AMD Zen 6 and Intel Meteor Lake processors may become more interesting than they seem now. If only the technical process is previously known about Zen 6, then Intel Core 14 generation will probably appear in a new unusual layout of the main elements.
AMD Zen 6 is a 2nm process and a possible change of manufacturer
An AMD engineer’s profile on LinkedIn revealed the codename and technical process of AMD Zen 6 chips. In 2024, their predecessor Zen 5 should appear, the microarchitecture of which will not appear in its final form until the summer. Much less is known about Zen 6, its development still has a long way to go before we can talk about details.
| Microarchitecture | Core | CCD | Process Node | Launch date |
|---|---|---|---|---|
| Zen 6 | Morpheus | ? | 2 nm | ? |
| Zen 5 | Nirvana | Eldora | 3 nm | 2024 |
| Zen 4 | Persephone | Durango | TSMC 5nm | 2022 |
| Zen 3 | Cerberus | Breckenridge | TSMC 7nm | 2020 |
| Zen 2 | Vahalla | Aspen Highlands | TSMC 7nm | 2019 |
| Zen+ | ? | ? | GlobalFoundries 12nm | 2018 |
| Zen | ? | ? | GlobalFoundries 14nm | 2017 |
The information was posted on the profile of Md Zaheer, who is apparently a senior chip development engineer at AMD. He worked on the power management of the Zen 4, Zen 5 and Zen 6 processors. Information about the processors is now hidden, but Twitter user Maro David took a screenshot of the original description (the tweet has also been removed now).
The Zen 6 architecture cannot be found in any of AMD’s official roadmaps, the manufacturer has only shared plans for Zen 5. The latest roadmap, dated June 2022, revealed that AMD expects to release Zen 5 in 2024. Zen 6 will most likely appear on the market no earlier than 2025.
According to Zaheer, who started working on Zen 6 earlier this year, the internal code name for AMD Zen 6 cores is Morpheus. It is reported that the Zen 6 chips will be manufactured using a 2nm process. Therefore, it is difficult to say whether AMD will continue to use TSMC for CPU production or switch to Samsung – both companies expect 2nm technology to be ready in 2025.
Intel Meteor Lake – L4 cache and no iGPU access to LLC
The new Linux patch implies that Intel’s 14th generation (Meteor Lake) processor will get L4 cache, which is not often used in CPUs. The patch description reads: “On MTL, GT can no longer highlight [ресурсы] on LLC – it can only CPU. This, along with adding ADM/L4 cache support, requires updating the MOCS/PAT table.” Nowhere in the description does it say what ADM is – it could be a marketing name for L4 cache. The size of the fourth level cache is also not disclosed.
During the Hot Chips 34 presentation, Intel shared the schematic of the Meteor Lake chip, which consists of four wafers: CPU, SoC, I/O, and integrated graphics. They are located on a large base plate that connects all the elements. The question is where Intel will place the ADM/L4 cache. It can be either on the base plate or inside the SoC, which may explain its considerable size.
Course
Game design for beginners
You will be taught by the person who integrated Chuck Norris into the feature for the World of Tanks project
Cool
Most processors are limited by the level of the L3 cache – the transfer rate decreases as the level increases. So, outside of interfacing with integrated graphics, there are very few scenarios where the L4 cache helps improve system performance.
Another interesting discovery is that Intel seems to have blocked the Meteor Lake graphics core from accessing the LLC (usually the L3 cache). According to the patch notes, “On MTL GT no longer allocates on LLC, set has_llc = 0”. In most modern LLC processors, the CPU and iGPU cores are shared. The described design is similar to the Ryzen console APU, where the compute and graphics cores do not use the same cache. Intel makes separate memory for integrated graphics?
Tenstorrent Head Predicts 30% AMD Zen 5 Performance Growth and CPU Leadership in Integer Computing in 2024-2025
Sources: Tom’s Hardware and again Tom’s Hardware
