AMD has come out with a new patent that will make it unrivaled in the GPU world. This published patent describes a GPU design split into multiple GPU chiplet sets. So what does multi-chiplet mean? Here are the details of AMD’s new patent…
AMD’s GPU patent pushes boundaries with multi-chiplet design
This patent from AMD splits the GPU design into multiple chiplet sets. Each chiplet is paired with a frontend die and several shader engine dies. These chiplets can be combined in various configurations.
For example, chiplet sets can come together to function as a single unified GPU or operate as independent GPUs. Some chiplet sets can work as a unified GPU, while others operate independently. This modular GPU design allows scaling of GPU resources and performance according to product needs or operating modes.
This flexibility allows various GPUs to be assembled with a small number of tape-outs using different numbers of chiplets. Additionally, this chiplet design allows the most critical components to be produced with the latest technology, while the supporting frontend logic can be manufactured with older and cheaper silicon.
AMD has already applied this new chiplet philosophy in a simple way in some RDNA 3 GPUs. However, this patent goes further, enabling various chiplet combinations and configurations. There are rumors that Nvidia will use chiplets in a similar approach for the GeForce RTX 5000 series. AMD’s patent could be part of future trends in chip designs and open the door to significant innovations in the GPU world.
What do you think about this new GPU design and multi-chiplet? How do you think AMD’s innovative approach will impact the GPU market? Don’t forget to share your thoughts in the comments section below…
{{user}} {{datetime}}
{{text}}