From games to computing
The 1990s brought 3D accelerators such as the 3dfx Voodoo. The GeForce 256 handled transformation and lighting on the card, and NVIDIA called it a GPU. It soon turned out that thousands of small cores were useful for more than games.
CUDA and AI
NVIDIA released CUDA, which let GPUs be programmed for general computing. In 2012 the AlexNet neural network was trained on two gaming cards and started the deep learning revolution. Today NVIDIA data-centre chips are the most sought-after hardware in the world.
Choosing a card
- Resolution and games decide, not just the model number.
- Watch the video memory (VRAM): 8 GB is little for new games today.
- A strong card needs a quality power supply and good case cooling.
- Technologies such as DLSS and FSR boost performance with AI; see Ray tracing and DLSS.