Title:Performance Improvement
Posted By:JanBing ( Jan Hofmeister )
Date Created:7 October 2015

I know, it's a bit unspecific to ask for improved performance, but I've just tried Nvidia's realtime flex demos, and pretty much all of them run a lot faster than what Lucid can do - which is especially apparent with fluids. Is 3ds max slowing down the calculations? Or are Nvidia's own demos just optimized much better?

Follow Ups

There are a few things at play here:

  • Nvidia can do things which we can't. Namely, they keep their assets on VRAM (GPU) so memory never has to travel between CPU and GPU. In Lucid it does because Max runs and processes things on CPU.
  • We also have to copy things into a format which 3dsmax supports, be it meshes, particles, or other geometry and then back to GPU
  • Displaying things in Max is slower because we can't "cheat" like the Flex demos do and use screen buffer to fake meshing particles
  • We do more calculations in Lucid. We have to add/modify forces, copy velocities back and forth, copy and assign parameters, UV coords, and so on. All of this chips away at performance.
  • Last but not least, we are currently focusing on stability and usability more than speed. I think once we get over the usability hump we will nail down and optimize performance in many areas.

In short, you can't expect it to be as fast as Flex demos for reasons above but it will be fast enough for everyone to enjoy I hope.

Marsel Khadiyev (Software Developer, EPHERE Inc.)

The speed is already quite impressive to me!

Speed IS amazing, I agree. And now I understand why the demos are faster, thanks for the rundown :)



Would distributed networrk simulation be possible with Flex?  


With ten network gpu's churning out one simulation you could push the particle count to tens of millions...  



I haven't heard anywhere of a possibility of distributing Flex animations over a network but I suspect that may not be possible. Since the main advantage is storing things in VRAM and using GPU to parallel process all this data one does have to simulate on the same machine. Maybe a technology like SLI could help double the particle count, however.

Marsel Khadiyev (Software Developer, EPHERE Inc.)