User Forum of Software BASEMENT

BASEMENT
Basic Simulation Environment for computation of environmental flow and natural hazard simulation
Laboratory of Hydraulics, Hydrology and Glaciology (VAW)
ETH Zurich
Basement_Logo

You are not logged in.

#1 2021-01-25 10:51:46

Matteo Facchini
Developer
From: Trento
Registered: 2014-09-05
Posts: 281

Runtime API erorr

Hello

I am encountering an error on some simulations on a very large domain (1.6M cells) and on some others with less cells (~ half, 800k cells).

The simulations stop after few minutes with the following error:

op_cuda_rt_support.c(96) : cutilSafeCall() Runtime API error : out of memory

This happens both on my windows 10 environment (NVIDIA Quadro P620 with 18347 MB of RAM) and on a ubuntu server with NVIDIA GPUs with 11019 MiB of RAM.

On my windows machine I was able to run the simulation just by simply restarting it from the beginning, on the server I do not have any control, so I am just trying some other solutions, i.e. run the simulation with the Cuda backend instead of the CudaO.

Cheers
Matteo

EDIT: included in the error there is the path to

op_cuda_rt_support.c

which is pointing to

externalLibs/OP2/op2/c/src/cuda

Last edited by Matteo Facchini (2021-01-25 10:54:32)

Offline

#2 2021-01-26 09:13:54

Lukas Vonwiller
Developer
Registered: 2014-09-04
Posts: 127

Re: Runtime API erorr

Ciao Matteo,

It looks like you dont't have enough memory on the GPU. The RAM on your Computer or Server does not matter. Check the memory of your GPU card using

nvidia-smi

hope this helps!
cheers, Lukas

Offline

#3 2021-01-26 11:30:40

Matteo Facchini
Developer
From: Trento
Registered: 2014-09-05
Posts: 281

Re: Runtime API erorr

Thanks Lukas, I was indeed talking about the GPU memory, I though it was also called RAM (or maybe VRAM?).

Anyway, the command you suggest reported an intense memory usage (1951MiB /  2048MiB) but the simulation runs.

Did you ever encounter problems with very large domains due to lack of GPU memory?

Furthermore, I see that the total available graphics memory (18347MB) is way larger than the dedicated one (2048 MB GDDR5). Do you know why?

EDIT: got it! The total available graphics memory is the system RAM, which I cannot share with the GPU.

Last edited by Matteo Facchini (2021-01-26 11:36:35)

Offline

#4 2021-01-26 15:16:18

Lukas Vonwiller
Developer
Registered: 2014-09-04
Posts: 127

Re: Runtime API erorr

Did you ever encounter problems with very large domains due to lack of GPU memory?

Yes, I ran out of memory for a very large mesh on a GPUs with 4GB Memory.

Offline

Board footer

Powered by FluxBB