I wrote this about a month ago when I completely tested the small cores of Intel Tremont and Zhaoxin Luizaui in processors.
I’m glad I wasn’t wrong with that hardware scheduler – Thread Director Technology 🙂
https://www.youtube.com/watch?v=SnRe7lcKOV8
“Indeed, the performance of small cores, and especially those from Intel, has moved into a whole new dimension than in the past.
Of course, small cores don’t really have the ability to outperform large ones by design as well as by targeting performance, but if the right alchemy that Intel apparently blended is mixed, it has potential not only in performance but also in efficiency.
Gracemont will be great for a small core. Already 6W Tremont achieves excellent IPC at the level of Ivy Bridge, while the Gracemont will have SIMD AVX, AVX2, a much higher power limit for TDP as well as a much larger usable L3 cache. Even the memory to advise DDR5 will move it somewhere else. And let’s not forget the six decoders in two clusters already in Tremont, (where one of the clusters is deactivated and the CPU only runs on three decoders), which Gracemont will probably slow down …
If Intel also implements the HW scheduler as a safeguard against cooperation with MS, it will be great.
The other successor of Alder Lak, ie. Raptor Lake will only raise those small cores up to 16 cores.
Probably in the autumn we will experience a revolution in changing the paradigms of the x86 / x86-64 world, especially in terms of efficiency.
Personally, I hope that the saving modes and profiles where you will not need max power will completely disconnect the large Golden Cove cores and leave it to the small Gracemont cores, ie. basically just some of them in one cluster. “
_______
My guess:
Gracemont IPC ~Haswell IPC (~ Centaur CNS IPC)
PS I believe that Intel keeps the consumption of Alder Lak at the very end as the high peaks will be related to the change of PARADIGMI concept x86-64 big.LITTLE and high IPC of eight large Golden Cove cores and how they can do it after decades of identical cores in the x86 world (not ) to be absorbed by reviewers and, to a large extent, by researchers (all the more so those who still longed for asymmetric multiprocessing at the turn of 2008/2009) …
In the rest, of course, with what he discussed in his blog David Ježek “How to carefully select the CPU for statistics”
source: https://diit.cz/blog/jak-peclive-vybrat-cpu-do-statistik
–
