4. • Blue Joule 120,000 Cores (1.4 Pflops)
• Blue Wonder 19,000 Cores
• 9PB of Disk Storage
• 15 PB of Tape Storage
• Big Data Analytics Cluster (288TB)
• At Rest
• Streaming
• Predictive
• Novel server technologies
• Open Power
• FPGAs
• Data Centric Architecture
• Intel Phi
Our Machines
5. Our People
• Engineers
• Chemists
• Life Scientists
• Mathematicians
• Software
Developers
• Data Scientists
7. Project Examples
• Engineering & Manufacturing
– Vehicle Design & Testing
– Consumer Electronics Design
– Consumer Packaged Goods Products and Packaging
• Environment
– Weather modeling
• Life Sciences
– Genomics for better crop yields
– Disease mapping
• Energy
– Advanced Battery Cell Design
– Efficient Well Head Oil extraction
• Financial Services
– Risk Management
– Service Modelling
• Transport
– Network simulation
8. 4 Forces of Change
Industrial
Engagement
• Funding models
demanding economic
impact
• Using existing
computational
modelling in industrial
workflows
• Better outcomes,
developed faster and
cheaper
• New engagement
models, delivery
disciplines, cadence
& language
DemocratisationPower
• Unprecedented scaling
of systems needed
• Availability and
Affordability
• Existing architectures
are not sustainable
• Moore’s Law
ending
• Emerging platforms and
new Architectures
• New languages,
compilers,
middleware and
applications
• Cloud
Big Data
• Everything will change
• 3 V’s
• Internet of Things
• Security standards
• Todays architectures will
be obsolete in 5 years
• Emerging platforms and
new Architectures
• New algorithms,
languages,
compilers,
middleware and
applications
• Open data is great – but
useless on its own
• Analytics is the key to
• Availability of skills will
remain a bottleneck
• Science and Knowledge
workers
• Build the Data Scientist in to
the Machine / Software
• New ways of interaction
• Speech
• Visualisation
• Mobile
• Emerging platforms and
new Architectures
• New algorithms,
languages,
compilers,
middleware and
applications
9. Chemistry
High Accuracy Chemistry
Simulation
Formulation in the Cloud
Engineering
Data Driven Multi Physics
and Multi Scale modelling
Acoustic modelling of
Large Scale CFD
Intelligent
manufacturing
Life Sciences
Agri-tech
Food Provenance and
security
Enabling Technologies
Machine Learning, Algorithms & Implementation for Data Centric Systems, HPC in the Cloud,
Visual Interactive Supercomputing, Ease of Use, Access , Uncertainty Quantification and Modelling
Research Areas:
10.
11. What is the ternary phase diagram for the
sds/c12e6/water mixture at 12% surfactant?
Determining…
High Throughput Chemistry
& Materials Simulation
• Automation
• Force-field Development
• Robust Simulation Analytics
• Customizable Simulation Code
• Uncertainty Quantification
Big Data for Chemistry &
Materials
• Strong Links to Experimental Facilities
• Data analytics
• Machine Learning
• Natural Language Processing
• Cognitive Computing
• Agreed Data Standards?
Chemistry & Materials Advisor
Data Centric Computing Infrastructures
14. A call to action!
Hartree is looking for strategic
engagement with industrial applications
of data and compute.
Merging of simulation and advanced
analytics are merging on to a single
compute platform.
How do these technologies affect your
business or sector. When do you
engage with these and how?
The Hartree Centre is developing capabilities in Big Data (See slide for subset of areas) and high throughput simulation (see slide for subset of areas) for the chemistry and materials sector. Combination of these two fields will give an integrated data analytics and simulation capability capable of addressing a number of real world applications. Our initial efforts are focused around computer aided formulation (fuels, lubricants & home and personal care products).
The deliverable we envisage will be a chemistry and materials advisor that enables non-expert users to ask questions relating to their specific area. The results will be determined from data where it exists or from simulation where data is not available. These ‘in-silico’ experiments are required to be robust, automated and reliable. This includes ensuring robust and accurate force fields that are parameterised to appropriately predict experimental behaviour. We are investigating a couple of candidate methods for this area.
Simulation results will be fed into the corpus of data for future advisor searches.
The results given back to the scientist will be a series of results each accompanied with a probability score on the reliability of the data
We are attempting to aid the scientist not replace them.
Addedsd IBM Research. Is another get traiing and skills in the emerging areas of data centric cognitive computing?