OpenPOWER Foundation Supercomputing Recap: Accelerating Innovation
-
Upload
openpowerorg -
Category
Technology
-
view
5.587 -
download
2
Transcript of OpenPOWER Foundation Supercomputing Recap: Accelerating Innovation
Acceleration Brought To You By
1. IBM Watson: 1.7x faster thanks to NVIDIA GPUs
Read More Here
2. Xilinx and IBM: Announced a four-year relationship
for data center and network function virtualization
collaboration based on Xilinx FPGA accelerators.
Read More Here
3. Accelerated Networking: Switch-IB 2 from Mellanox
and ExpEther Tech. 100GB/s speeds. Perfect for
POWER systems.
Read More Here
4. New OpenPOWER-Based Systems: E4 Computing
Engineering and Penguin Computing have new systems
based on OpenPOWER.
Read More Here
5. IBM Ports Key Applications: New Internet of Things, Spark,
Big Data and Cognitive Era applications ports.
Read More Here
OpenPOWER—the perfect platform for developers.
Expanded GPU services on SuperVessel: GPU
computing as-a-service capabilities for Caffe, Torch
and Theano.
Read More Here
Expanded FPGA Services on SuperVessel: Coherent
reconfigurable accelerators are now available to
developers via the cloud.
Read More Here
New Cluster at University of Texas at Austin: A
POWER8 accelerated cluster is now available to
academic researchers.
Read More Here
Oregon State University Expands OSUOSL:
Additional compute and memory capacity
for POWER8-based systems in the Open
Source Lab.
Read More Here
Find out how to use all of these here
CENTERS OF EXCELLENCE
POWER ACCELERATION AND DESIGN CENTERS
RESEARCH
USER GROUP
CLIENT CENTERS
INNOVATION CENTERS
Watson Explorer Health Care Annotator Presented by Peter Hofstee/Tim Kaldewey/Kubilay Atasu
Challenge: Show the different sequence of
word clouds from health care documents
processed without acceleration versus with
FPGA acceleration.
Result: The accelerated searches process
more documents than the non-accelerated
searches in the same amount of time.
Real world takeaway: Doctors can learn
about allergy risks and find connections
between historical lab results for new patients
faster with FPGAs.
Adverse Drug Reaction Prediction Presented by: Randy Swanberg/Minsik Cho/ Rajesh Bordawekar
Challenge: Use a logistic regression model
with technology from OpenPOWER, Spark and
NVIDIA on a dataset from 2011 to predict drug –
drug interactions (DDIs).
Result: Predicted 73% of DDIs discovered after 2011.
Real world takeaway: Using a GPU accelerator to
evaluate a similar dataset from 2015 can accurately
predict yet undiscovered DDIs.
Watson Retrieve and Rank Presented by: Tim Kaldewey/David Tam/David Wendy
Challenge: Compare response times of
generic browser queries against Wikipedia data
for accelerated and un-accelerated searches.
Result: With NVIDIA Accelerators Watson is able
to answer questions 1.64x faster than before.
Real world takeaway: Practical implementation of
Watson comes one step closer to being real-time.
Degrees of Social Separation Presented by: Randy Swanberg/Jan Rellermeyer
Challenge: Calculate the degree of separation
for every actor to Kevin Bacon in a database of
10,000 movies with SPARK social analytics.
Result: With SPARK adjusted to ‘spill’ to a flash
device rather than a hard disk drive, a 4x reduction
in DRAM requirements was achieved.
Real world takeaway: POWER8 plus CAPI flash
acceleration makes it more practical to use SPARK
with highly iterative and demanding workloads.