A patent application from Intel reveals that the company is looking at a modular design for its GPUs.
Intel wants to build future graphics cards via so-called Multi Chip Modules (MCM). This is evident from a patent application. The manufacturer is thereby going after AMD, and Nvidia also wants to embrace the technology. An MCM-GPU combines multiple graphics processors in a hierarchical structure. Those different modules then work together to deliver a single frame. Thus, the performance of a GPU goes up without having to drastically increase the complexity of the graphics processors themselves.
Workload sharing
The basic principle is not that complex. One GPU module is in charge. That one analyzes the frame to be rendered for so-called visibility data. The chip sees which objects in a scene are visible and which are not, without rendering them. That’s a quick and easy job. Then the chip generates a grid based on that initial analysis. The different boxes of the grid are then assigned to the other GPU modules, which work together to render the frame. Think of the concept as an extended version of SLI or Crossfire, but on the GPU itself and optimized.
read also
Intel also wants Multi Chip Modules in its GPUs
Intel is not giving away any further details, but in doing so it gives a nice indication of the direction it wants to go with its own GPUs. The company has been working for some time on its own graphics cards to compete with AMD and Nvidia. You can already taste the fruits of that development when you have a Tiger Lake (or Alder Lake) chip in your laptop. Some modest discrete cards have also already appeared on the market, but 2022 will really be the year of the Intel GPU. It is unlikely that those cards will already use MCM technology, although Intel will likely continue to develop the architecture to work with future module-based GPUs.