As singular chips are getting smaller and smaller it is becoming harder for manufacturers to keep the trend going. The next GPU architecture will be based on the 7nm process but where will companies go after that? Companies are looking into other methods of boosting performance in future products and Nvidia Multi-Chip-Module GPUs could be an answer to this problem.
Nvidia is working with Arizona State University, the University of Texas, and the Barcelona Supercomputing Centre in order to make this happen and is also trying different Nvidia Multi-Chip-Module GPU designs using in-house simulators. In theory, the Nvidia Multi-Chip-Module GPUs should scale well and should provide higher SMs and higher performance as compared to current GPUs.
The idea of Nvidia Multi-Chip-Module GPUs is all great in theory but seeing how this is still in research phase you should not expect anything to come out until a few generations later. This is more of a long-term plan. Intel is also working on a new core that could help them face AMD in the CPU market in the long run.
Seeing how companies are working on drastic design changes it is possible that we will see major improvements in hardware performance in the next decade or so. AMD on the other hand already has the Zen architecture which is scalable and AMD has a multi-generational roadmap ready up till 2020. The Ryzen CPUs, as well as the AM4 platform, will be relevant for a few more years.
According to the official Nvidia website:
The need for higher performing GPUs continues to exist in many domains. To address this need, in this paper we demonstrate that package-level integration of multiple GPU modules to build larger logical GPUs can enable continuous performance scaling beyond Moore’s law. Specifically, we propose partitioning GPUs into easily manufacturable basic GPU Modules (GPMs) and integrating them on package using high bandwidth and power efficient signaling technologies.
Let us know what you think about Nvidia Multi-Chip-Module GPUs.