GPU Accelerated Computing with Python

川长思鸟来 2022-03-30 12:58 123阅读 0赞

https://developer.nvidia.com/how-to-cuda-[Python][]

[python][Python] is one of the fastest growing and most popular programming languages available. However, as an interpreted language, it has been considered too slow for high-performance computing. That has now changed with the release of the NumbaPro Python compiler from Continuum Analytics.

CUDA Python – Using the [NumbaPro][] Python compiler, which is part of the [Anaconda Accelerate][] package from Continuum Analytics, you get the best of both worlds: rapid iterative development and all other benefits of Python combined with the speed of a compiled language targeting both CPUs and NVIDIA GPUs.

## Getting Started ##

1.  If you are new to Python, the python.org website is an excellent source for [getting started][] material.
2.  Read this [blog post][] if you are unsure what CUDA or GPU Computing is all about.
3.  Try CUDA by taking a self-paced lab on [nvidia.qwiklab.com][]. These labs only require a supported web browser and a network that allows Web Sockets. [Click here][] to verify that your network & system support Web Sockets in section "Web Sockets (Port 80)", all check marks should be green.
4.  Watch the first CUDA Python CUDACast:  
    
5.  Install Anaconda Accelerate
6.  First install the free Anaconda package from [this location][].
7.  Once Anaconda is installed, you can install a trial-version of the Accelerate package by using Anaconda’s package manager and running conda install accelerate. See [here][] for more detailed information. Please note that the Anaconda Accelerate package is [free][] for Academic use.

## Learning CUDA ##

1.  For documentation, see the Continuum website for these various topics:
    
     *  Learn more about [libraries][]
     *  See how to use [vectorize][] to automatically accelerate functions
     *  Writing CUDA [directly][] in Python code
2.  Browse through the following code examples:
    
     *  You can download the following [IPython Notebooks][] and (after installing Anaconda Accelerate) execute them locally on your own system which has an NVIDIA GPU:
        
         *  [A NumbaPro Mandelbrot Example][]
         *  [A Monte Carlo Option Pricer Example][]
3.  Browse and ask questions on NVIDIA’s [DevTalk forums][], or ask at [stackoverflow.com][].

So, now you’re ready to deploy your application?  
You can [register today][] to have FREE access to NVIDIA TESLA K40 GPUs.  
Develop your codes on the fastest accelerator in the world. Try a Tesla K40 GPU and accelerate your development.

## Performance/Results ##

*  It’s possible to get enormous speed-up, 20x-2000x, when moving from a pure Python application to accelerating the critical functions on the GPUs. In many cases, with little changes required in the code. Some simple examples demonstrating this can be found here:
    
    1.  A MandelBrot [example][A NumbaPro Mandelbrot Example] accelerated with CUDA Python. 19x speed-up over the CPU-only accelerated version using GPUs and a 2000x speed-up over pure interpreted Python code.
    2.  A Monte Carlo Option Pricer [example][A Monte Carlo Option Pricer Example] accelerated with CUDA Python. Achieved a 30x speed-up over interpreted Python code after accelerating on the GPU.

## Alternative Solution - PyCUDA ##

Another option for accelerating Python code on a GPU is [PyCUDA][]. This library allows you to call the CUDA Runtime API or kernels written in CUDA C from Python and execute them on the GPU. One use case for this is using Python as a wrapper to your CUDA C kernels for rapid development and testing.

[Python]: http://lib.csdn.net/base/python
[NumbaPro]: https://docs.continuum.io/numbapro/quickstart
[Anaconda Accelerate]: https://developer.nvidia.com/anaconda-accelerate
[getting started]: http://www.python.org/about/gettingstarted
[blog post]: http://blogs.nvidia.com/blog/2012/09/10/what-is-cuda-2/
[nvidia.qwiklab.com]: http://nvidia.qwiklab.com/
[Click here]: http://websocketstest.com/
[this location]: http://www.continuum.io/downloads
[here]: https://store.continuum.io/cshop/accelerate/
[free]: https://store.continuum.io/cshop/academicanaconda
[libraries]: http://docs.continuum.io/numbapro/cudalib.html
[vectorize]: http://docs.continuum.io/numbapro/CUDAufunc.html
[directly]: http://docs.continuum.io/numbapro/CUDAJit.html
[IPython Notebooks]: http://ipython.org/notebook.html
[A NumbaPro Mandelbrot Example]: http://nbviewer.ipython.org/f5707335f40af9463c43
[A Monte Carlo Option Pricer Example]: http://nbviewer.ipython.org/835a8ca39ced77fe751d
[DevTalk forums]: https://devtalk.nvidia.com/default/board/57/cuda-programming-and-performance/
[stackoverflow.com]: http://stackoverflow.com/questions/tagged/numbapro
[register today]: http://bit.ly/RegDevPages
[PyCUDA]: http://mathema.tician.de/software/pycuda