Installing the CUDA Deep Neural Network library

The CUDA Deep Neural Network (cuDNN) library enables Echoview to utilize graphics processing unit (GPU) acceleration for its machine learning (ML) operators. GPU acceleration has the potential to provide significant performance improvements. Refer to the table in Machine learning operator performance for a comparison.

Besides possessing the right hardware, you must install CUDA and cuDNN to harness GPU acceleration. Follow this guide to download CUDA 11.2.0 (2.9 GB) and cuDNN 8.1.0 (665 MB) and install them.

Note that this procedure can take some time. The files to download are relatively large and there are several stages involved in the sequence. We recommend you read through all the steps to understand them before proceeding.

• Before starting
• Check if you have a CUDA compatible GPU for cuDNN
• Obtain administrator permissions for the PC
• Register for the Nvidia Developer Program
• Remove unsupported versions of CUDA and cuDNN
• Remove Nvidia Frameview SDK
• Installing CUDA and cuDNN
• Verifying the installation with Echoview
• Troubleshooting

Before starting

Check if you have a CUDA compatible GPU for cuDNN

From the Start menu open the Device Manager and expand the Display Adapters drop-down menu to determine your GPU.

If it is listed on https://developer.nvidia.com/cuda-gpus your PC supports GPU acceleration for cuDNN.

Obtain administrator permissions for the PC

You may not be able to complete the installation without administrator access.

Register for the Nvidia Developer Program

Nvidia requires you join their developer program to download cuDNN. Create an account at https://developer.nvidia.com/developer-program.

Remove unsupported versions of CUDA and cuDNN

Echoview requires CUDA 11.2.0 and cuDNN 8.1.0 for GPU acceleration with the version of TensorFlow^TM used with the ML operators.

If you have other versions installed on your PC, remove them before continuing. Follow the uninstallation instructions on

• https://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html#uninstall-cuda-software, and
• https://docs.nvidia.com/deeplearning/cudnn/install-guide/index.html#upgrade-windows

and restart your PC.

Remove Nvidia Frameview SDK

You may need to remove the Nvidia Frameview SDK software if it is installed. Do this via the Control Panel with Add or remove programs.

Installing CUDA and cuDNN

1. Close any instances of Echoview that are open
2. Update your Nvidia graphics drivers
• Visit https://www.nvidia.com/Download/index.aspx to download and install the latest drivers for your GPU. You may need to restart your PC afterward.
• Remove Nvidia Frameview SDK if it came bundled with the update for the driver.
3. Download and install CUDA 11.2.0
• Download CUDA 11.2.0 from https://developer.nvidia.com/cuda-11.2.0-download-archive. If needed, select your operating system (Windows) and version (e.g., 10), architecture (x86_64) and the exe (local) installer type.
• Open the downloaded executable file and follow the instructions. You may omit Visual Studio when the installer prompts you.
• Remove Nvidia Frameview SDK if it came bundled with the CUDA installer.
4. Download and install cuDNN 8.1.0
1. Visit https://developer.nvidia.com/rdp/cudnn-archive.
2. Select the Download cuDNN v8.1.0 (January 26th, 2021), for CUDA 11.0, 11.1 and 11.2 option from the list, then choose cuDNN Library for Windows (x86). Use your credentials for the Nvidia Developer Program to initiate the download.
3. Extract the zipped contents.
4. Create the C:\Program Files\NVIDIA\CUDNN\v8.1\ directory on your PC, and copy the extracted bin\ lib\ and include\ folders from the previous step to this directory.
5. Download and install Zlib (a data compression software library used by cuDNN)
1. Download http://www.winimage.com/zLibDll/zlib123dllx64.zip.
2. Extract the zipped contents.
3. Copy the extracted dll_x64\ folder from the previous step to the C:\Program Files\ directory.
6. Update your PATH variable to include cuDNN and Zlib
1. From the Start menu type Run and hit Enter.
2. Type control sysdm.cpl and hit Enter to open the System Properties dialog box.
3. On the Advanced tab, click on the Environment Variables... button.
4. In the User variables section, double-click on Path to open the Edit environment variable dialog box.
5. Click New and input C:\Program Files\NVIDIA\CUDNN\v8.1\
6. Click New and input C:\Program Files\dll_x64\
7. Click OK to close the dialog boxes in the above steps to save your changes.

For more information, refer to https://docs.nvidia.com/deeplearning/cudnn/install-guide/index.html#install-windows.

Verifying the installation with Echoview

Close all applications that are currently using the GPU.

Open the Windows Task Manager and on the Performance tab, select your Nvidia GPU from the list of system resources. The usage statistics will be at or close to 0% for this resource.

In Echoview, open (or zoom into) an echogram of a ML operator. If your installation worked, the graph(s) will show an increase in resource utilization for your Nvidia GPU.

Troubleshooting

• Review the procedure in Installing CUDA and cuDNN, particularly the
• versions for CUDA and cuDNN
• paths to Zlib and cuDNN 8.1.0
• Toggle the High performance custom graphics settings for Echoview (Windows 10 onward)
1. Open Windows Settings
2. From the System page, click on Display > Graphics
3. Navigate to and add Echoview.exe (Refer to the Installing Echoview and Utilities page for executable file location.)
4. Then select Echoview and click on the Options button
5. On the Graphics preference dialog box, choose High performance for your Nvidia GPU
6. Click Save
• Open the NVIDIA Control panel on your PC and verify the status of your GPU
• Restart your PC

If you are still encountering problems, contact support@echoview.com.

See also

About Machine Learning in Echoview