Parallel Computing with GPUs

One of the key advantages of GPU computing is its ability to perform parallel computations efficiently. GPU architectures consist of thousands of cores capable of executing multiple tasks simultaneously, enabling significant speedups for parallelizable algorithms.

# Example Python code snippet for parallel computing with GPUs using CUDA
import numpy as np
from numba import cuda

# Define a CUDA kernel function
@cuda.jit
def gpu_parallel_computation(input_array, output_array):
    idx = cuda.grid(1)
    if idx < input_array.size:
        output_array[idx] = input_array[idx] * 2

# Generate input data
input_data = np.arange(1000000)

# Allocate GPU memory
input_gpu = cuda.to_device(input_data)
output_gpu = cuda.device_array_like(input_gpu)

# Configure kernel launch parameters
block_size = 256
grid_size = (input_data.size + block_size - 1) // block_size

# Launch kernel on GPU
gpu_parallel_computation[grid_size, block_size](input_gpu, output_gpu)

# Copy results back to host
output_data = output_gpu.copy_to_host()

print(output_data)