GPUArray

GPUArray[array] yields an array stored in GPU memory, making it available for GPU-accelerated computation.

The input array can be:

• a packed list of real or complex numbers
• a NumericArray
• a SparseArray

Data is automatically transferred to the GPU. Use Normal to copy it back to the CPU.

Examples

Basic usage

Create a GPUArray from a vector and compute element-wise cosine:

g = GPUArray[Range[1.0, 5.0]];
Cos[g]

Retrieve the result as an ordinary list:

Normal[Cos[g]]

Matrix operations

Perform a matrix–vector dot product on the GPU:

mat = GPUArray[RandomReal[1, {500, 500}]];
vec = GPUArray[RandomReal[1, 500]];
Normal[mat . vec]

Fourier transform on the GPU

data = GPUArray[RandomReal[1, 2^16]];
Normal[Fourier[data]]

Random number generation on GPU

g = GPUArray[RandomReal[{0, 1}, 10^6, Method -> "GPU"]];
Mean[g]

Linear algebra

Solve a linear system entirely on the GPU:

a = GPUArray[RandomReal[1, {200, 200}]];
b = GPUArray[RandomReal[1, 200]];
Normal[LinearSolve[a, b]]

Arithmetic

Standard arithmetic operations are dispatched to the GPU automatically:

x = GPUArray[Range[1.0, 1000.0]];
Normal[x^2 + Sin[x]]

Please visit the official Wolfram Language Reference for more details.