Welcome to the quick setup guide created to help you utilize CUDA 11.8 for enhanced image processing performance with our AI powered products. After this tutorial, you should be able to use your NVIDIA GPU to speed up processing from minutes to just seconds.
Step 1: Checking GPU Compute Capability
In order to utilize CUDA with our programs, your GPU needs to have the proper compute capability to run the PyTorch binaries.
Checking your compute capability
1. Find your GPU on this website.
2. Ensure your GPU has a compute capability of at least 3.7. If not, you will be unable to utilize CUDA acceleration.
Step 2: Verifying Your NVIDIA Driver Compatibility
Since our programs are built with PyTorch and come bundled with the necessary CUDA toolkit, your primary focus should be on ensuring your NVIDIA driver is compatible with CUDA 11.8. I will provide a download link later in the tutorial to a folder containing the needed PyTorch binaries.
Checking Your Current Driver Version
1. Open the NVIDIA Control Panel on your computer by right clicking on your desktop.
2. Go to the “Help” menu and select “System Information”.
3. Look for the “Driver Version” in the System Information dialog.
4. If your driver version is below 450.80.02, you will need to update.
Updating Your NVIDIA Driver
If your driver version does not support CUDA 11.8, you’ll need to update it:
1. Visit the NVIDIA Driver Downloads page.
2. Select your GPU model and operating system.
3. Download the latest driver that supports CUDA 11.8 (>=450.80.02).
Step 3: Downloading and replacing with the GPU-compatible "torch" folder
Click on the following link to download the GPU-compatible “torch” folder: torch_gpu
Locating the product install directory
1. Navigate to the directory where the product is installed on your system. This is typically found in ‘C:\Program Files\{product}’ by default.
2. In the install directory, find the _internal folder.
3. Within the _internal folder, locate the existing “torch” folder. This folder contains the default CPU-only version of PyTorch.
4. Rename the existing “torch” folder to “torch_backup” to keep it as a backup.
5. Extract the downloaded GPU-compatible “torch” folder and place it into the _internal folder.
Step 4: Testing the functionality
1. Launch the program and perform a simple task to ensure that the application is functioning correctly with the new GPU-compatible “torch” setup.
2. If you encounter any issues, revert to the backed-up “torch_backup” folder by renaming it to “torch” and contact zach@nebulosityai.com for further assistance.