(CLOSED) Physx 3.3.2 enabling GPU acceleration on Linux

Gaming and Visualization Technologies APIs Physics Modeling (closed)
#1

I installed the latest version of physx (3.3.2) on Linux, and I would like to enable the GPU, but the program crash when calling the function “mCudaContextManager-> getGpuDispatcher ()”, because PxCudaContextManager not created.

Are there any examples or guidelines on how to enable physx GPU on linux?

The documentation refers to an “application GUID”, how can I get it?

Thanks in advance.

#2

Did you create an instance of the CUDA context manager? Object creation does not differ between the different versions of PhysX afaik, so I don’t see how the CUDA context manager creation would be different on Linux than say on Windows.

#3

Thanks for the reply …

I’m new to Physx.

I have followed the user guide and I think I created the CUDA context manager but when I run “mCudaContextManager->getGpuDispatcher();” the program crash.

I tried to remove the CPUDispatcher and put it back but it does not change anything.

However enclose the entire statement Physx:

mFoundation = PxCreateFoundation(PX_PHYSICS_VERSION,mAllocator,mErrorCallback);
    PxProfileZoneManager* profileZoneManager = &PxProfileZoneManager::createProfileZoneManager(mFoundation);
    mPhysics = PxCreatePhysics(PX_PHYSICS_VERSION, *mFoundation, PxTolerancesScale(),true);

    mCPUDispatcher = PxDefaultCpuDispatcherCreate(2);

    PxCudaContextManagerDesc cudaContextManagerDesc;
    mCudaContextManager = PxCreateCudaContextManager(*mFoundation,cudaContextManagerDesc,profileZoneManager);

    PxSceneDesc sceneDesc(mPhysics->getTolerancesScale());
    sceneDesc.gravity = PxVec3(0.0f, -9.81f, 0.0f);
    sceneDesc.cpuDispatcher	= mCPUDispatcher;
    <b>CRASH HERE ---></b> sceneDesc.gpuDispatcher = mCudaContextManager->getGpuDispatcher();
    sceneDesc.filterShader	= PxDefaultSimulationFilterShader;
    sceneDesc.flags = PxSceneFlag::eENABLE_ACTIVETRANSFORMS;
    mScene = mPhysics->createScene(sceneDesc);
    mCooking = PxCreateCooking(PX_PHYSICS_VERSION, *mFoundation, PxCookingParams(PxTolerancesScale()));
    mControllerManager = PxCreateControllerManager(*mScene);
#4

From first glance everything with the CUDA context manager seems ok. I would suggest putting a breakpoint in your code and checking to see if its actually created. Also what HW are you running on.

CUDA is Nvidia proprietary so if you are not using a Nvidia GPU, it will not work. Also what version of the drivera are you using?

#5

Searching the net I read that GPU Physx 3.3.2 only works with gpu Fermi and higher …
Damn, my gpu is a Nvidia GTS250 and I think that is not supported (it’s a bit old) …

Driver Version : 340.46

I ask for confirmation …

#6

Yes, I think you will have to use a more recent card. However the function calls should be returning an error code that will indicate that the driver or GPU is not up to spec.

#7

Thank you all for the answers …
Now I have everything clear.