FSI won't work

[Rebeca Guimarães]

Good afternoon!

I've donwloaded and compiled Chrono and been able to run many projects. Since I'm running terramechanics analysis, I am now trying to run FSI SPH.

I was able to code, configure and geenrate, but when I run the exe, it will halt at the Initialize and close the cmd.

I think I've narrowed the issue down to CUDA. I'm using version 13.0.1. I read at the website I need to use 12.3, but my MSVC says it expects version 12.4 or newer.

Can someone help me how to downgrade my MSVC so it will accept 12.3?

Thanks!

D:\Program Files\Microsoft Visual
Studio\2022\Community\VC\Tools\MSVC\14.44.35207\include\yvals_core.h(902):
error : static assertion failed with "error STL1002: Unexpected compiler
version, expected CUDA 12.4 or newer."

[Dan Negrut]

Rebeca – I don’t know if you are interested in this, but there is PyChrono support for this, see instructions here: https://uwmadison.box.com/s/lbxgqv1qu4zd25y8naq9fkzeenmf0ri6 .

Then you wouldn’t have to downgrade MSVC and be up and running quickly.

Dan

---------------------------------------------

Bernard A. and Frances M. Weideman Professor

NVIDIA CUDA Fellow

Department of Mechanical Engineering

Department of Computer Science

University of Wisconsin - Madison

4150ME, 1513 University Avenue

Madison, WI 53706-1572

608 772 0914

http://sbel.wisc.edu/

http://projectchrono.org/

---------------------------------------------

--
You received this message because you are subscribed to the Google Groups "ProjectChrono" group.
To unsubscribe from this group and stop receiving emails from it, send an email to projectchron...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/projectchrono/b3ff10c6-ac6c-42f7-bcb5-af0b30085b6fn%40googlegroups.com.

[Radu Serban]

Hi Rebeca,

 

Could you please clarify what Chrono version you are using?  Older releases will obviously not work with new CUDA releases.

 

The current Chrono code in the main branch of the repository should work fine with any new CUDA 12.* versions (e.g. 12.8 or 12.9), as well as CUDA 13.0 (although I only tested the latter on Linux).  Any of these CUDA versions will work fine with the latest VS versions.

 

Please make sure that:

• you use the latest Chrono code from the development branch (‘main’).
• Update your copy of Visual Studio to the latest version.
• You can test with CUDA 13.0.1.  If that doesn’t work, please let me know.  But if you can install a different version of CUDA, I suggest you use 12.9 which is what I currently use on my Windows machine.

 

--Radu

 

 

From: projec...@googlegroups.com <projec...@googlegroups.com> On Behalf Of Rebeca Guimarães
Sent: Thursday, September 18, 2025 10:45 PM
To: ProjectChrono <projec...@googlegroups.com>
Subject: [chrono] FSI won't work

 

Good afternoon!

--

[Rebeca Guimarães]

Hello, everyone!

Thank you for the suggestion, Dan! I will definetely try out PyChrono. I need to present a proof of concept to my undergrad thesis advisor in the next weeks so I had decided to go with C++ since I struggled a bit with Anaconda. But I'll try again :)

Radu, I've been using the latest version, I downloaded it from the git. I'm able to compile and build Chrono and run Vehicle module projects. 

The FSI single wheel demo that comes in the pre compiled pack works just fine in my computer. But whenever I try to compile a new project, even the demo cpp itself, it will run and then simply close the prompt before the visual interface ever starts. 

This has been happening with CUDA 12.3, CUDA 12.9, CUDA 13.0. 

What baffles me is the fact I'm able to run the pre compiled exe, so it doesn't seem to be a hardware incompatibility. I really have no idea what's going on :(

Have any of you ever faced a similar problem, where the prompt closes right at the FSI Initialize() line?

Rebeca

[Rebeca Guimarães]

Also, I just checked my prompt again and saw I'm getting this error when building Chrono. It happens in many lines of BceManager. 

D:\Rebeca\Desktop\Terramechanics\chrono-main-NASA\chrono-main\src\chrono_fsi\sph\physics\BceManager.cu(373): error : no
 instance of overloaded function "atomicAdd" matches the argument list [D:\Rebeca\Desktop\Terramechanics\chrono-main-NA
SA\chrono-main\build\src\chrono_fsi\sph\Chrono_fsisph.vcxproj]
              argument types are: (chrono::fsi::sph::Real *, chrono::fsi::sph::Real)
            atomicAdd(&body_forces[body_ID].x, sharedForces[0].x);
            ^

Rebeca

[Radu Serban]

Rebeca,

 

This error suggests that you either have a very old GPU (which is unlikely, as atomicAdd for double is available on devices with compute capability 6 and newer) or not really using CUDA 12.9 or 13.0 (but rather still 12.3).

 

You seem to have multiple versions of CUDA installed on your machine. How do you specify which one to use when you configure Chrono with CMake?

Could you please post the CMake messages during Chrono configuration that indicate which version of CUDA was found and will be used? Something like the following:

Searching for CUDA

  CUDA compiler:           C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.8/bin/nvcc.exe

  CUDA toolkit version:    12.8.61

  CUDA toolkit root dir:   C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.8

  CUDA archs (from cache): all-major

  CUDA found and enabled.

Searching for Thrust

  Thrust version:     2.7.0

  Thrust include dir: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.8/include

To view this discussion visit https://groups.google.com/d/msgid/projectchrono/3bf93f42-2f59-43b3-b63e-ee793140f586n%40googlegroups.com.

[Radu Serban]

I would suggest holding off on using the PyChrono conda package that Dan mentioned. That is not based on the current Chrono code (and was set up as a temporary solution for some tutorials we recently ran at UW-Madison). Updated official PyChrono conda packages with Chrono::FSI support will come soon.

To view this discussion visit https://groups.google.com/d/msgid/projectchrono/96f01d65-23b5-487b-9402-c739d258cd17n%40googlegroups.com.

[Radu Serban]

Hi Rebeca,

 

So, you do have a pretty old GPU.  Anything before 7.5 is now considered legacy.  In fact, 7.5 is also marked for obsoletion.

I’m afraid you’ll need to get access to newer hardware.

 

--Radu

 

 

From: Rebeca Guimarães <rebec...@gmail.com>
Sent: Friday, September 26, 2025 1:48 PM
To: Radu Serban <ser...@wisc.edu>
Subject: Mensagem particular referente a: [chrono] FSI won't work

 

Hey there, Radu!

 

My last try had indeed been with CUDA 12.3 yet again, since my last try with 13.0 resulted on the exe simply closing right from the start.

 

Searching for CUDA

CUDA compiler: D:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.3/bin/nvcc.exe

CUDA toolkit version: 12.3.52

CUDA toolkit root dir: D:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.3

CUDA archs (from cache): 60

CUDA found and enabled.

Searching for Thrust

Thrust version: 2.2.0

Thrust include dir: D:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.3/include

 

I was able to surpass the atomicAdd error by defining sm_60 instead of all_major as the arch, since I have a GPU with compute capability of 6.1. On the other hand, now I'm getting a new error of too many resources requested for launch when I compile and run the single wheel FSI demo. But that only happens when I run my own compiled demo, the pre compiled demo available on the Chrono website runs normally.

 

I don't know if there are any differences between the source code of the pre compiled demo and latest one available on git that would explain the higher demand for GPU. Or if it's still an error due to CUDA issues.

 

If you could shed any light on the matter, it would help a lot!

 

Thanks,

Rebeca

[Rebeca Guimarães]

Hey there, Radu!

I'm thrilled to tell you I was able to run the FSI module in my computer! 

It turned out FsiForceWCSPH was requesting 1024 threads and my GPU was unable to give that. I then adapted it and changed to fewer threads and it worked out.

I'm still working on the visualization, since I only had Irrlicht, so I'll now install VSG. 

It's just a temporary solution, since I'm requesting approval to use my uni's laboratories' computers, but up until then, I'm able to continue my VIPER single wheel simulations. Once I have access to the labs, I hope to be able to run complete rover simulations.

[Radu Serban]

That’s fantastic, Rebeca.

 

We’ll make a note about the number of threads issue (in case other users run into similar issues). The truth is we have not used older GPUs in such a long time, we take for granted certain capabilities.

 

If you can get access to a cluster with newer GPUs, you will see a dramatic increase in computational performance.

 

Best,
Radu

To view this discussion visit https://groups.google.com/d/msgid/projectchrono/fec67aab-1bfe-452c-a2f1-80ad05f8390dn%40googlegroups.com.