View Single Post
Old 03-26-2015, 04:01 PM   #20 (permalink)
aerohead
Master EcoModder
 
aerohead's Avatar
 
Join Date: Jan 2008
Location: Sanger,Texas,U.S.A.
Posts: 15,879
Thanks: 23,955
Thanked 7,219 Times in 4,646 Posts
PC

Quote:
Originally Posted by jamesqf View Post
Or on a PC if you spend a few hundred on a GPU, and do some programming :-)
As I understand it,DNS can only be performed with a supercomputer,working in teraflops.
The mesh size and time steps are so vast that a PC cannot possibly handle the volume of information.
It is not a matter of a graphical processing unit.
The Full Navier-Stokes equation works in a spherical coordinate domain.It's millions of individual points in space,defined each by X-Y-and Z coordinates, which are interacting and solved for simultaneously in Eulerian differential calculus.All turbulence must be modeled,something only DNS can perform.
There is a RISC computing approach involving perhaps dozens of parallel computers operating together,and their cost,for DNS capability,as of 1992 was over $100,000,000.
DNS is quite old.It's only until recently that computers attained the speed and capacity to take on DNS.
Daimler-Benz' supercomputer requires over 48-hours continuous run time to analyze a single car.
The other thing about PCs is that,we need a digitized full-scale car with every,and I mean every small detail,down to the fraction of a millimeter.
I know of only Southwest Research Institute,in San Antonio,Texas which has the capability to do this.I've no idea what the cost would be,or if one could even gain access to their facility.
No doubt,Moore's Law will,over time,bring PC capacity up to otherwise 'impossible' levels,but I don't want to predict when they'd attain 'super' status.
__________________
Photobucket album: http://s1271.photobucket.com/albums/jj622/aerohead2/
  Reply With Quote
The Following User Says Thank You to aerohead For This Useful Post:
BamZipPow (03-27-2015)