Services

Completed Projects

The table below outlines just a few of the many and varied projects that we have completed for clients and the results achieved. These results are typical, and illustrate the potential value of Acceleware's expertise.

INDUSTRY

APPLICATION

DELIVERABLES

HIGHLIGHTS

Computer Aided Engineering

SIMULIA Abaqus

  • Develop a GPU accelerated version
  • Conduct a finite-element analysis and develop a library to offload LDLT factorization portion of the multi-frontal solver to GPUs
  • Implemented 2-3x accelerated solution that supports NVIDIA and AMD GPUs

Finance

Option Pricing

  • Debug & optimize existing CUDA code
  • Implemented the Leisen-Reimer version of the binomial model for stock option pricing
  • Improved performance of 30-50x compared to single-threaded CPU code

Medical

CT Reconstruction Software

  • Develop a GPU accelerated application for image reconstruction on CT scanners and implement advanced features including job batch manager, filtering, and bad pixel corrections
  • Achieved 31x accelerated back projection

Mining

Particle Modelling

  • Multi-threaded a particle simulator that had many serial dependencies
  • Improved maintainability and testability of code
  • Improved performance by 2x for certain data sets

Oil & Gas

Reservoir Simulation

  • Develop iterative linear algebra solvers optimized for reservoir simulation on NVIDIA GPUs
  • Implement GMRES, QMR, AMG, and a variety of preconditioners
  • Demonstrated a 5x improvement on NVIDIA Tesla C1060 over multi-core Intel Nehalem CPUs

Oil & Gas

Controlled-Source Electromagnetic (CSEM) Modelling

  • Port existing Matlab code to C/C++ multi-threaded code
  • Saw signigicant performance improvements moving to multi-threaded C++
  • Performance gains of approximately 12.5x over single threaded Matlab

Oil & Gas

Seismic Algorithm

  • Demonstrate the viability of GPUs to accelerate seismic algorithms
  • Implemented more efficient summed area table algorithm
  • Demonstrated speed-ups of 50-150x on NVIDIA GPUs

Oil & Gas

Petrophysics

  • Add multi-threading support to existing C# petrophysics inversion code to optimize performance on multi-core CPUs
  • Study the feasibility of porting inversion algorithms to GPUs
  • Delivered performance improvements of 8x on an 8 core CPU, and 14x on a GPU

Oil & Gas

Seismic Analysis

  • Develop a high-performance CPU/GPU library that implements statistical data sampling algorithms
  • Develop a high-performance CPU/GPU library that builds upton the data sampling library and implement statistical algorithms for quantitative seismic interpretation
  • Found optimal CPU/GPU balance to deliver maximum performance

Oil & Gas

Controlled-Source Electromagnetic (CSEM) Modelling

  • Port key sparse linear algebra routines (QMR, BiCStab, and SpMV) to NVIDIA GPUs to improve performance
  • Refactor existing Fortran/MPI code base to improve maintainability
  • Implement dynamic load balancing to optimize performance on large clusters
  • Delivered 5x speedup on GPU

Oil & Gas

Seismic

  • Prove native C/C++ implementation of prototype Matlab code
  • Reduced the software's memory footprint to support simulations 2-3x larger
  • Implemented algorithmic and data structure improvements
  • Improved performance on portions of the code by 20-30x through algorithmic optimizations

Oil & Gas

Drilling Simulation

  • Develop high-performance forward modelling simulation code for analyzing wellbore data
  • Port existing Matlab code to native C/C++
  • Optimize code to take advantage of multi-core CPUs
  • Implement more effiecient data structures to optimize certain operations
  • Delivered 18x speedup on average over pre-existing customer code

Oil & Gas

Beam Migration CPU Optimization

  • Optimize CPU Beam migration code
  • Improved single threaded execution by a facor of 8-10x via CPU optimizations
  • Delivered overall node performance by a factor of 3x

Oil & Gas

Seismic Filters

  • Deliver optimized anisotropic diffusion, median, convolution, and coherent noise removal filters
  • Optimize existing CPU and CUDA filter code, and port additional filters to CUDA
  • Improved performance of CUDA code by 3x
  • Improved performance of CPU code by a factor of 10-50x via algorithmic optimizations
  • Customer quote: "It flies!"

Oil & Gas

Seismic - Kirchoff Depth Migration

  • Implement GPU version of depth migration
  • A single GPU delivered the same throughout as 4 dual-socket Xeon nodes

Oil & Gas

Seismic - Full Waveform Inversion

  • Port existing CPU FWI code to run on GPUs via OpenCL
  • Achieved 8.4x speedup using 2 GPUs vs. a 30 core CPU system
  • Customer quote: "I appreciate the time and effort everyone spent on the project. It is good to see a company with similar work ethics as GeoTomo. It has been a pleasure working with your team"

Oil & Gas

Passive Seismic Emission Tomography (PSET)

  • Optimize existing GPU code
  • Improved performance over existing GPU code by 2-3x

Oil & Gas

Controlled-Source Electromagnetic (CSEM)

  • Refactor existing Matlab/Fortran code base into modern, modular, and maintainable Fortran code
  • Improved performance and significantly reduced memory footprint
  • Improved performance by a factor of 2.5x over original solution
  • Memory footprint reduced by up to a foctor of 3x

Oil & Gas

Seismic - Kirchoff Time Migration

  • Port seismic imaging code to CUDA to run on NVIDIA GPUs
  • Improved job runtime by a factor of 2.5x

Oil & Gas

Phase-shift Plus Interpolation (PSPI) Migration

  • Port seismic imaging code to CUDA to run on NVIDIA GPUs
  • Achieved 3x over original CPU code

Oil & Gas

Seismic - Depth Imaging

  • Optimize seismic imaging code
  • Reduced memory footprint and improved performance on multi-core X86 clusters

Oil & Gas

Full Waveform Inversion

  • Port an existing multi-threaded CPU FWI code to run on GPUs
  • Achieved a 15x speedup with 2 GPUs vs. a 30 core system

Oil & Gas

Controlled-Source Electromagnetic (CSEM)

  • Refactored Matlab/Fortran code into testable, modular, and maintainable Fortran code
  • Improved performance by 2.5x and reduced memory footprint by a factor of 3x

Oil & Gas

Finite Difference Forward Modelling

  • Advanced features for forward modelliing
  • Delivered improved data handling, cluster management, and visualization features

Security & Defense

Cryptography

  • Port & optimize a cryptography algorithm for FPGAs
  • Exceeded performance requirement 15x faster than expected

Security & Defense

Image Reconstruction

  • Ported & optimized a cryptography algorithm for FPGAs
  • 15x faster than expected performance requirements