MATLAB: Matlab reboots the computer when running parellel computation

MATLABmatlab ubuntu memory reboot parellel-computationParallel Computing Toolbox

I'm running Matlab 2018b on Ubuntu 18.04 with a 14-cores CPU and 32GB memory. When I try to run my code (using parellel computation to call c++ and CPLEX), the computer will automatically reboot without any warning. Sometimes the computer does not reboot but crashs (no display, no audio output, no usb) and the small screen on motherboard says I need to detect my memory. If I shutdown some workers, the reboot or crash will happen later. I have tested all my hardwares without finding any problems. The CPU temperature is normal and the memory is occupied by at most 1/3 before rebooting. And my code runs successfully on my old computer with 4-cores based on Windows 10.

Is there anyone having some idea to solve it? Many thanks.

Best Answer

If the mainboard display tells you to check the memory, check the memory exhaustively. I'm using the free version of memtest86 https://www.memtest86.com/ or the Ultimate Boot CD: http://www.ultimatebootcd.com/ . A damaged RAM can cause the observed behavior: The more cores are running, the more likely the failing RAM cells are used.

Related Solutions

MATLAB: Help diagnosing PC restart issue while computing matrix pseudo inverse

Thank you all for your help. I figured out the issue, I had OC'd my CPU a while ago, and despite running on 'standard performance' in my OC tool, I was still running at 4.5 GHz. Manually changing it back to default and everything works now! (12 hours later)

MATLAB: Can the MEX-file in MATLAB 7.3 (R2006b) not reliably access mxArrays after the main mexFunction() exits

These mxArrays can be removed from MATLAB's tracking by passing them to mexMakeArrayPersistent(). However, the caller of mexMakeArrayPersistent() implicitly takes ownership of that array and must ensure that at some point it is freed (via mxDestroyArray()) or else this will produce a memory leak.

The attached example MEX-files demonstrate a failure (memtest1.cpp) as well the correct management of the persistent mxArray (memtest2). To execute these, initialize the variable "x" and call each MEX-function twice, as demonstrated below.

memtest1 will cause a segmentation violation:

mex -v memtest1.cpp
x = 7;
memtest1 
memtest1
 Initializing pointer
 Ptr: 287730336
 Value of x(1): 7.000000
 Existing pointer
 Ptr: 287730336
 Probably going to segv:

 ------------------------------------------------------------------------
        Segmentation violation detected at Mon Oct 02 19:45:57 2006
 ------------------------------------------------------------------------

memtest2 will execute successfully:

mex -v memtest2.cpp
x = 7;
memtest2
memtest2
 Existing pointer
 Ptr: 324198688
 Value of x(1): 7.000000
 Existing pointer
 Ptr: 324198688
 Value of x(1): 7.000000

This type of interface will allow you to control the managment of certain mxArrays. However, they must be correctly managed using mxDestroyArray() in your program, as they are no longer tracked by the MATLAB memory manager.

Best Answer

Related Solutions

MATLAB: Help diagnosing PC restart issue while computing matrix pseudo inverse

MATLAB: Can the MEX-file in MATLAB 7.3 (R2006b) not reliably access mxArrays after the main mexFunction() exits

Related Question