Chrono becomes impossibly slow - after a certain number(arbitrary) of steps

a_meduri2312

Hi all,

I am trying to simulate a two body floating system in water with an initial velocity. This is the GPU I am using

Device 0: "NVIDIA A100-PCIE-40GB"
  CUDA Capability Major....: 8.0
  Global memory............: 40536 MBytes
  CUDA Cores...............: 6912 (108 Multiprocessors,  64 CUDA Cores/MP)
  GPU Max Clock rate.......: 1410 MHz (1.41 GHz)
  Memory Clock rate........: 1215 Mhz
  Memory Bus Width.........: 5120-bit
  L2 Cache Size............: 40960 KBytes
  Constant memory..........: 64 KBytes
  Shared memory per block..: 48 KBytes
  Registers per block......: 65536
  Maximum threads per MP...: 2048
  Maximum threads per block: 1024

For the first n time steps, the average part (0.01s) completes in around 60 s each. After a point, the number keeps increasing to very very high values

Part_0062      0.620010         18271      290    7416.15  01-12-2021 02:14:52
Part_0063      0.630011         18561      290    8340.36  01-12-2021 02:20:27
Part_0064      0.640007         18851      290    9396.37  01-12-2021 02:28:34
Part_0065      0.650008         19141      290   10568.82  01-12-2021 02:39:29
Part_0066      0.660012         19431      290   12006.74  01-12-2021 02:53:41
Part_0067      0.670018         19721      290   13372.18  01-12-2021 03:10:53
Part_0068      0.680026         20011      290   17478.97  01-12-2021 03:37:38
Part_0069      0.690031         20301      290   22696.46  01-12-2021 04:16:13
Part_0070      0.700001         20590      289   28586.20  01-12-2021 05:07:33
Part_0071      0.710012         20880      290   34909.09  01-12-2021 06:12:30
Part_0072      0.720022         21170      290   41675.10  01-12-2021 07:31:20
Part_0073      0.730034         21460      290   49305.96  01-12-2021 09:05:26
Part_0074      0.740011         21749      289   57715.70  01-12-2021 10:55:35
Part_0075      0.750023         22039      290   65813.61  01-12-2021 13:01:12
Part_0076      0.760003         22328      289   74698.31  01-12-2021 15:22:33
Part_0077      0.770016         22618      290   85062.92  01-12-2021 18:03:09
Part_0078      0.780032         22908      290   97917.52  01-12-2021 21:07:11
Part_0079      0.790010         23197      289  108760.41  0

As you can see, the simulation time/sec keeps climbing. This has happened every time I run Chrono in 3D. I have attached my XML file.

I have two questions.

1. The collision between the two bodies occurs much before 0.62 s, so I believe there is no way it is linked to the collision itself.
2. What is this number dependent on and is there any way I can reduce this? I was reading that we can use multi-GPU with Project Chrono. How feasible is this?

a_meduri2312

ClosedBodySim.zip

This is my XML file along with STL files for the simulation.

Asalih3d

First a question: What is your experience with using A100 Nvidia GPU? What is the maximum number of particles you can do? I see it has 40 gb of RAM quite impressive. The simulation speed should be impressive too!

Then for your question, have you tried playing with these parameters:

Perhaps they would explain why the simulation takes a lot of time suddenly. Could it be that after the bodies have collided it keeps checking if they are still colliding etc.?

I will be honest and say I don't know too much about this aspect, but I hope you are able to figure it out, so simulation time does not become prohibitively long :-)

Kind regards

Alex

The simulation performs around 290 steps between output files... that is not changing.

Can you send to us the final complete Run.out?

Alex