search

tdlukulu / CUDA-CFD

CUDA accelerated Computational Fluid Dynamics
2 stars 0 forks source link

Use of single/double precision #3

Open GoogleCodeExporter opened 9 years ago

GoogleCodeExporter commented 9 years ago 
Currently single precision is used for the solution of the linear system. 
Single and double precision use should be compared in terms of

1. Solution accuracy.
2. Convergence of nonlinear iterations.
3. Speed.

This should be done for both CPU and GPU.

Original issue reported on code.google.com by cuneytsert on 7 Apr 2012 at 10:27

GoogleCodeExporter commented 9 years ago 
[deleted comment]
GoogleCodeExporter commented 9 years ago 
[deleted comment]
GoogleCodeExporter commented 9 years ago 
[deleted comment]
GoogleCodeExporter commented 9 years ago 
[deleted comment]
GoogleCodeExporter commented 9 years ago 
For the GPU part nvcc must be used with
-arch=sm_13
parameter and without
-DSINGLE
parameter. The code compiles and runs without problems. Tests should be done.

Original comment by cuneytsert on 12 Oct 2012 at 12:37

GoogleCodeExporter commented 9 years ago 
With the tests I did for the Poisson solver, for large meshes single precision 
may give unexpected results and should be avoided.

More systematic tests are necessary.

Original comment by cuneytsert on 8 Dec 2012 at 1:47