The 38-core Xeon Ice Lake has not been abolished, there will also be a 40-core

Looking at the history of the news, it is clear that the first report explicitly mentions 10nm Xeon Ice Lake appeared on our website on June 25, 2018. It would probably not be extra interesting in itself. But the article was not a translation of Intel’s official statement, but a translation of AMD’s statement. Forrest Norrod then announced the 7nm Epyc Rome (Zen 2) was designed as a competition for Xeons Ice Lakebut will not stand against them. More than 2.5 years have passed and Norrod’s words are confirmed. While Intel is still preparing for the Xeony Ice Lake, AMD has completed and released Epyc Rome (Zen 2), which will replace Epyc in a few weeks Milan (Zen 3).

About Xeons Ice Lake were briefly referred to as up to 38 – core processors, but there have been no reports of more than 28 – core models for quite some time, which logically raised concerns about Intel ‘s future. The good news that leaker momomo_us brings is that it counts:

We even learn about the 40-core model, so it looks like the 38-core assumed deactivation of the two cores to increase yield. However, Intel will probably be interested in using a fully active configuration, which can move Xeon 5% higher in tests and reviews.

  • 2.7x density scaling vs 14nm
  • Self-aligned Quad-Patterning
  • Contact Over Active Gate
  • Cobalt Interconnect (M0, M1)
  • 1st Gen Foveros 3D Stacking
  • 2nd Gen EMIB

The above is a list of technologies that Xeony Ice Lake is to be equipped with according to WCCFTech.

The situation in which Intel has been slapping for several years is primarily due to the delayed development of 10nm production. This is true, a well-known fact, but it is reasonable to doubt whether it is the only one. For years, we have heard that Intel has a number of architectures ready in the drawer, which are just waiting for an available process. But it seems more and more that the process will not be the only problem.

Let’s take a desktop example Rocket Laketo be released in March. It builds on the same cores as Ice Lake (architecture Sunny Cove) to be completed in the years. Rocket Lake are from Ice Lake differs by process, uses 14nm. It has been completed for over five years. Why does a product built on years of ready-made technologies enter the market in the spring of 2021 and still with a delay? We can say that Intel simply did not think to combine the finished technologies and wasted time waiting for a 10nm process before it was attacked by the kernel Sunny Cove produce on 14 nanometers.

But what doesn’t explain why Ice Lake, published in 2019 in notebooks, is waiting for release in servers by 2021. If Intel has finished architectures in a drawer and expects the 10nm process to be ready for Xeon production in early 2021, why does it get the green generation that was obsolete two years ago and not something built on newer architecture?

A certain clue may be the just quoted list of technologies used to implement Xeons Ice Lake. These require, among other things, Foveros layering technology and the second generation of EMIB jumpers. Let’s sum it up once again: Intel needs technologies to make a 40-core processor that AMD doesn’t even need to implement a 60% higher-core processor with 10% higher IPC, which also gets the same TDP that Intel needs for a significantly less equipped configuration.

Even if Intel’s 10nm process were already in a state in 2019 that Xeons could be manufactured on it, Intel would not be able to manufacture them, as the mentioned technologies were then at the level of the first experimental products (nothing greater than Lakefield pro Foveros a Kaby lake-g for the first generation of EMIB bridges).

In fact, Intel has built its product on technologies that are complicated to implement the product for a given market and performance segment. The successful implementation of several such technologies conditioned the release of a product which in practice will have a problem with the competitiveness of a more powerful product, which does not require any special jumpers or an exclusive encapsulation method.

Pouzdro Foveros u SoC/SiP Lakefield

I dare not guess whether Intel has launched the development of these technologies with the idea that one day it will need them for further meaningful development – and on Ice Lake they came out as Black Peter, in short, it was time to use them in something to justify the funds spent on their development. Or, indeed, at the beginning of development, Intel believed that it would not be able to achieve the given performance goals other than using these technologies, and it waited for them (after the process) for years.

EMIB bridge for connecting chipsets

Whether one option or the other is closer, it seems that Intel (at least at the time these decisions were concerned) has lost the ability to solve problems in the simplest way possible. Instead, he developed technologies that were not needed to achieve the set goals. It is possible that Intel has recovered from this flight. For example, the computational Xe HPC looks like a product that was created in an effort to produce something that would not have been possible without these technologies. This can be good with regard to secondary use (in terms of Ice Lake) of unnecessarily developed technologies. On the other hand, it is also a product for which it means successful serial production every This technology is another condition in terms of its industrial mastery and another possible bottleneck in terms of production capacity. It is sufficient that any of these technologies become a constraint on production capacity or yield and that series production of the product is jeopardized.

Intel has obviously chosen a more difficult path. Whether it’s good or bad, only time will tell. In the short term, this has led to a number of delays and some reduction in competitiveness. However, Intel has the funding to bridge these issues. Only the long-term horizon will show whether these technologies will significantly help it increase competitiveness in future generations of products (as opposed to AMD, which currently chooses an easier path), or whether the risks associated with them will accumulate to what Intel sees as unsustainable. strategy change. Finally, with regard to the development time of new generations, we do not know whether simpler (KIS – “Keep It Simple”) solutions are no longer being created on drawing boards. Or at least solutions that do not require even more complex housing and connection systems to implement, which have not yet been developed. After all, it also depends on the company’s philosophy: Is the goal an interesting technology or a successful product?


Source: Diit.cz by diit.cz.

*The article has been translated based on the content of Diit.cz by diit.cz. If there is any problem regarding the content, copyright, please leave a report below the article. We will try to process as quickly as possible to protect the rights of the author. Thank you very much!

*We just want readers to access information more quickly and easily with other multilingual content, instead of information only available in a certain language.

*We always respect the copyright of the content of the author and always include the original link of the source article.If the author disagrees, just leave the report below the article, the article will be edited or deleted at the request of the author. Thanks very much! Best regards!