Jump to content
digitallysane

Nice research paper (Wavelet Turbulence For Fluid Simulation)

Recommended Posts

I can't stress enough that CPU usage is NOT related to performance. One can easily have more CPU usage and still be slower! Use wall clock timing tests to compare between using OpenMP and not using OpenMP.

Share this post


Link to post
Share on other sites

It's true, but we are talking of the same program with or without the openmp c++ flag. The only difference of with/without openmp cpu behavior graphic, are thats five small peeks when all cores work together, the rest is normal one core execution. The conclusion is clear, I think.

Share this post


Link to post
Share on other sites
I can't stress enough that CPU usage is NOT related to performance.

Somewhat unrelated, but using a GPU to accelerate computation really throws this assumption for a loop as well. Because all of the computation is offloaded to the GPU, the CPU sits idle (or near idle). In those cases, the machine appears to be doing nothing but yet gets the job done 20x faster. It is a little strange to see the first time :) In fact, with power-saving technologies enabled (Cool n'Quiet or SpeedStep), the CPU will sometimes even power down to a lower GHz level.

Share this post


Link to post
Share on other sites
Well, now what I really want is Jeff's MasterClass files (knew I shoulda gone to SIGGRAPH this year :) )

The raw version of the presentation (.hip file clips, powerpoint, etc) is now online on the Website with the other mastercourse materials.

http://www.sidefx.com/index.php?option=com...&Itemid=264

Note that you may still want to use the plugin created earlier - 9.5 Gas Field Vop isn't multithreaded and processes the curl noise once per vx, vy, and vz channels, so is around 12x slower than it could be.

- Jeff

Share this post


Link to post
Share on other sites
Note that you may still want to use the plugin created earlier - 9.5 Gas Field Vop isn't multithreaded and processes the curl noise once per vx, vy, and vz channels, so is around 12x slower than it could be.

Is this gonna change (soon)? It kinds of frighten me that fluids (and dynamics) which are notoriously slow generally speaking, are not multithreaded fully in Houdini.

I can understand that it might be hard to rewrite for multithreading older modules of Houdini, but for DOPS (which are new) this was part of the design consideration, right?

Dragos

Share this post


Link to post
Share on other sites
Is this gonna change (soon)? It kinds of frighten me that fluids (and dynamics) which are notoriously slow generally speaking, are not multithreaded fully in Houdini.

I can understand that it might be hard to rewrite for multithreading older modules of Houdini, but for DOPS (which are new) this was part of the design consideration, right?

Multithreading is something very important for Fluids, which is why a lot of it is multithreaded.

The specific case of the Gas Field Vop, however, was not multithreaded in time for the 9.5 release. It is multithreaded in the latest internal version, but the change is rather dangerous to put into 9.5.

Share this post


Link to post
Share on other sites
The raw version of the presentation (.hip file clips, powerpoint, etc) is now online on the Website with the other mastercourse materials.

http://www.sidefx.com/index.php?option=com...&Itemid=264

Note that you may still want to use the plugin created earlier - 9.5 Gas Field Vop isn't multithreaded and processes the curl noise once per vx, vy, and vz channels, so is around 12x slower than it could be.

- Jeff

nice thx for posting this Jeff

jason

Share this post


Link to post
Share on other sites

For what it's worth, the presentations from SIGGRAPH 2008 are now availble at Siggraph Encore. This particular one is $9.99, I think, and does a pretty good job of explaining the paper. Just search for "wavelet turbulence", or enter 2008, Papers, Jiggly Fluids in the various dropdown lists on the main page.

Share this post


Link to post
Share on other sites

this is an old but great thread. does houdini in its current build implement the same technology to get the same results as demoed here?

Edited by queglay

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×