GPU Computing in MATLAB
To speed up your code, first try profiling and vectorizing it. For information, see
Performance and Memory. After profiling and vectorizing, you can also try using your computer’s GPU to speed up
your calculations. If all the functions that you want to use are supported on the GPU, you
can simply use
gpuArray to transfer input data to the GPU, and call
gather to retrieve the output data from the GPU. To get started with GPU
computing, see Run MATLAB Functions on a GPU.
For deep learning, MATLAB® provides automatic parallel support for multiple GPUs. See Deep Learning with MATLAB on Multiple GPUs (Deep Learning Toolbox).
Key Functions and Classes
|Number of GPU devices present|
|Table of properties of GPU devices|
|Reset GPU device and clear its memory|
|Wait for GPU calculation to complete|
|Time required to run function on GPU|
|Determine if gpuArray or CUDAKernel is available on GPU|
|Determine whether input is |
|Apply function to each element of array on GPU|
|Apply function to each page of distributed or GPU array|
|Control random number generation for GPU calculations|
|Random number stream on a GPU|
|Query and set forward compatibility for GPU devices|
Run MATLAB Code on GPU
- Run MATLAB Functions on a GPU
Hundreds of functions in MATLAB and other toolboxes run automatically on a GPU if you supply a
- Identify and Select a GPU Device
This example shows how to use
gpuDeviceto identify and select which device you want to use.
- GPU Support by Release
Support for NVIDIA® GPU architectures.
- Establish Arrays on a GPU
gpuArrayin MATLAB represents an array that is stored on the GPU.
- Using FFT2 on the GPU to Simulate Diffraction Patterns
This example uses Parallel Computing Toolbox™ to perform a two-dimensional Fast Fourier Transform (FFT) on a GPU.
- Run MATLAB Functions on Multiple GPUs
This example shows how to run MATLAB code on multiple GPUs in parallel, first on your local machine, then scaling up to a cluster.
- Train Network Using Automatic Multi-GPU Support (Deep Learning Toolbox)
This example shows how to use multiple GPUs on your local machine for deep learning training using automatic parallel support.
Improve Performance on GPU
- Improve Performance of Element-wise MATLAB® Functions on the GPU using ARRAYFUN
This example shows how
arrayfuncan be used to run a MATLAB® function natively on the GPU.
- Improve Performance of Small Matrix Problems on the GPU using PAGEFUN
This example shows how to use
pagefunto improve the performance of applying a large number of independent rotations and translations to objects in a 3-D environment.
- Measure and Improve GPU Performance
Use benchmark tests in MATLAB to measure the performance of your GPU.
- Benchmarking A\b on the GPU
This example looks at how we can benchmark the solving of a linear system on the GPU.
- Profile Your Code to Improve Performance
Use the Profiler to measure the time it takes to run your code and identify which lines of code consume the most time or which lines do not run.
Revise loop-based, scalar-oriented code to use MATLAB matrix and vector operations.
- Random Number Streams on a GPU
Control the random number streams on a GPU to generate the same sequences of random numbers as on the CPU.
- Generating Random Numbers on a GPU
This example shows how to switch between the different random number generators that are supported on the GPU.
- Stencil Operations on a GPU
This example uses Conway's "Game of Life" to demonstrate how stencil operations can be performed using a GPU.