I have added a GPU but the project doesn't use it
When you add a GPU, you need some requirements for BOINC to use it:
- Drivers from the GPU manufacturer.
- A BOINC project capable of using this GPU.
- Telling the project & BOINC that it's all right to use the GPU.
While Windows will probably install drivers for your GPU, these drivers are usually lacking the necessary components to do calculations on that hardware. This is done to make the drivers be available for a wide range of hardware. It's also that a lot of the components are open source, a thing that Microsoft doesn't like, and most of these components are direct competitors with things that Microsoft makes:
- OpenGL -> DirectX
- CUDA/OpenCL -> DirectComputing
So it is better to use the drivers from the GPU's manufacturer, being AMD (or ATI), Intel or Nvidia:
- AMD drivers
- Intel drivers
- Nvidia drivers
Wherever, try to make a clean installation of the drivers. This may be an option in the driver installer (mostly in the case of Nvidia drivers, and it's hidden away in a non-obvious place so read all the options carefully), but it may also require you to uninstall the previous drivers via Device Manager or Uninstall a program.
A GPU capable BOINC project
To check if a BOINC project is capable of using your GPU, you can check the list of BOINC projects, via the BOINC web site or via the Join Projects wizard.
If you want to know if your GPU is capable of doing these calculations, you can check various lists available on the internet. But just simple rule of thumb is:
- AMD/ATI GPU: Radeon HD 4000 and higher have OpenCL capability, but better to use is HD 5000 and above.
- Nvidia: to use CUDA, you need at least a Geforce 8800. All Nvidia GPUs are OpenCL capable (*).
- Intel: the minimum requirement is a 3rd generation Intel iCore CPU with HD Graphics. (**) For a list see here
(*) Because OpenCL is in a way a direct competitor to Nvidia's CUDA, Nvidia decided at one time to develop only CUDA and no longer add the newest OpenCL to their GPUs. Therefore it is possible you have a top-of-the-range latest model Nvidia GPU with just OpenCL 1.2 capability, while OpenCL is already in version 3 or 4.