GPU Processing Using Embedded Python in Origin

Numba is a Python compiler that can compile Python code for execution on CUDA-capable GPUs or multicore CPUs.  In this blog, we will show steps to introduce such capability into Origin’s embedded Python.

Step 1 Find out the CUDA version of your GPU

First  we need to find out the CUDA version. Go to the control panel (Search Control Panel in the Windows OS Search  Entry), in  the  Control  Panel, find NVIDIA Control Panel and open it.  

Click System Information link at bottom left. Select Components tab and find NVCUDA64.DLL in the table. You should find the CUDA version higher than 11.1.1  in this case.

Step 2   Download CUDA toolkit

Go to the webpage https://developer.nvidia.com/cuda-toolkit-archive and download the toolkit for the  corresponding CUDA version. Execute the .exe file to install the toolkit. 

The tool kit should be installed in C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.1.  Go to Windows System Properties dialog to verify if environment variable CUDA_PATH is correctly set as above path; if not exist, set it as shown below.

Step 3. Install Python Package numba to the embedded Python in Origin  

Launch Origin. In the Origin workspace, select from menu Connectivity: Python Packages… If the numpy version is  newer than 1.21, we need to uninstall the numpy package first as numba has requirement numpy <1.21  and >=1.17. Select  numpy  in  the  table,  click  Uninstall  button to uninstall numpy 1.21.4.  

Click Install… button.  Type numba in the Package Names entry and click OK to install numba package. After  the  installation,  restart  Origin.  

Step 4. Test Python code.

Select from menu Connectivity: Open untitled.py…  and copy paste the following code. Press F5 key to execute.

from numba import cuda
import numpy as np
@cuda.jit
def cudakernel0(array):
    for i in range(array.size):
        array[i] += 0.5

array = np.array([0, 1], np.float32)
print('Initial array:', array)

print('Kernel launch: cudakernel0[1, 1](array)')
cudakernel0[1, 1](array)

print('Updated array:',array)

You should be able to see the result indicating the CUDA kernel is properly launched.

Initial array: [0. 1.]
Kernel launch: cudakernel0[1, 1](array)
Updated array: [0.5 1.5]

コメントを残す

メールアドレスが公開されることはありません。 が付いている欄は必須項目です