The computer should have enough memory, but it failed. This error appeared when I tried to do a NVT simulation. The simulation box is 768.00000 768.00000 768.00000 [nm**3], in which there are 100 chains. The force-field is OPLSAA. I attached the md parameter file for this simulation. Hope someone who can help me find the problem. Thanks in advance.
I think it is logical. I have never seen this kinda big system even in coarse grained model. Maybe it works in at least 32 cores but am not sure about that. Probably, you have to consider using smaller systems.
That's a truly massive system. The box size itself doesn't matter so much as how many atoms you put inside it. As much as you can say, what are you simulating? If by "chains" you mean segments of a protein, then I'm guessing your simulating a lot of stuff in that box. It very likely will not have enough memory on any single machine unless you have a high-memory compute node available or some program that will distribute the data across multiple machines.
it also depends how this is parallelized. if thru OMP then the memory footprint is higher than MPI. Using GPU it is even higher. apart from technical stuff, this is indeed large system.
Yes, it is big. Thanks all of you. I am simulating a polymer melt and I used the experimental density to set the box size (I don't know if I am correct or not for setting based on density). I tried to find out how the memory was allocated. The memory should be fourier-nx*fourier-ny*fourier-nz*4(for float type) byte, according to the code. But number of the allocated elements are different between error exhibition and my calculation. For example, for the case which has 50 chains, 300 carbon for each chain, the allocated elements is 3,306,243,099 from error and 28,991,029,248 from my calculation. The difference is quite big. Do you know why?
By the way, changing fourierspacing can change the point for integration in PME. But how pme-order works?
I used OMP. Yes, the technical stuff is also complicated for the calculation of big system.
Thanks again! Maybe I need to scale my system a bit smaller if the density is not important, or change another computer.
I Assume its gromacs. it is better you should provide system information such as the number of atoms and better to provide the nvt.log file
- Badges
- Science topic