0

What is the speed of running tensorflow 2.3.0 with GPU relative to tensorflow 2.18.0 with only CPU?

Hardware

  • Laptop: MacBook Pro 15-inch 2012 64-bit.
  • OS: Windows 10 Pro 22H2
  • Processor: Intel(R) Core(TM) i7-4850HQ CPU @ 2.30GHz
  • GPU: GeForce GT 750M, compute capacity 3.0, GeForce Game Ready Driver 425.31 WHQL.

Purpose

  • Learning keras in R.

Why tensorflow 2.3.0?

  • Because my laptop's internal GPU has a compute capacity 3.0 whereas most current tensorflow applications require a compute capacity 3.5+, I had to compile my own tensorflow installer wheel file "tensorflow-2.3.0-cp37-cp37m-win_amd64.whl" and use compatible CUDA 10.1 and cuDNN 7.6 to take advantage of the existing GPU.
  • Accordingly, I need to run python version 3.7.11 (default, Jul 27 2021, 09:42:29) [MSC v.1916 64 bit (AMD64)], numpy 1.18.5, keras 2.11.0 (reported by pip list | grep keras, but keras:::keras_version() reports 2.4.0), TensorFlow v2.3.0 but not later versions.
  • The tensorflow session detects 1 CPU and 1 GPU, as listed in the hardware

What about tensorflow 2.18.0?

  • I can update to python 3.12.7, keras 3.6.0, and tensorflow 2.18.0 as of 11/15/2024.
  • But I will lose access to GPU because latest versions of software are not compatible with older GPUs.

I wonder whether I should choose tensorflow 2.3.0 with GPU or tensorflow 2.18.0 with only CPU, given my limitations of hardware capacity and software versions. I understand that some keras functions (e.g. layer_random_rotation() for image rotation and data enhancement requires Tensorflow version >=2.6) are only available in later tensorflow versions, but I have not started computer vision applications or huge data sizes (in multiple GB) yet. Does the GPU access provide enough benefits to justify skipping latest software versions and retaining the earlier versions?

DrJerryTAO
  • 101
  • 2

0 Answers0