Dyverso Crashes when using OpenCL-GPU or CUDA

Describe your realflow bugs
duke3d
Posts: 22
Joined: Tue Aug 30, 2016 11:54 pm

Re: Dyverso Crashes when using OpenCL-GPU or CUDA

Postby duke3d » Thu Mar 30, 2017 8:25 am

I have the same problem. I use Dyverso Granular particles. Every time I simulate with GPU it Realflow crashes usually at random frame. I tried it on Windows 10 it was crushing every time Video memory was filled up to 4Gigadytes or so. Now, in Windows 7 it crashes every time it reaches 1 GByte of Video memory. Sometimes it crashes right at the very first frame. I tried from 200K to 20M particles. I tried different settings, nothing helped. I tried Command line mode as well. That is what I got:

>>> Setting up project
>>> 7%
>>> 14%
>>> 21%
>>> 28%
>>> 35%
>>> 42%
>>> 50%
>>> 57%
>>> 64%
>>> 71%
>>> 78%
>>> 85%
>>> 92%
>>> 100%

Using 40 threads for this simulation ...

>14:45:14: Using 40 thread(s) for this simulation.
>14:45:14: Using CUDA device GeForce GTX 1070 for Dyverso for this simulation.
>14:45:15: ================ Frame 0 started ================
CUDA ERROR: an illegal memory access was encountered cudaMallocHost 880
CUDA ERROR: an illegal memory access was encountered cudaMemcpyHostToDevice 216
CUDA ERROR: an illegal memory access was encountered cudaMemsetAsync 276

I get these errors all the time. The difference is that the memory addresses vary.
I made a simple scene with 2 Dyverso Granular Domains, added a filter. Made it so that particles with particular speed moved from first domain to the second one. I got Error. I even had few crashes without GPU utilization. Without Filter there is now problem.

My PC:
2x Xeon 2630 v4
ASUS Z10PE-D8
64 GB ECC Registered RAM
256 GB SSD
2x 2TB HDD
Gigabyte GTX 1070 8GB
Win 7 64bit


duke3d
Posts: 22
Joined: Tue Aug 30, 2016 11:54 pm

Re: Dyverso Crashes when using OpenCL-GPU or CUDA

Postby duke3d » Mon Apr 03, 2017 3:52 pm

hmm.. today I got the same problem.
This time I had no filters at all. One Dyverso (SPH) domain and emitter, 5 k_Volume Daemons. Again, it reached about 4GB of vRAM in Win 10, and crushed.

Elapsed time: 00:26:30
Remaining time (estimated): 05:08:56
Average time (frame): 00:00:20

>22:35:29: ================ Frame finished ================

>22:35:29: ================ Frame 1079 started ================
.....CUDA ERROR: out of memory cudaMalloc 712
CUDA ERROR: out of memory cudaMalloc 712
CUDA ERROR: out of memory cudaMalloc 712
CUDA ERROR: out of memory cudaMalloc 712

Crushed at about 8.4 M particles


Return to “Realflow bug report”

Who is online

Users browsing this forum: No registered users and 3 guests