Moore’s law is dead, or so the NVIDIA CEO claimed a few years back. So, the fact that chip miniaturization has started to near its limits doesn’t really come as a surprise. Intel didn’t bother and the results are clear, the company is struggling to move to the 10nm node with a shortage of the older 14mn CPUs. AMD has migrated to the MCM design using chiplets instead of a monolithic die, and this has greatly contributed to the success of the Ryzen CPUs, allowing HEDT levels of performance in the consumer space. Now, it seems like NVIDIA will be taking a page from AMD’s rule-book, using the mult-chip module design for its next-gen design.
This is still a rumor so take it with a grain of salt, but looking at industry trends I’d say it’s very plausible. This design suits GPUs more than it complements CPUs. The reason being that GPUs are more parallel while CPUs tend to be sequential. In simple words, CPUs pack a dozen or fewer cores and focus on single-threaded performance while GPUs consist of up to several thousand cores, running in tandem, doing different calculations simultaneously by dividing the workload between the various components.
The MCM or chiplet design induces a latency penalty which can have an impact on sequential tasks that favor higher IPC and faster response times. The result is that AMD chips are relatively slower in games compared to Intel’s. Although GameCache greatly reduces the impact of this issue, no doubt it’s still not completely eradicated.
With graphics cards, such a latency increase won’t as much of an impact as GPUs focus on parallel workloads rather than a sequential speedy pipeline. One of the main benefits of this design will be better yields, resulting in larger, more powerful GPUs and that too at much lower costs.
WCCFTech did a small test and found that shifting a medium-sized 815mm² GPU to MCM-design results in a 116% yield gain with much lower wastage. The benefits increase as we move to higher-end parts. Although there are other caveats such as load-distribution, VRAM to GPU latency and so on, but like AMD has already demonstrated with its Ryzen and Epyc CPUs that the MCM design is the future. We expect both NVIDIA and Intel to adopt it soon enough.