Intel’s IDM 2.0 Strategy
To understand Intel’s Roadmap we have to understand its strategy and consequently how they intend to move the different pieces in order to strengthen their dominance over the market. We have already talked about this before in HardZone. Which is based on three different branches:
- The vast majority of products will be manufactured at Intel itself.
- External foundries will be used for certain specific products or when Intel’s production capacity is not enough for the huge portfolio of products to be developed.
- Intel will cede part of its production to third parties so that they can manufacture its chips.
It must be taken into account that in the last point it is not a relationship like that of AMD with TSMC, since Intel will be the one that develops the Full Custom libraries for use in its processors to later license them in semi-custom mode to third parties. . That is the difference with its rivals in terms of design and production, due to the fact that they have a vertical integration between design and manufacturing that allows them to take full advantage of their manufacturing nodes.
In the same way, Intel will use TSMC nodes for certain components, but the relationship will not be to create custom hardware libraries, but rather they will make use of existing ones in order to relieve stress on their foundries. One example is in Intel ARC GPUs; which are designed using TSMC’s semi-custom libraries.
Intel Manufacturing Node Roadmap
|Performance per additional watt
|Alder Lake, Raptor Lake, Sapphire Rapids y Emerald Rapids.
|Second half of 2022
|Second half of 2023
|Granite Rapids, Sierra Forest
|Second half of 2024
|Second half of 2025
|Si, High NA
|Optimized RibbonFET (GAA)
To better understand Intel’s roadmap, we need to know which are the nodes or chip manufacturing processes that they have already developed or are in process of and that will be in charge of manufacturing the chips. Future Intel Core and Xeon. That is why we have summarized it in a simple table. As for the performance per watt, this has been measured by reference to the Intel 10nm SuperFin process, so we are not talking about a jump in performance from one generation to another. In this aspect, the table indicates the level of theoretical improvement of the same architecture through all the nodes.
Intel CPU Roadmap for PC
Intel’s idea is to launch a new architecture every year, so apart from Raptor Lake for 2022, Meteor Lake in 2023 and Arrow Lake in 2024, we have both Lunar Lake and Nova Lake on the roadmap for 2024 and 2025 respectively.
As if that were not enough, Intel intends annual launches with a new manufacturing node from its move to tiles, thus taking advantage of smaller chips. Thus, Meteor Lake will make use of the Intel 4 node, Lunar Lake of the 20A node and pear finish Nova Lake of the 18A. The packaging and number of tiles or chiplets on each node will also change. 3 in the case of Meteor Lake, 4 in the case of Lunar Lake and 5 in the case of Nova Lake.
During 2022 we will see the launch of Raptor Lake, an improved version of the Alder Lake architecture that will be launched as the thirteenth generation of Intel processors, but that will bring with it a series of interesting news.
It is a processor fully compatible with the current socket for desktop Alder Lakes, but it will bring news such as a more advanced overclocking system and support for M.2 modules for Intel artificial intelligence. As for the number of cores, it will increase from 16 to 24 cores, but it will do so by increasing the number of E-Cores from 8 to 16 and not the P-Cores, so the number of execution threads will increase to 32 cores. .
Intel promises double-digit CPI improvements of at least 10%. We know that one of the novelties is a reorganization in the internal power network or PDN of the processor. Which should allow you to get higher base and boost clock speeds than current Alder Lake architecture processors. Although this is not the only reason for the performance improvement.
Intel has improved the P-Core in order to compete with Zen 4, and it will be Raptor Lake on PC along with Emerald Rapids on servers and workstations, which will give it that performance boost to compete against the Ryzen 7000 and EPYC Genoa .
The abandonment of monolithic CPUs in the Intel roadmap
We must point out that both Alder Lake and Raptor Lake, corresponding to the twelfth and thirteenth generation of Intel Core processors, are going to be the last ones to use monolithic cores. Since from Meteor Lake From now on they will choose to separate the processor into different tiles or chiplets, specifically into three different ones:
- CPU o Compute Tile: it includes the CPU with its different cores, as well as its cache structure. Compared to AMD the Compute Tile is equivalent to the CCD Chiplet. This part will be built using Intel’s most advanced manufacturing nodes.
- SoC: These are the Northbridge, memory controller, and the Southbridge, device controller, united on a single chip. Equivalent to the IOD of the AMD Ryzen. This part will be built using old or established Intel manufacturing nodes or external foundries.
- Tile GPU: it’s about what used to be the integrated GPU, but separate from the main processor. We’ve already seen this on laptop CPU variants, but it will be the first time this has been done on desktop CPUs. This part is expected to be manufactured by TSMC under its N3 node.
There are several advantages in using a processor made up of tiles or chiplets, the first of which is that it allows breaking the limit of the lattice in wafers, but especially the fact of being able to use less advanced manufacturing nodes for parts of the processor that do not require state-of-the-art technology to function, such as the memory and device controller.
Meteor Lake and beyond
And what can we highlight in the Intel Roadmap regarding Meteor Lake and what will come after? The processor will make use of the Intel 4 node in its manufacturing process, specifically in its Compute Tile whose P-Core will again receive an update at the architecture level. It is expected to be released in 2023 if there are no delays. Beyond we have Arrow Lakewhich will be the first Intel product to use its process 20A. The safest thing is that it will do it in the Tile CPU, while the rest of the components will repeat with respect to its predecessor. The last processor on the roadmap is Nova Lake manufactured at 18A.
Intel CPU Roadmap for Servers
In servers, Intel is not going to sit idly by and will make use of the same architectures and technologies that it will deploy for PCs. although with the difference that they are going to give preference to the architecture for personal computers in terms of launch dates. That is, the new manufacturing nodes will be deployed first on the Intel Core and not on the Xeon.
For this 2022 we expect Diamond Rapids, which will make use of the new P-Core cores that we have seen in the Intel Core 12, but unlike this it is not a hybrid architecture. So we are not going to see a single E-Core. Among its novelties at the server level are the adoption of DDR5 memory and the fifth generation PCIe that we have already seen on the desktop, but also the Compute Express Link or CXL.
The next stop will be in 2023 and also using the Intel 7 node. If Sapphire Rapids is a CPU built from the P-Cores of the Intel Core 12, Emerald Rapids will be the same with those of the Intel Core 13 or Raptor Lake.
Intel will not only launch servers based on its P-Core, but also intends to create server processors based on its E-Colors. Sierra Forest to be released in 2024 will be the first of them.
With regard to servers, Intel intends to launch its future Diamond Rapids architecture using its Intel 3 node, with which they promise a performance per watt 18% higher than that of the Intel 4 node. All this to build an Intel Xeon made up of 144 cores and 288 threads each.
The only case where it seems that they will break with this tradition will be with Granite Rapids, which will use the Intel 3 node exclusively. At the moment in the Intel Roadmap they have not mentioned the architectures making the 20A node or more advanced. Rather, they have not been named with a specific name, but it is known that we will also see Intel Xeon using nodes 20A and 18A.