The presentation of Zen 3, the AMD architecture that will succeed Zen 2, was scheduled for later this year. This architecture was going to use the manufacturing process of 7nm TSMC, and would have improvements in performance and efficiency that would allow a considerable increase in the CPI (up to 20% compared to Zen 2) and a significant reduction in consumption.
Everything we had been seeing so far was truly promising, and placed Zen 3 as an architecture that will have everything necessary to deal a hard blow to Intel. Now, thanks to a recent leak from the prolific and well-known DigiTimes, it is said that AMD has preferred to slightly delay the presentation of Zen 3 for make the leap to the 5nm + manufacturing process of TSMC.
If this is confirmed there will be a very important jump, so much so that the improvement in terms of the CPI it could be even bigger than we imaginedAnd efficiency gains would not be left behind either. As many of our readers will know, we cannot directly compare the manufacturing processes used by TSMC and Intel, but there is no doubt that if AMD manages to launch Zen 3 in the 5nm process + the Santa Clara giant will be in serious trouble.
Zen 3 at 5nm +: why would it be a very hard blow for Intel?
We all know the situation that Intel has lived in for a couple of years. The chip giant still number one in the world of x86 processors, and it is also one of the world’s largest semiconductor manufacturers, but it is located stuck in the 14nm process for too long.
Until the arrival of Kaby Lake at Intel they had been able to maintain a strategy called “tick-tock”, focused on the change of architecture in one year, with an increase in the CPI, and the reduction of the manufacturing process in another year, without an increase in the CPI. This was truncated in 2017, a date by which Intel should have transitioned to the 10nm process. It was not possible, and they had to recycle the 14nm process used in Skylake.
This situation has continued to this day, since the recently released family of Intel Comet Lake-S processors use the 14 nm ++ process, the same that we have seen in Coffee Lake and Coffee Lake Refresh. It does not bring improvements at the IPC levelAnd neither at the level of efficiency, in fact that generation has reached the limit of silicon at 14 nm, and confirms what is an open secret, that Intel has to adopt, in general, the 10 nm process.
AMD uses a different architecture than Intel, a topic that we already had the opportunity to deal with when we talked about what the processor is and what elements make it up. The first uses a multichip module design that makes it easy to build processors with many cores, at the cost of using multiple small packages to create a “big processor.” By cons, Intel maintains a monolithic core design that integrates all the elements into a single silicon pill, and in a single package.
The decision AMD made to turn towards an MCM (multi-chip module) design with Zen was a success, since despite its limitations in terms of working frequency, it has allowed it to reach Intel in IPC, and has managed to outperform it in multithreading, cost-performance, and efficiency. Zen 2 processors are built under a much simpler design than the Core 10 series design and use the 7nm process instead of the 14nm process.
For the reason we have stated, they cannot be directly compared, but it is a fact that their efficiency is much higher than that of the Intel equivalents, which have a cooler operation, and offer a more solid value. If AMD releases the 5nm + Zen 3 process architecture in early 2021 it will be a blow to Intel because the company it will have nothing to hold on to, nothing to compete with AMD.
The launch of Rocket Lake-S is expected later this year, a generation of high-performance general-purpose processors that will be manufactured in the process of 14 nm +++, and that it will have a maximum of 8 cores and 16 threads. To improve the IPC they will use an adaptation of the Willow Cove cores, present in Tiger Lake (10 nm ++), while AMD will have on the market processors in 5 nm + with configurations of 16 cores and 32 threads that could also offer a higher IPC and greater efficiency.
When will Intel complete the transition to 10nm?
It is the question that we all ask ourselves, and unfortunately no answer, at least for now. The chip giant has said on several occasions that it would complete the transition in different years, and none of those dates has been met.
Right now the only thing that Intel has in the market in the process of 10 nm + are the processors Ice Lake U series and Y series, energy efficient variants with up to four cores and eight threads that, in addition, they are having a very limited distribution, and that they do not make a significant difference in performance.
The next jump will be Tiger lake, a generation that will come manufactured in 10 nm ++, but that will not compete directly with Zen 3, since it will be used to create low-power processors for laptops and convertible devices. They will have up to four cores and eight threads.
Following the data that the latest leaks have left us, everything seems to indicate that we will have to wait for the arrival of Intel’s Alder Lake, between 2021 and 2022, for that transition to complete. These processors will have a big.LITTLE configuration, that is, they will mount high-performance cores and high-efficiency cores, the same approach that we have seen for years in the mobile sector.
It is clear that Intel failed to achieve good success rate per wafer With the 10nm process on chips with a monolithic design and configured with more than four physical cores, it is the only reasonable explanation that we can draw from everything that has happened in the last three years.
DigiTimes Report on 2020.05.28: pic.twitter.com/50vuxqUrAi
– RetiredEngineer® (@chiakokhua) May 28, 2020