GPUaccelerated simulation of flow through porous medium

1
GPU-accelerated simulation of flow through porous
medium

• by Asbjørn Bydal

2
Introduction

• Goal
• Explore the use of GPUs to accelerate simulation
  of flow through porous medium.
• Motivation
• The GPU performs better than the CPU
• Higher arithmetic throughput.
• More power efficient.
• Cheaper in terms of flops per dollar.
• Contribution
• Knowledge of how a reservoir simulation can be
  implemented on a GPU.

3
Background

• Graphical Processing Unit (GPU)
• General Purpose Computing on GPU (GPGPU)
• Nvidia Compute Unified Device Archtechture (CUDA)
• Single Instruction Multiple Thread (SIMT)
• Reservoir simulation
• Injection of water into the reservoir
• Production of oil

4
Prototypes

• Lax-Friedrich finite volume scheme (simple)
• High resolution scheme (advanced, computationally
  demanding)
• Heterogeneous computing model

5
Performance testing

• Lax-Friedrich

• High resolution

6
Single and double precision comparison
7
Mixed precision
8
Heterogeneous computing model - tests
9
Conclusion

• GPUs provide lots of benefits in reservoir
  simulations
• Faster and cheaper
• The precision dilemma is possible to circumvent
  and will become less significant in future GPU
  hardware generations
• Mixed precision schemes and more support for
  double precision.
• Still aspects of reservoir simulations that needs
  to be explored
• Multicomponent flow
• Gravity
• Scaling of problem size and multi GPU setups