10 bucks

Share with us how are you using the powerrrr of the force

Moderator: Alain

10 bucks

Postby Bird » Sun Jan 04, 2015 11:23 pm

Newton flawlessly handling an attempt to stuff 1000 pennies into a jar. Newton took about 3 hours to solve the simulation and Lightwave took about 20 hours to render out the 20 second animation.

https://www.youtube.com/watch?v=jjDZKQiAU-M

-Bird
Bird
 
Posts: 623
Joined: Tue Nov 22, 2011 1:27 am

Re: 10 bucks

Postby Sweenie » Mon Jan 05, 2015 3:45 am

Awesome :D
Seems you will need a bigger jar though. :lol:
Sweenie
 
Posts: 498
Joined: Mon Jan 24, 2005 7:59 am
Location: Sweden

Re: 10 bucks

Postby manny » Mon Jan 05, 2015 10:52 am

Very nice!
http://www.instaLOD.io - InstaLOD - State of the art 3D optimization
manny
Site Admin
Site Admin
 
Posts: 131
Joined: Tue Feb 11, 2014 6:49 pm

Re: 10 bucks

Postby Julio Jerez » Mon Jan 05, 2015 3:40 pm

woow, very impressive.

It makes one wonder if doing the GPU thong will help accelerating the simulation.
I started and I have neglected for a while now after the big disappointment of the AMP.

I now recognized that this is nor AMP, it is a the way the PC architecture is designed.
so the bother neck is there regardless of the language.

But for things like this a GPU implementation can easy make anywhere for 10 to 100 time faster.
Basically the cost will be all the data coping with will be Fixed, while on the CPU the cost all go to the CPU collision and physics simulation.

Did you run that multithreaded or single threaded?
Julio Jerez
Moderator
Moderator
 
Posts: 12249
Joined: Sun Sep 14, 2003 2:18 pm
Location: Los Angeles

Re: 10 bucks

Postby Bird » Mon Jan 05, 2015 6:39 pm

woow, very impressive
Very nice!
Awesome

Thanks! I've made a few of these in the past and this came out the best so far I think.

Did you run that multithreaded or single threaded?

Multithreaded. I am on a quad core laptop and I have set NewtonSetThreadsCount to 4. Although while Newton was updating, I noticed in Task Manager that CPU utilization was less than 50% throughout the whole simulation.

In the past I haven't noticed Newton taking so long to solve. I think that the new slowdown is when a dynamic body is inside a Static mesh collision tree. I've noticed that in other scenes too in more recent versions of Newton. I'll investigate when I get a chance and start a new thread elsewhere.

But for things like this a GPU implementation can easy make anywhere for 10 to 100 time faster.


That would be great of course :)

Thanks
-Bird
Bird
 
Posts: 623
Joined: Tue Nov 22, 2011 1:27 am

Re: 10 bucks

Postby JoeJ » Fri Jan 09, 2015 6:29 am

Julio Jerez wrote:I started and I have neglected for a while now after the big disappointment of the AMP


Some more lessons i've learned the last months:
Beside the problem of slow data transfer there is also the problem of slow shader invocation.
I have a dynamic tree rebuild that needs to invoke tasks per tree level, so i can't avoid to start GPU threads multiple times. GPU work is <1ms in summary, but it lasts >5ms to complete.
I hope multiple command queues can fix that, or OpenCL 2.0 which can start work by GPU itself.
This seems again an GPU<->CPU issue only - APUs are much better here.

But the more important lesson:
Changing 5 lines of C++ code often means changing 100 lines of GPU code.
And changing the entire algo because you have a cool idea that is 10 times faster means:
Trash all GPU code and begin from scratch :( - even if the basic idea is not that different.
User avatar
JoeJ
 
Posts: 1453
Joined: Tue Dec 21, 2010 6:18 pm


Return to User Gallery

Who is online

Users browsing this forum: No registered users and 7 guests