Encountered a CUDA error when add two Transform node to a Group node

//create first Transform node
GeometryGroup object_group = LoadDataMesh(furniture->GetMesh());
Transform optix_mesh = m_Context->createTransform();
optix_mesh->setMatrix(true, (float*)&furniture->GetMatrix(),0);
optix_mesh->setChild(object_group);

//create second transform node similarly	
GeometryGroup object_group1 = LoadDataMesh(furniture->GetMesh());
Transform optix_mesh1 = m_Context->createTransform();
optix_mesh1->setMatrix(true, (float*)&furniture->GetMatrix(), 0);
optix_mesh1->setChild(object_group1);

//add Transform to Group, which is also the top object of ray tracing
m_RootGroup->addChild(optix_mesh);
m_RootGroup->addChild(optix_mesh1);

when add optix_mesh and optix_mesh1 to the Group node, then call context->launch(0,0,0), it fails with “Unknown error (Details: Function “_rtContextLaunch2D” caught exception: Encountered a CUDA error: Kernel launch returned (719): Launch failed, [6619200])”
However, when I just add one mesh (no matter optix_mesh or optix_mesh1) to Group node, it works normally and i get the correct image. And more, if optix_mesh and optix_mesh1 share the same GeometryGroup child, it also works. But when they do not share the same GeoemtryGroup , the error above still appears.
Please help me…
System information:
Windows 7 Utimate 64 SP1, Nvidia Titan X,driver version 358.91, otpix 3.8.0, cuda v7.0.27. I’m using 32bit application.

Hi,
be sure, that you have set acceleration structure for m_RootGroup correctly.

Yes, this is how I set up the m_RootGroup

m_RootGroup = m_Context->createGroup();
m_RootGroup->setAcceleration(m_Context->createAcceleration("Trbvh", "Bvh"));
m_Context["_TOP_OBJECT"]->set(m_RootGroup);

When reporting issues please always list at least the following information:
OS version, installed GPU(s), display driver version, OptiX version, CUDA Toolkit version used.
Application bitness should be 64-bit, there won’t be support for 32-bit anymore.

CUDA error 719 is a CUDA_ERROR_LAUNCH_FAILED. You’ll find these enums in cuda.h of the CUDA toolkit.
Unfortunately that’s just the CUDA generic error for “something went wrong” and doesn’t help much to isolate what exactly didn’t work.

If the error happens with a dummy launch which validates the OptiX state, compiles the kernel, and builds the acceleration structures, this could be a problem with the acceleration structure (AS) build.
In such cases try to isolate this by using other AS builders, for example Bvh instead of Trbvh.

If you aren’t using the newest available OptiX 3.8.0 version, try that.
If OptiX 3.8.0 doesn’t help, but Bvh worked, try again with the next upcoming OptiX release which should be out very soon.

If Bvh or that newest OptiX version still fails you might have a different problem. Then an OptiX API Capture (OAC) trace would be needed to see what API calls you’re really doing. Please see this thread for instructions how produce an OAC trace: https://devtalk.nvidia.com/default/topic/803116/?comment=4436953

It turns out that my LoadDataMesh has loaded mtl file wrongly. Although it doesn’t explain why it always fails at the second Transform node, the problem does disappear now.Thanks Detlef and Smejkal for help