Out of memory
ww w
I've copied the log below:
Device 1 of 4: NVIDIA A100-PCIE-40GB
Compute Capability: 8.0
Global Memory: 42298834944 B
Constant Memory: 65536 B
Shared Memory: 49152 B
Registers: 65536
Clock Speed: 1.41 GHz
Number of SMs: 108
Number of Cores: 6912
Auto-thread: 442368
Auto-block: 64
============================= GPU Information ================================
Device 2 of 4: NVIDIA A100-PCIE-40GB
Compute Capability: 8.0
Global Memory: 42298834944 B
Constant Memory: 65536 B
Shared Memory: 49152 B
Registers: 65536
Clock Speed: 1.41 GHz
Number of SMs: 108
Number of Cores: 6912
Auto-thread: 442368
Auto-block: 64
============================= GPU Information ================================
Device 3 of 4: NVIDIA A100-PCIE-40GB
Compute Capability: 8.0
Global Memory: 42298834944 B
Constant Memory: 65536 B
Shared Memory: 49152 B
Registers: 65536
Clock Speed: 1.41 GHz
Number of SMs: 108
Number of Cores: 6912
Auto-thread: 442368
Auto-block: 64
============================= GPU Information ================================
Device 4 of 4: NVIDIA A100-PCIE-40GB
Compute Capability: 8.0
Global Memory: 42298834944 B
Constant Memory: 65536 B
Shared Memory: 49152 B
Registers: 65536
Clock Speed: 1.41 GHz
Number of SMs: 108
Number of Cores: 6912
Auto-thread: 442368
Auto-block: 64
ans =
4x1 struct array with fields:
name
id
devcount
major
minor
globalmem
constmem
sharedmem
regcount
clock
sm
core
autoblock
autothread
maxgate
>> >> >> >> >> >> >> >> >> >> >> >> >> >> Launching MCXLAB - Monte Carlo eXtreme for MATLAB & GNU Octave ...
Running simulations for configuration #1 ...
mcx.respin=1;
mcx.nphoton=1e+09;
mcx.maxdetphoton=1e+08;
mcx.gpuid='1111';
mcx.srctype='pencil';
mcx.unitinmm=1;
mcx.savedetflag=5;
mcx.issaveref=1;
mcx.isspecular=1;
mcx.autopilot=1;
mcx.srcdir=[0 0 1 0];
mcx.srcpos=[99 59 0 1];
mcx.issrcfrom0=1;
mcx.detnum=4;
mcx.tstart=0;
mcx.tend=1e-08;
mcx.tstep=2e-11;
mcx.dim=[200 200 200];
mcx.mediabyte=1;
mcx.medianum=5;
MCXLAB ERROR -2 in unit mcx_core.cu:3006: out of memory
Error from thread (1): out of memory
Qianqian Fang
the error was raised on line 3006 of mcx_core.cu, which reads
https://github.com/fangq/mcx/blob/v2024.2/src/mcx_core.cu#L3006
this is the line where the main volumetric output array, gfield,
on the GPU is allocated.
the needed size of the output array is explained in
https://github.com/fangq/mcx?tab=readme-ov-file#requirement-and-installation
"For simulations with large volumes, sufficient graphics
memory is also required to perform the simulation. The minimum
amount of graphics memory required for a MC simulation is
Nx*Ny*Nz bytes for the input tissue data plus Nx*Ny*Nz*Ng*4*2
bytes for the output flux/fluence data - where Nx,Ny,Nz are the
dimensions of the tissue volume, Ng is the number of concurrent
time gates, 4 is the size of a single-precision floating-point
number, 2 is for the extra memory needed to ensure output
accuracy (#41). MCX does not require double-precision support in
your hardware."
let's do the math in your case
your domain volume is Nx=Ny=Nz=200, your time-gate Nt=tend/tstep = 1e-8/2e-11=500. the total bytes just for the volumetric output alone is 200*200*200*500*4*2=30GB.
each of your GPU has a max 40GB global memory, but there are other programs (run nvidia-smi to list all programs using the GPU) and other smaller variables needed by mcx. In the end, it is not surprising it raise this error.
a related question is what spatial resolution and temporary resolution is sufficient for your simulation? can yo work with a 100x100x100 volume? do you really need 2e-11 as your tstep?
Qianqian
--
You received this message because you are subscribed to the Google Groups "mcx-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to mcx-users+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/mcx-users/305cd539-b0ee-4b91-886c-b86241a92b95n%40googlegroups.com.