The Cornell
University Center
for Advanced Computing (CAC) announced that it is testing the performance of
general-purpose GPUs with MATLAB applications in a new research collaboration
with NVIDIA, Dell, and MathWorks.
This research will explore GPU computing capabilities for
data manipulation on NVIDIA GPUs using MATLAB applications. In particular,
Cornell will focus on the use of multiple GPUs on the desktop via the MathWorks
Parallel Computing Toolbox, and a GPU cluster via MATLAB Distributed Computing
Server.
Cornell is conducting this
research on Dell C6100 servers with the C410x PCIe expansion chassis, which
supports server connections to NVIDIA Tesla M2070 GPUs.
"The launch of this GPU
capability with eight nodes (each with eight CPU cores) and eight NVIDIA Tesla
M2070 GPUs (each with 448 CUDA cores) is extremely valuable particularly for
researchers needing to process large blocks of data in parallel," says David
Lifka, Cornell CAC director.
For example, researchers from
Weill Cornell
Medical Center,
University of Michigan Health System, and Rutgers
Laboratory for Computational Imaging and Bioinformatics are currently using the
NVIDIA GPUs and MATLAB to accelerate and improve the diagnosis of cancer cells
using template matching. Using MATLAB's built-in GPU functions, the researchers
experienced a 14.7-times speedup in code processing time (from 86.9 sec to 5.9
sec). That’s a significant improvement for pathologists who would like to
process many large scale images each day. By comparison, MATLAB code running on
GPUs performed 4.8-times faster than code that was implemented in C++ without
GPUs. And, because MATLAB is optimized for use with GPUs, users can take
advantage of the GPUs’ compute power without needing to learn another
programming language or leaving the MATLAB environment.
In another project, Theo
Damoulas, a research associate with the NSF-established Institute for
Computational Sustainability (ISC) directed by Prof. Carla Gomes, benefited
from a 12-times speedup in Dynamic Time Warping (DTW) computation by using a
combination of built-in MATLAB GPU functions and CUDA code. DTW is the
computationally expensive part of the code which uses machine learning and
signal analysis techniques to automatically identify bird species from their
flight calls. Automatic flight call classification is much faster and arguably
more accurate than manual classification, and the first step in creating large
scale networks of recording stations that can provide a detailed understanding
of the migration patterns of individual species.
This project is
representative of the research of the ISC, whose aim is to provide solutions
for balancing environmental, economic, and societal needs for a sustainable future
by bringing computational thinking to sustainability research. The ISC is a
joint venture involving scientists from Cornell
University, Bowdoin College,
the Conservation Fund, Howard University, Oregon State University, and the
Pacific Northwest National Laboratory.
"As GPU performance testing
and production runs continue at Cornell, CAC will be developing lessons learned
and best practices for porting MATLAB code in order to improve the overall
experience with MATLAB GPU computing for scientific researchers," notes Lifka.
Cornell previously deployed a
National Science Foundation-sponsored 512-core experimental MATLAB resource for
the research community in partnership with Purdue University
to provide a bridge to high-end national resources. Over 550,000 jobs ran on
the experimental resource which facilitated research, student learning, and
Science Gateway applications.
Cornell University,
www.cornell.edu
