GPU PROCESSING
Group plans to revisit this decision based on
performance and user feedback. For other
GPU programming tools, consider the Magma
project, led by the linear algebra research
groups at the University of Tennessee, UC
Berkeley and UC Denver, which aims to
develop a library similar to Lapack, but for
heterogeneous/hybrid architectures, starting
with the current multicore+GPU systems. In
addition, GPU VSIPL from the Georgia Tech
Research Institute is an implementation of
the VSIPL (Vector Signal Image Processing
Library) Core Lite Profile that targets GPUs
that support Cuda. Next is GPULib, a library
from Tech-X that executes vectorised maths
functions on Nvidia GPUs. It provides
bindings for development environments
including Matlab and IDL (from ITT Visual
Information Solutions).
Developers are also getting assistance
in creating Cuda code from firms such as
Acceleware. That company writes specialised
algorithms and solvers for selected industries,
Using algorithms developed for GPUs by Acceleware, Ansys has significantly improved the
at this time focusing on electromagnetic and
execution of some of its solvers.
antenna design, oil/gas exploration and general
matrix solvers with some activity in finances, Research has announced that Mathematica lead in what it feels is the inevitable shift of
biomedical and imaging. Acceleware creates 7 is a Cuda-accelerated release of the GPU capabilities into CPUs.
libraries – similar to traditional numerics software scheduled for this spring. Nvidia has One of the big names in HPC, IBM, is
libraries but at a much higher level – that ISVs developed a Matlab plug-in for 2D FFT calls, taking a more reserved attitude towards
then integrate into their programs. For instance, and AccelerEyes has created a GPU engine GPUs. Dave Turek, VP of Deep Computing,
explains Schneider, Ansys software makes a call for Matlab-based on Cuda called Jacket (see notes that the company currently has no
to his libraries to offload the work of a direct page 28), which uses a compile on-the-fly GPUs in its portfolio. ‘The use of GPUs is
solver, and it also provides software that handles system to allow GPU functions to run in very embryonic and we are proceeding at an
the calls and data transfers from the CPU to Matlab’s interpretive style. appropriate pace.’ He believes the industry
the GPU. For its LabView software, National has entered a period of evaluation that will
Besides many pure-play graphics processors, Instruments has announced prototype last between 18 to 24 months and there
Rapidmind is starting to support Nvidia and development systems. Ansys has adapted will be a gradual dissemination into more
AMD GPUs with its software-development two of its solvers in Ansys Mechanical to conventional segments.
environment. Here developers write code run with Cuda. And if these leading software GPU clusters will be slow in coming, insists
in C++ and the RapidMind platform then companies are migrating towards Cuda, many OCF’s David Yip. He notes that as customers
‘parallelises’ it across multiple cores. The more are sure to follow so their software can move from GPU boards to clusters, they will
application is then run on RapidMind’s remain competitive. need to restructure data from ‘coarse grain’
development platform, which dynamically data decomposition to ‘fine grain’ on a GPU
compiles those sections into a parallelised An embryonic market to run it on multiple GPUs in cluster nodes.
program object. Concerning future trends, consider that At the moment, he adds, nobody has really
Certain industries that rely heavily on Intel announced its Larrabee ‘many-core’ come up with a system of managing the
proprietary codes, such as oil/gas exploration x86 architecture last autumn, but the initial data decomposition, and this will slow the
and the financial industry, have teams of indications are that it is emphasising graphics adoption of GPU clusters.
programmers who can take advantage of acceleration. The first product based on A final thing to watch for is the release of
Cuda. However, the average scientist Larrabee is expected this year or next. Apple’s Mac OS X Ver. 10.6 Snow Leopard,
or engineer simply wants to be able to The established GPU companies aren’t which is expected for release this year. That
purchase software that is GPU-enabled. standing still. Nvidia has doubled its floating- OS extends support for modern hardware
Fortunately, ISVs are starting to see the point performance every 15 to 18 months and with OpenCL, which Apple says ‘lets any
enormous potential of GPU acceleration of expects to stay on that track. AMD’s Harrell application tap into the vast gigaflops of
their software and have either issued Cuda feels that with her company’s expertise in GPU computing power previously available
versions or are working on them. Wolfram both GPUs and multicore CPUs that it can only to graphics applications’.
32
SCIENTIFIC COMPUTING WORLD february/march 2009 www.scientific-computing.com
SCWfeb09 pp30-33 GPU.indd 32 4/2/09 14:04:11
Page 1  |  Page 2  |  Page 3  |  Page 4  |  Page 5  |  Page 6  |  Page 7  |  Page 8  |  Page 9  |  Page 10  |  Page 11  |  Page 12  |  Page 13  |  Page 14  |  Page 15  |  Page 16  |  Page 17  |  Page 18  |  Page 19  |  Page 20  |  Page 21  |  Page 22  |  Page 23  |  Page 24  |  Page 25  |  Page 26  |  Page 27  |  Page 28  |  Page 29  |  Page 30  |  Page 31  |  Page 32  |  Page 33  |  Page 34  |  Page 35  |  Page 36  |  Page 37  |  Page 38  |  Page 39  |  Page 40  |  Page 41  |  Page 42  |  Page 43  |  Page 44