Intel says that if they were based on something similar to AMD Zen they would reduce their investment in R&D by 50%

Monolithic processors have their days numbered, and the future is modular processors from AMD. Lisa Su has been able to innovate in this segment by offering high quality solutions. Perfecting the process for better results has been tricky, but easier than on monolithic processors. This is acknowledged by Intel's Director of Engineering, Murthy Renduchintala, who believes that a change is needed.

Renduchintala is possibly one of the best engineers in the world and he has extensive knowledge. While his words are surprising at first glance, the future of Intel lies in the Foveros architecture. An architecture that is based on the stacking of chips, while AMD is committed to dispersing them in the encapsulation.

[amazon box="B005404P9I"]

The future passes through modularity

AMD has the upper hand in this field in a bid designed to make the design, development and manufacture of processors easy. A money saving process compared to monolithic processors. NVIDIA is already working on Hopper, which could be the world's first modular GPUs. Intel has Foveros, which is in the development phase and could arrive between 2021-2022.

Developing modular processors has many advantages, such as reducing costs. Not only are manufacturing costs reduced, since you can produce elements in various lithographs, but in R&D. The research and development process takes a lot of money and can end up creating problems. Let them tell Intel with 10nm.

With correct R&D you can obtain a better position than the competition and improve profitability. Renduchintala highlights that one of the problems they have had is monolithic integration. A problem that directly affects the logical scalability of IPs. And the fact is that the processor is not just cores and threads, there are other elements such as cache, circuits and security elements.

Renduchintala says that processing engines, mixed signals, interconnections and other elements require a laborious R&D process to work together. Contrary to what it may seem, IPs do not benefit from the logical scale of the cores, causing Intel to spend 50% of its R&D budget on all this. Something that causes that the same benefit is not obtained as the logical scale of the processing engine.

Modularity and various lithographs

Although we all associate modular processors with AMD's creation for the Zen architecture, that's not the case. A few days ago, we saw the appearance of modular Intel Xeon SoMa processors . These processors have four dies integrated into a single package and predate the Zen processors. These processors were intended for a very specific function and were only sold to industrial users.

Renduchintala says that changing the way of working and decoupling the IPs will allow for a different and exclusive lack. It also indicates that a more efficient and faster roadmap would be generated. This translates into product renewals faster and faster and with more news, come on, new processors every six months instead of every year.

Intel currently spends about $ 15.200 billion on the processor business. Going to an MCM architecture would allow for lower costs and possibly prices. One of the factors of the price reduction would be the ability to produce in several lithographs. AMD manufactures its cores in 7nm but the I / O DIE is manufactured in 14nm, significantly reducing costs.

Future MCM for Intel

That the engineering director recognizes this is very important and that although it sounds new, it is not. What it tells us is that not only server processors could become MCM, but all processors, even domestic ones. Foveros is the foundation for Intel's new processor technology combined with EMIB.

We could see a processor with a 10nm CPU and iGPU, the cache made at 22nm and the Northbridge made at 14nm, for example. This will not only facilitate the reduction of costs and therefore the reduction of prices, but also the improvement in production. It is intuited from what Renduchintala said is that integrating all the elements in 10nm is really complex and hence the problems.

We will have to wait a couple of years to see the steps that Intel takes towards the field of MCM processors. It is evident that the industry is moving towards modular processors and that monolithic processors have their days numbered.

Source: PCGN