Why GPU-acceleration makes sense

Why GPU-acceleration makes sense#

In this notebook we demonstrate speedup through GPU-acceleration using a Gaussian blur filter. We’ll be using the clEsperanto library which uses OpenCL and is compatible to a wide range of Intel, AMD and NVidia GPUs. Feel free to run it on your GPU and measure the speedup!

See also

Note: benchmarking results vary heavily depending on image size, kernel size, used operations, parameters and used hardware. Use this notebook to adapt it to your use-case scenario and benchmark on your target hardware. If you have different scenarios or use-cases, you are very welcome to submit your notebook as pull-request!

import pyclesperanto_prototype as cle
from skimage import filters
import time

# to measure kernel execution duration properly, we need to set this flag. It will slow down exection of workflows a bit though
cle.set_wait_for_kernel_finish(True)

# selet a GPU with the following in the name. This will fallback to any other GPU if none with this name is found
cle.select_device('RTX')

<GeForce RTX 2080 Ti on Platform: NVIDIA CUDA (1 refs)>

# test data
import numpy as np

from skimage.io import imread
test_image = imread('Lund_000500_resampled-cropped.tif')

sigma = 10

# convolve with scikit-image
result_image = None

for i in range(0, 10):
    start_time = time.time()
    result_image = filters.gaussian(test_image, output=result_image, sigma=sigma)
    print("skimage Gaussian duration: " + str(time.time() - start_time))
    

skimage Gaussian duration: 0.644662618637085
skimage Gaussian duration: 0.63631272315979
skimage Gaussian duration: 0.6193966865539551
skimage Gaussian duration: 0.6499156951904297
skimage Gaussian duration: 0.6301307678222656
skimage Gaussian duration: 0.6531178951263428
skimage Gaussian duration: 0.6489198207855225
skimage Gaussian duration: 0.6308994293212891
skimage Gaussian duration: 0.7410404682159424
skimage Gaussian duration: 0.8148434162139893

# convolve with pyclesperanto
result_image_gpu = None

test_image_gpu = cle.push(test_image)

for i in range(0, 10):
    start_time = time.time()
    result_image_gpu = cle.gaussian_blur(test_image_gpu, result_image_gpu, sigma_x=sigma, sigma_y=sigma, sigma_z=sigma)
    print("pyclesperanto Gaussian duration: " + str(time.time() - start_time))

pyclesperanto Gaussian duration: 0.026170730590820312
pyclesperanto Gaussian duration: 0.002056121826171875
pyclesperanto Gaussian duration: 0.015659093856811523
pyclesperanto Gaussian duration: 0.019225597381591797
pyclesperanto Gaussian duration: 0.01566314697265625
pyclesperanto Gaussian duration: 0.015616178512573242
pyclesperanto Gaussian duration: 0.01566910743713379
pyclesperanto Gaussian duration: 0.015576839447021484
pyclesperanto Gaussian duration: 0.01562190055847168
pyclesperanto Gaussian duration: 0.023794889450073242

Let’s just check if the results look similar

import napari

viewer = napari.Viewer()
napari.run()

viewer.add_image(test_image)
viewer.add_image(result_image)
viewer.add_image(result_image_gpu)

<Image layer 'result_image_gpu' at 0x1d3b659c460>

napari.utils.nbscreenshot(viewer)

../_images/4a0c06150aa306a63bbfd3e87118e17c6c574806176ff96bb857b24264cbcb40.png