High Performance Iterative Tomography Reconstructions on GPU and Intel Xeon Phi Coprocessor

Period of Performance: 01/01/2015 - 12/31/2015


Phase 1 STTR

Recipient Firm

33 Bowman Ln
Kings Park, NY 11754
Firm POC
Principal Investigator

Research Institution

Brookhaven National Laboratory
Building 817
Upton, NY 11973
Institution POC


Tomographic reconstructions with insufficient data, such as the projections scanned with inadequate angular range or contaminated with noise, are often confronted for transmission electron microscopy and full-field transmission X-ray microscopy. Iterative reconstructions can provide a viable solution by numerical optimization with a cost of intensive computational overhead. Although super computer clusters help to reduce computing time from days to hours, they are expensive, costly in energy and have limited CPU hours assigned to users. Recently emerging high performance computing (HPC) hardware, such as the general-purpose graphical processing unit (GPGPU) or Intel Xeon Phi coprocessor, provides a cost-efficient solution. However, the software tools that satisfy specific big data requirement for transmission microscopy and accommodate these hardware architectures are either outdated or rarely available. More over, it is unclear which devise to choose for different imaging modality and data scale. How this problem is being addressed: This STTR project is developing an open-source HPC software tool to solve intensive computation problem for the reconstruction in transmission X-ray microscopy and transmission electron microscopy. By porting existing iterative reconstruction algorithms to GPGPU/Intel many-core coprocessors and implementing new accommodated acceleration methods, we aim to enable near real time reconstruction with insufficiently scanned high-resolution images and suggest suitable devise for related computation problems. What is to be done in Phase I: In Phase I, we will (1) implement an open source high performance software toolkit of representative iterative tomographic reconstruction algorithms executing on both GPU and coprocessor, (2) perform benchmark testing for associated applications and make the architecture recommendation with consideration of budget plan and programming capability, and (3) engage research publications on designing and developing new acceleration algorithms on the aforementioned HPC architectures. Commercial applications and other benefits: These software tools can dramatically enable battery design for material science, drug design for biology research, and dosage control for clinical practice with a largely reduced budget and energy cost.