Softimage Nvidia Flex

[Pedrito]

Hi everyone, a long time xsi user new to the forum!

I'm developing a custom implementation of the nvidia flex gpu multisolver into ICE as a full nodal framework . Is a very low level plugin but fully integrated so you could just use your normal foces and velocity as if their where just normal points.
Take a look at my vimeo to see what is able of.
As a resume:
-Fluids
-Grains
-Cloth
-Custom Constraints
-Soft and RigidBodies(finishing the implementation)
-Ropes
-FULL ICE Integrated

All sim times in my videos are calculated without caching or capturing for benchmark purposes.

Stay tunned if you are interested, a beta release will be available when I finish polishing the code and creating a few toplevel compounds for the basic user.
I am going to upload some samples the next weeks that will be part of the relase as soon as possible!

Hope you like it!

Make Softi great Again!!!

[rray]

Hi Pedro,

Welcome to the forum! This is very cool, super exciting

I was just trying an older ICE NVidia PhysX integration last week, but couldn't get it work.

Most definitely staying tuned

[Pedrito]

Hi, Thanks!

Yes I also tried the ICE NVidia PhysX a few weeks ago and didn't manage to get it working, something probably changed in the SDK since the last working version? :S

The bad thing about this new nodal framework is that you can't decide to run on the CPU, so you are limited to the amount of GPU memory you have, in my samples with 2GB i'm able to 1M and a bit... in the office, with tow 1080ti I'm able to run more than 6M and would try to make a capture with the max I can, just for test (obviusly not at the same speed but...you know, lets force the machine!

[owei]

Hey Pedrito!

Looking forward to this! looks promising and thanks for sharing!

cheers,
oli

[wireframex]

Thanks for sharing this

[whurst]

Hi Pedro,

Welcome to the forum! This is very cool, super exciting

I was just trying an older ICE NVidia PhysX integration last week, but couldn't get it work.

Most definitely staying tuned

With my XSI 2012 it is working, but, you have to copy the DLL files into bin manually, than it is working.

Copy

PhysX3Common_x64.dll
PhysX3Cooking_x64.dll
PhysX3Gpu_x64.dll
PhysX3_x64.dll
PhysXCooking64.dll
PhysXCore64.dll
PhysXDevice64.dll
PhysXLoader64.dll
physxcudart64_20.dll

into

C:\Program Files\Autodesk\Softimage 2012 SP1\Application\bin

than it should work. It doesn't work in the User Directory, didn't know why



ANYWAY


I saw yesterday these videos on VIMEO and i hope (i serious hope) it will run on XSI 2012 ...

And welcome Pedrito

[Pedrito]

Hi Wolfgang thanks for the tips !
Probably that's the key, also I was trying it on 2015 SP1

And about running this on 2012... I don't have anyone now, but if I get acces to the SDK I think I can compile it and it should work, nothing fancy happening in the code I think. I will give it a change once finished!

[rray]

Good tips thanks. I will check some old old HDs for the 2012 SDK.

2GB->12GB should make a big difference, not that it looks really good aleady... still staring at these super low simulation times

[Pedrito]

Hi everyone again,

Sorry for the delay on the updates, I've been a bit busy at work and can't finish this

This weekend I''ve finished the soft body work (as in flex, no ICE powered stuff at all right now) and the rigidbody in a very early stage( just one rigid body per geometry). I am going to take a look at the "jj _island indexer" sources ( from Julian Johnson ->http://julianjohnsonsblog.blogspot.com ) and we will be able to create a rigid per island, nex step are rigid bodies from particle shapes

I am trying to implement a faster caching method caching directly from buffers, but I have never worked with c++ IO files and I am a bit.lost, even with your help Whurst, I will ask you a bit more in DM , thanks again for the help anyway!

I will try to post some cloth/softbody samples tonight!

Any thoughts or feelings please let me know! If you want to colaborate with me someway contact by DM and any help is always welcome and apreciated!!

Have a good week!

[Pedrito]

He there, some Softbody Emision, new videos today, I promise

[anhungxadieu]

Cool stuff, i really like the cloth one!

[mattmos]

I love what you're doing here, feels like Lagoa but with an extra speed boost! I would definitely be interested in taking the beta for a spin when you get to that stage.

[wireframex]

any news Pedrito

I saw next stage in FB ...

[Pedrito]

Hi, well I have not news at all, not so much updates, I have just been testing what I have right now and preparing some sample Scenes to test all the aspects, I hope at ending month I can have the first release.
I still have to perform a lot of security cheks, once everything is setted, is very stable, but, while preparing the setup, if you miss something, it may crash

Also I need a Tip on how to properly clear a pointer created by a c++ node once the node gets deleted ( because I can't get the SICALLBACK TERM to get fired at all, and I don't want to be cleared at the ENDEVALUATE as I want to persist until I delete the node) so right now a toogler called FLUSH need to manually be cheked and evaluated before closing or changin the scene to Clear the memory ( and that is very importatn because user maximun number of particles are allocated at the creation of the node.. Any thoughts?



I also prepared this scene, in wich I create a complex setup inside a pCloud representing various assets( rope,cloth and softBody), this need to be done inside a pCloud right now for contraining, but I'm working on a way to allow all this setup to be built even without creating a single PointCloud

Thanks for the interest! I try to keep all the places updates, but the first one is always VIMEO so to keep up to date, please follow my acount

[rray]

Sounds and looks great! Is the air around the parachute actually simulated? Would be so nice if something like this could be simulated:



No idea about the pointer question atm.

[Pedrito]

I dont know if Iunderstand what do you mean by simulated air in this case, but here we have gravity, mass, and the cloth has a special drag(flex built in) with takes care of the normals and apply forces a bit different.

Anyway, that scene could be simulated by adding a custom turbulent wind as force directly to the particles(as shown in the NON PROFILED TESTS in the small cloths sample), or creating a VDB velocity field( as shown in the vortex sample) and setting a few CROSS constraints inside the bag to simulate a fake volume constraint (if the bag is oppen), or if the bag is a clossed mesh, actually can have volume presservation!

I will try yo replicate that and add it to the bundle if you want