In LAMMPS package, I change the random number seed in velocity command to
simulate independent several calculations.
But I couldn't find it in GPUMD package. I tested several time and thought
that running step could relate to perform the several calculations.
In GPUMD, the seed for the random number generator is set according to the
system time in your computer. So normally you will have different seeds for
different runs. If you want to have the SAME seed for different runs (we
only use it for debugging), you can keep (perhaps it is already there)
-DDEBUG in the following line of the makefile:
CFLAGS = -std=c++11 -O3 -arch=sm_50 -DDEBUG
If you want to have different seeds in different runs (which is the normal
case for production run), remove -DDEBUG in this line before compiling the
code. This has been stressed in the online manual:
https://gpumd.zheyongfan.org/index.php/Installation
Did I misunderstand the process of GPUMD?
Is there any difference in operation of thermal conductivity calculation
b/w two packages? (If It is just that I couldn't understood, sorry for not
studying more hard..)
I am not sure what you are asking in this question. Perhaps you can make it
more clear...
On Tue, Sep 24, 2019 at 4:40 PM "소순성" <soonsung2001@xxxxxxxxxx> wrote:
Thank you for your kindness and for fast answers.
Ah! I have one additional question.
Before I used this wonderful GPUMD package, I've used LAMMPS package to
calculate the thermal conductivity.
In LAMMPS package, I change the random number seed in velocity command to
simulate independent several calculations.
But I couldn't find it in GPUMD package. I tested several time and thought
that running step could relate to perform the several calculations.
Did I misunderstand the process of GPUMD?
Is there any difference in operation of thermal conductivity calculation
b/w two packages? (If It is just that I couldn't understood, sorry for not
studying more hard..)
--------- 원본 메일 ---------보낸사람 : Bruce Fan <brucenju@xxxxxxxxx>
받는사람 : <gpumd@xxxxxxxxxxxxx>
받은날짜 : 2019년 Sep 24일(Tue) 22:00:09
제목 : [gpumd] Re: With one gpu(gtx 1080ti), can i simulate several gpumds
on several terminals?
Hi,
Yes you can. Depending on your purpose you can do it these ways:
1) Run several executable at the same time (e.g., you submit a job, and
then another user of the computer submit another job to use the same GPU).
This requires to set your GPU to the default mode instead of the exclusive
process mode. To check the compute mode, run nvidia-smi
in your terminal. If you see "E. process" within the output, it means that
the current compute mode of the GPU is exclusive process mode and does not
allow running more than one (computing type) CUDA application. You can use
the following command to change to the default mode: sudo nvidia-smi -i
0 -c 0
In the default compute mode, you can run more than one CUDA executable.
2) If you want to run a few simulations one by one, but with a single
submission of your job, you can prepare the inputs for all the simulations
(each has a folder) and put the paths to these folders in a single "driver
input file". An example is provided in the GPUMD package. In the file
"examples/input_gpumd.txt" you can find the following text:
5
examples/gpumd/ex1
examples/gpumd/ex2
examples/gpumd/ex3
examples/gpumd/ex4/ballistic
examples/gpumd/ex4/diffusive
This means that the five simulations corresponding to the above five
folders will be run one by one after submitting a job with this command:
src/gpumd < examples/input_gpumd.txt
Zheyong
On Tue, Sep 24, 2019 at 3:27 PM "소순성" <soonsung2001@xxxxxxxxxx> wrote:
Thank you to Zheyong Fan for using this good GPUMD program.
I have question. I'm newbie in gpu calculation and I use this program in
my desktop with one gpu(gtx 1080ti, Ubuntu OS).
And I want several calculations at the same time.
So, can I use gpumd program using several terminals for several gpumds at
the same time?
