Hi Mohammed, Just to add to what Dan said and be a bit more specific...
First I'd like to address the speed issue. I see you used " SetForceCalcThreadsPerBlock(1)". Please remove it. As noted in the error message, it is for custom force model users (so that they can tweak it to prevent some CUDA failures when their model uses too much local memory per thread). It seems you are not using a custom model, so what it solely does here is launch 1 thread per CUDA block for force calculation, slowing things down some 100 times. The bigger problem appears to be that everything in your simulation combined, probably occupies a cube in space not larger than several millimeters across. Yet the simulation world is around 10 meters in size (see the InstructBoxDomainDimension line). The solver needs to subdivide the domain to distribute the workload, and such a size discrepancy probably leads to a situation, where however the subdivision is created, one of the subdomains will contain almost all simulation entities, and bust its memory; it cannot create a finer subdivision either, as that busts the pool of available domain IDs. You *should scale the simulation world* with everything else, too, though I seem to have forgotten to put scaling in that line. I thought about a more dynamic subdivision strategy, but right now it's not a high development priority, and usually, domain sizes just somewhat physically meaningful should work fine. After you make these changes it might run, but I wouldn't be sure if it represents anything like you wanted to do---this demo's clump templates are quite complex, and there's definitely no guarantee that a demo runs without some fine-tuning after you scaled a variable some 1000 times. You should consider using ParaView to create a rendering and include a screenshot (even if it fails at step 1, we can still have a look at the initial setup) for further discussion, which can help us rule out so many possibilities. Thank you, Ruochun On Saturday, April 12, 2025 at 3:58:01 AM UTC+8 Mohammed Saffarini wrote: > Dear Chrono-DEM Team, > > I am trying to simulate a setup similar to the demo DEMdemo_Repose, but > for particles that are much smaller in radius. > > The original demo file ran normally to the end. However, when I tried to > build my own setup with particles sizes that I need, the setup did NOT run. > In fact, after so many trials that didn't work out to create a setup of my > own, I finally decided to take the demo file and just use a scaling factor > that can get me the particle sizes that I am looking for. However, I am > still unable to get the simulation working for that range of particles > sizes. > > The particle sizes I am targeting is between 10 and 20 micrometers. > However, since I am not able to simulate this size, I tried particles sizes > in the range of 100 - 200 micrometer (Using a scale factor of 0.01 instead > of the 2 that is in the original demo file). I made the necessary > modification to get everything else properly scaled as well. However, the > simulation takes an average of 8 - 12 minutes on 2 GPUs for every 0.1 > seconds of simulation time. The system has 8,850 particles only. > > The following warning is shown for small particles range: > > WARNING! A type of clump is instructed to have near-zero (or negative) > mass or moment of inertia (mass: 2.0000003e-10, MOI magnitude: > 3.080585e-18). This could destabilize the simulation. > Please make sure this is intentional. > > > WARNING! A type of clump is instructed to have near-zero (or negative) > mass or moment of inertia (mass: 1.00000015e-10, MOI magnitude: > 1.5402925e-18). This could destabilize the simulation. > Please make sure this is intentional. > > > WARNING! A type of clump is instructed to have near-zero (or negative) > mass or moment of inertia (mass: 4.00000061e-10, MOI magnitude: > 6.16117001e-18). This could destabilize the simulation. > Please make sure this is intentional. > > > WARNING! A type of clump is instructed to have near-zero (or negative) > mass or moment of inertia (mass: 2.0000003e-10, MOI magnitude: > 3.080585e-18). This could destabilize the simulation. > Please make sure this is intentional. > > > WARNING! A type of clump is instructed to have near-zero (or negative) > mass or moment of inertia (mass: 2.0000003e-10, MOI magnitude: > 3.080585e-18). This could destabilize the simulation. > Please make sure this is intentional. > > > WARNING! A type of clump is instructed to have near-zero (or negative) > mass or moment of inertia (mass: 2.0000003e-10, MOI magnitude: > 3.080585e-18). This could destabilize the simulation. > Please make sure this is intentional. > > > > And finally, the following error is obtained: > > -------- Simulation crashed "potentially" due to too many geometries in a > bin -------- > The dT reported max velocity is 0 > > ------------------------------------ > If the velocity is huge, then the simulation probably diverged due to > encountering large particle velocities. > Decreasing the step size could help, and remember to check if your > simulation objects are initially within the domain you specified. > > ------------------------------------ > If the velocity is fair, and you are using a custom force model, one thing > to do is to SetForceCalcThreadsPerBlock to a small number like 128 (see > README.md troubleshooting for details). > If you are going to discuss this on forum > https://groups.google.com/g/projectchrono, please include a visual > rendering of the simulation before crash. > > terminate called after throwing an instance of 'std::runtime_error' > what(): GPU Assertion: unspecified launch failure. This happened in > /home/4h1/src/DEM-Engine/src/algorithms/DEMCubContactDetection.cu:384 > > > srun: error: mod-condo-g28: task 0: Aborted (core dumped) > > > > Here below, I am providing the latest code I used for the range of 10 - 20 > micrometers in particle’s radius. Please let me know what the problem is > here: > > #include <core/ApiVersion.h> > > #include <core/utils/ThreadManager.h> > > #include <DEM/API.h> > > #include <DEM/HostSideHelpers.hpp> > > #include <DEM/utils/Samplers.hpp> > > #include <cstdio> > > #include <chrono> > > #include <filesystem> > > using namespace deme; > > using namespace std::filesystem; > > int main() { > > DEMSolver DEMSim; > > DEMSim.UseFrictionalHertzianModel(); > > DEMSim.SetVerbosity(INFO); > > DEMSim.SetOutputFormat(OUTPUT_FORMAT::CSV); > > DEMSim.SetNoForceRecord(); > > srand(42); > > float scaling = 0.001; > > int num_template = 6; > > int min_sphere = 1; > > int max_sphere = 5; > > float min_rad = 0.01 * scaling; > > float max_rad = 0.02 * scaling; > > float min_relpos = -0.01 * scaling; > > float max_relpos = 0.01 * scaling; > > auto mat_type_walls = DEMSim.LoadMaterial({{"E", 1e8}, {"nu", 0.3}, > {"CoR", 0.3}, {"mu", 1}}); > > auto mat_type_particles = DEMSim.LoadMaterial({{"E", 1e9}, {"nu", > 0.3}, {"CoR", 0.7}, {"mu", 1}}); > > DEMSim.SetMaterialPropertyPair("CoR", mat_type_walls, > mat_type_particles, 0.3); > > auto funnel = > DEMSim.AddWavefrontMeshObject(GetDEMEDataFile("mesh/funnel.obj"), > mat_type_walls); > > funnel->Scale(0.01); > > float funnel_bottom = 0.f; > > std::vector<std::shared_ptr<DEMClumpTemplate>> clump_types; > > for (int i = 0; i < num_template; i++) { > > int num_sphere = rand() % (max_sphere - min_sphere + 1) + 1; > > float mass = 0.1 * (float)num_sphere * std::pow(scaling, 3); > > float3 MOI = make_float3(2e-5 * (float)num_sphere, 1.5e-5 * > (float)num_sphere, 1.8e-5 * (float)num_sphere) * 50. * std::pow(scaling, 5); > > std::vector<float> radii; > > std::vector<float3> relPos; > > float3 seed_pos = make_float3(0); > > for (int j = 0; j < num_sphere; j++) { > > radii.push_back(((float)rand() / RAND_MAX) * (max_rad - > min_rad) + min_rad); > > float3 tmp; > > if (j == 0) { > > tmp.x = 0; > > tmp.y = 0; > > tmp.z = 0; > > } else { > > tmp.x = ((float)rand() / RAND_MAX) * (max_relpos - > min_relpos) + min_relpos; > > tmp.y = ((float)rand() / RAND_MAX) * (max_relpos - > min_relpos) + min_relpos; > > tmp.z = ((float)rand() / RAND_MAX) * (max_relpos - > min_relpos) + min_relpos; > > } > > tmp += seed_pos; > > relPos.push_back(tmp); > > int choose_from = rand() % (j + 1); > > seed_pos = relPos.at(choose_from); > > } > > > > auto clump_ptr = DEMSim.LoadClumpType(mass, MOI, radii, relPos, > mat_type_walls); > > clump_types.push_back(clump_ptr); > > } > > float spacing = 0.1 * scaling; //0.08 * scaling; > > float fill_width = 0.01f; > > float fill_height = 0.02f * fill_width; > > float fill_bottom = funnel_bottom + fill_width + spacing; > > PDSampler sampler(spacing); > > std::vector<std::shared_ptr<DEMClumpTemplate>> > input_pile_template_type; > > std::vector<float3> input_pile_xyz; > > float layer_z = 0; > > while (layer_z < fill_height) > > { > > float3 sample_center = make_float3(0, 0, fill_bottom + layer_z + > spacing / 2); > > auto layer_xyz = sampler.SampleCylinderZ(sample_center, > fill_width, 0); > > unsigned int num_clumps = layer_xyz.size(); > > for (unsigned int i = 0; i < num_clumps; i++) { > > input_pile_template_type.push_back(clump_types.at(i % > num_template)); > > } > > input_pile_xyz.insert(input_pile_xyz.end(), layer_xyz.begin(), > layer_xyz.end()); > > layer_z += spacing; > > } > > auto the_pile = DEMSim.AddClumps(input_pile_template_type, > input_pile_xyz); > > > > DEMSim.InstructBoxDomainDimension({-10, 10}, {-10, 10}, {funnel_bottom > - 1.f, funnel_bottom + 20.f}); > > DEMSim.InstructBoxDomainBoundingBC("top_open", mat_type_walls); > > DEMSim.SetInitTimeStep(1e-7); > > DEMSim.SetGravitationalAcceleration(make_float3(0, 0, -9.81)); > > DEMSim.SetMaxVelocity(25.); > > DEMSim.SetErrorOutAvgContacts(10000); > > DEMSim.SetForceCalcThreadsPerBlock(1); > > DEMSim.Initialize(); > > path out_dir = current_path(); > > out_dir += "/DemoOutput_Repose"; > > create_directory(out_dir); > > > > std::chrono::high_resolution_clock::time_point start = > std::chrono::high_resolution_clock::now(); > > for (int i = 0; i < 140; i++) { > > char filename[200], meshfile[200]; > > sprintf(filename, "%s/DEMdemo_output_%04d.csv", out_dir.c_str(), > i); > > sprintf(meshfile, "%s/DEMdemo_funnel_%04d.vtk", out_dir.c_str(), > i); > > DEMSim.WriteSphereFile(std::string(filename)); > > DEMSim.WriteMeshFile(std::string(meshfile)); > > std::cout << "Frame: " << i << std::endl; > > DEMSim.DoDynamics(1e-1); > > DEMSim.ShowThreadCollaborationStats(); > > } > > std::chrono::high_resolution_clock::time_point end = > std::chrono::high_resolution_clock::now(); > > std::chrono::duration<double> time_sec = > std::chrono::duration_cast<std::chrono::duration<double>>(end - start); > > std::cout << time_sec.count() << " seconds (wall time) to finish the > simulation" << std::endl; > > DEMSim.ShowTimingStats(); > > DEMSim.ClearTimingStats(); > > std::cout << "DEMdemo_Repose exiting..." << std::endl; > > return 0; > > } > -- 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 projectchrono+unsubscr...@googlegroups.com. To view this discussion visit https://groups.google.com/d/msgid/projectchrono/9de47d87-bb5e-4fe3-8898-e8a423fdf959n%40googlegroups.com.
