Nvidia GPU Installation Guide
If you want to use Nvidia GPU acceleration for subtitle extraction in SubsFast, several prerequisites are needed:
1. Your OS should be Windows 64-bit version.
2. CUDA Toolkit must be installed in your OS.
After satisfying the above conditions, you can open the switch in the figure below to use the GPU for acceleration during subtitle extraction.
How to install CUDA Toolkit?
1. Download the installation package
CUDA 12.1.1+ is required, you can download CUDA from link below.
2. Start installation
The installation process has certain requirements for the size of the system disk (usually C:), please reserve at least 4G of disk space, otherwise the installation may fail.
Click on the installation package and you will see the following interface, click "Agree and Continue".
Select "Custom Install" - "Next".
Only check "CUDA" and "Driver components". Then "Next".
The Display Driver under Driver components is the graphics card driver, which means that when installing CUDA, the graphics card driver will also be upgraded.
The 536.23 in the figure indicates the currently installed graphics card driver version on the system, and 537.13 indicates the graphics card driver version after installation.
3. Verify installation status
To confirm the installation results are successful, you can verify as follows. But if you feel there's no problem, you can completely skip this step and directly try using GPU acceleration in SubsFast. If subtitle extraction is normal, it means the installation was successful.
First, open "Command Prompt(CMD)".
1️⃣ Press the Windows logo + R keyboard shortcut to open the "Run" window
2️⃣ Enter "cmd" and press Enter, and a black command prompt will appear
Before installation, I had executed the following commands in the command line to get some pre-installation status.
First, enter "nvidia-smi" in the command line and press Enter to see results similar to below.
You can see the current graphics driver version from the figure above, as well as the recommended CUDA version corresponding to this driver version. It does not necessarily have to be this version - generally a slightly lower CUDA version can also be compatible with a slightly higher graphics driver version.
Then, enter "nvcc -V" in the command line and press Enter to find out the currently effective CUDA version through the return result, and confirm it is the same version you just installed, indicating successful installation.
After installation, I execute "nvidia-smi" again for comparison - the graphics driver has been updated.
Then I execute "nvcc -V" again for comparison - the currently effective CUDA version has been upgraded from the previous 11.8 to 12.2, indicating this installation was successful.
At this point, you can start normally using the GPU acceleration function of SubsFast. If problems still exists, feel free to contact us by email.