At the Intel Architecture Day in 2018, the company talked about the design changes it will make in future products. Someone told us that in the future, Intel will design the CPU in a way that allows it to deploy CPUs flexibly, instead of strictly locking a given CPU design on a process node. Although the company has not announced publicly, the idea that Intel may adopt a 10-nanometer CPU and back transplant it to 14-nanometer has surfaced to solve its dilemma.
During the AMA for the Rocket Lake platform, Intel revealed that the CPU back-end project to create Cypress Cove was completed in the first quarter of 2019. The reason why Intel targeted the Sunny Cove processor in Ice Lake instead of the more advanced Willow Cove chip in Tiger Lake is that the TGL design has not yet been completed. Rocket Lake was delivered at the end of the first quarter of 2021, which shows that Intel spent about two years building a new floor plan and backporting the core.
This timetable provides you with a window into how the chip design cycle works. In the first quarter of 2019, Intel’s high-end chip is 9900K, and the company still has its own solid performance advantage at the top of the CPU market (especially in the gaming field). Intel has made a green light transformation of Cypress Cove, and began to work hard to port Ice Lake to 14nm, while preparing Core i9-9900KS (launched in October 2019) and the 10th generation Core i9-10900K (launched in May 2020) .
This is how semiconductor manufacturing tends to work. AMD is currently delivering Zen 3, finalizing Zen 4, and working on Zen 5. Intel just launched Rocket Lake with Alder Lake later in 2021, and is reported to launch 7nm follow-up products in 2023. This means that both Intel and AMD are choosing design features and goals. According to them, the competition situation will appear in 1-2 years.
AMA pointed out that the eight-core chip represents the largest chip Intel can manufacture + UHD graphics, without clarifying whether this is related to the design of the LGA1200 socket or other product limitations. We believe that the years since the launch of Ryzen illustrates the factual difference in product design methods between AMD and Intel.
Although not every AM4 motherboard supports every Ryzen CPU, AMD has proven that its platform design has a longer life cycle and better upgrade path than Intel, although their respective revenues and resources vary greatly.If AMD can design AM4 to support the conversion from a monolithic die to a small chip with a central I/O die with If the number of CPU cores doubles, then Intel has no reason to permanently find that its products run out of margins so easily-unless they do so deliberately. We continue to suspect that the limiting factors of Rocket Lake are related to heat and power. Compared with the eight-core RKL + integrated Xe GPU, Intel is definitely capable of making larger chips. Claiming that it can’t issue misinformation about Intel’s manufacturing capabilities without mentioning the limiting factors-unless you think the same company that managed to build a 28-core monolithic Xeon cannot manage 10-core desktop CPUs with IGPs.
Pat Gelsinger promised to first introduce Alder Lake to desktop computers by the end of 2021. If Intel sticks to this timetable, it will be the fastest desktop platform ever to be eliminated. In contrast, Rocket Lake’s stack is as low as possible, but in terms of power consumption, the top chips compete fast and are thirsty to not compete.
Some of the interesting examples we have seen here show how a company can make a series of reasonable decisions in 2019 and find itself under a lot of pressure on the final product two years later. However, Alder Lake has developed rapidly after Rocket Lake, and this fact also shows the scale and strength of becoming Intel. Intel has maintained the work of multiple engineering teams on multiple chips, so that it (probably) can start a new desktop replacement CPU in less than a year after launching the first desktop replacement CPU. This transition represents the assurance that I have made for failed launches, and we see that this does not happen often.
The closest analogy I can think of is in June and August 2002. AMD launched the 130nm purebred core Athlon XP in June, the so-called “T-bred A”. The chip is very hot and does not bring too much clock improvement to AMD. Two months later, AMD introduced Thoroughbred B, which is a new version of the chip with a 9th metal layer and higher frequencies. The highest frequency of T-Bred A CPU is 1.8GHz, but the CPU frequency of Purebred B can reach 2.25GHz, which makes AMD even stronger against Intel in the second half of this year. In any case, Rocket Lake occupies a strange place in the pantheon of Intel desktop CPUs, and may have encountered difficulties in adapting the CPU design to process nodes that it was not built for, which is somewhat inadequate in the objective curriculum.