GPU Computing





Latest Results

edit SideBar

GPU Computing

GPU Computing

June. 2014.

We have acquired PBRF funding to leverage recent advances in graphics processing hardware. Our two NVidia Tesla K40 GPU Accelerators are each equivalent to 2880 CUDA cores, with a peak double precision performance of 1.43Tflops.

We are developing code for solving quantum dynamics problems on GPU hardware to take advantage of the high performance for a certain class of dynamical problems.

Mandelbrot set

July 25, 2014.

First GPU calculations done by Matt Reeves to evaluate the Mandelbrot set, an embarrassingly parallel problem.

To test the Tesla K40 Matt evaluated the Mandelbrot set on an 8192^2 grid, with 1000 iterations at each point.

The timing comparison is:

CPU (C implementation): ~ 116 s , serial execution, Intel(R) Core(TM) i7-4770.

GPU (CUDA C): ~ 650 ms, optimized, Tesla K40.

Mandelbrot set on 8192^2 grid.

Numerical zoom-in on a region located on the lower branch, enclosed within the black square on the above figure.

Papers using GPU simulations

Spectral energy transport in two-dimensional quantum vortex dynamics,
T. P. Billam, M. T. Reeves, A. S. Bradley, [arXiv]

Identifying a superfluid Reynolds number via dynamical similarity,
M. T. Reeves, T. P. Billam, B. P. Anderson, A. S. Bradley, [arXiv]

Recent Changes (All) | Edit SideBar Page last modified on November 27, 2014, at 12:28 PM Edit Page | Page History
Powered by PmWiki