Using Emgu TF with GPU (Commercial Release)

From EMGU
Jump to navigation Jump to search

GPU support

Emgu TF's commercial release will let you run Tensorflow graph on Windows with GPU.

Software requirements

Once the commercial license is purchased, the first step of running Emgu TF with GPU support is to download our commercial release the contains GPU support. Please make sure you have downloaded the package that contains "-cuda" in the package name. The commercial CPU release that do not contains "-cuda" in the package name do not support GPU.

Hardware and Driver requirements

  • Make sure your system can run 64-bit application. Our GPU build only support running on x64 system.
  • A NVidia GPU is required. You Graphic card also need to support the CUDA release that our commercial release is based on. For example, our Emgu TF 1.5.0 release is based on CUDA 8 and you will need a CUDA device that supports CUDA Compute Capability of at least 3.0. If you are not sure, check out this page to see if your graphic card has sufficient compute capability to support GPU mode in Emgu TF.
  • Make sure you have the latest Graphic card driver installed. e.g. our Emgu TF v1.5.0 release was tested on GeFore drive 390.65, you should make sure your graphic card driver is at least at this version when running Emgu TF v1.5.0.

How to check if GPU mode is enabled

The following function call will let you know if CUDA is enabled in Emgu TF.

TFInvoke.IsGoogleCudaEnabled

Still not able to use GPU mode?

Please email us with the following information:

  • The name of the package used. e.g. "libemgutf-windesktop_x64-cuda-1.5.0.128c.zip"
  • The Windows operation system version. e.g. Windows 10 Pro 64-bit
  • The NVidia Graphic card version. e.g. GTX 1080
  • The NVidia Driver version. e.g. 390.65

We will take a look and try our best to help.

Retrieved from "https://www.emgu.com/wiki/index.php?title=Using_Emgu_TF_with_GPU_(Commercial_Release)&oldid=2133"