How to interpret hardware usage?


Hi all and happy new year!
Apologies if this question has already been asked, but I didn’t find a relevant discussion here.

I’m currently working on a project involving fluid simulation and the Leap Motion sensor, running on my laptop equipped with a Nvidia GTX 1050 Ti and an Intel i7-8700. After a certain amount of particles, the framerate drops significantly. If that makes any sense, I tried to set all the nodes with the same resolution and data format, to prevent conversions, and with the preview disabled. Also, opening the performance monitor I can see that the Nvidia FlexSolver comp is taking a lot of time to cook, ~100ms per frame. That’s nothing special, though, and of course explains the low framerate.

The weird thing (to me) is that when opening the Windows task manager I notice that neither the CPU nor the dedicated GPU in any of its parameters seem to be “exploited” to the extreme. What I can’t understand is how to justify that low framerate with such a minimal usage of the hardware. Am I missing any setting? Is there any bottleneck hidden to my inexperienced eye, or some technical explanation?

Just to clarify, I am running TD on a fresh Windows installation, completely debloated and optimised. Also, I made sure that TD runs on the dedicated GPU.

Any help would be much appreciated, thank you!

Would you be able to post your. toe file so we can look at this directly?

Hi @malcolm here is it
Debug.toe (17.1 KB)

I made some optimizations to the updating of collision shapes and transforms of static shapes in Flex which should significantly bring down the cook time of the Nvidia Flex Solver COMP in your case.

And to further optimize your project (with the aforementioned changes), I’d recommend changing actor2 so that instead of updating the transform via the collision shape/transform SOP, it’s done via the Xform page parameters on the Actor COMP itself. Updating the transform that way will have less overhead.

I can send you a build with the changes once one is ready.

Here’s a link to a build with the optimizations:

Let us know how that works for you.

Hi @eric.b thank you very much for this build. The time the flexsolver takes to calculate things is indeed massively lower. However I still have the same low framerate and resources consumption