.download.nvidia.com/compute/cuda/2_3/toolkit/docs/NVIDIA_CUDA_Programming
_Guide_2.3.pdf
http://info.cancerresearchuk.org/prod_consump/groups/cr_common/@nre/@sta/documents/image/
crukmig_1000img-12614.jpg
Further Information
For further information, please contact [email protected], [email protected]
[email protected].
An
online
pdf
version
is
available
http://www.scribd.com/doc/24094085/.
or
at