Intel Alder Lake-U Review - The fine line between Efficiency and insufficient Performance

[NikoB]

AVX512 is supported by all new Intel server lines. It's just that in the consumer segment, they, with their lagging technical processes, in the AVX512 mode of operation, could not cope with heat dissipation.

If these instructions really speed up at least some classes of tasks, in Meteor Lake Intel will most likely return the AVX512 to the consumer series. Well, or in the next generation after this.

It is the x64 model from AMD that you are currently using in your processor from Intel. They are doing things with varying degrees of success. But now AMD is undoubtedly the leader in innovation, although they only introduced the AI ��module in Zen4, when Intel already has its third generation of NeuroDSP ...

AMD has the weakest point, as I have written many times - a terribly slow memory controller. It loses from 1.5 to 2 times the Alder Lake controller even at a higher memory frequency (6400 vs 5200 in the case of LPDDR5).

But if you look at this problem broadly - both companies, as I already detailed everything in the news about Dell CAMM modules, have obvious problems with RAM bandwidth. They are much worse than what is required now. And in the case of 7945HX, the lack of speed already reaches 8-10 times, which means that both companies urgently need to look for ways to switch to HBM memory in mass systems, otherwise they will suffocate in their races to increase L1..L3 cache. It cannot fix the general problem of lack of bandwidth at times, just like the SLC cache in an SSD cannot solve the problem of read / write speed beyond it if the NAND chips themselves are too slow.

Most well-known analysts write practically nothing about this.

[h4]

2 atom cores = 1 Skylake w/ HT

This is not strictly true; Gracemont 2C2T delivers the same performance as Skylake 1C2T at much less power. approximately, the following three have comparable MT performance.
- Golden Cove 1C2T @ 15W
- Gracemont 2C2T @ 15W
- Skylake 1C2T @ 35W

Test results run at maximum clock are shown below.
Golden Cove 4C8T @ 4.1 GHz, 55W - ST 1687, MT 8517 [12100]
Gracemont 8C8T @ 3.7 GHz, 50W - ST 1064, MT 7692 [part of 12900K]
Skylake 4C8T @ 4.0 GHz, 72W - ST 1119, MT 5449 [6700K]

[h4]

The 6800U is certainly excellent, but the problem is that we rarely see it on the market. In fact, in Q3 2022, AMD's client division experienced a devastating sales decline (which, given the timing, means that Rembrandt lost out to Alder lake). I speculate on this problem as follows:

Rembrandt dies in fab are binned as
 - majority -> 6800H, which is overwhelmed by ADL-H
 - low-voltage tolerance -> 6800U is superior, but the amount that can be shipped is limited to the production of 6800H

Instead, they are releasing large quantities of 5625U to the market, which we can easily obtain. We find that there is no interference by so-called 'mindshare'.

On the other hand, ADL-H/U allocates dies produced from the same mask to H for high-clock(good)/high-power(bad) and P for low-clock(bad)/low-voltage(good), so there is no bottleneck caused by the sorting process. ADL-U has a small die and no productivity issues, so much so that it is rare to see a selective dropout, i3 or pentium, on the market.

My own view is that AMD's strategy has failed commercially, although it has been able to satisfy those who discuss it on the forum in this way.

[h4]

Note that in the following announced sequence, there are major node or architecture changes at each step.
Raptor (1H 2023) -> Meteor (2H 2023?) -> Arrow (2024) -> Lunar (2024-2025)

If the conference presentation is correct, each major node change should decrease power consumption by a factor of 0.6 (when clock won't be changed) so the battery life should increase 1.3 times for each step in the calculation. In terms of efficiency, it is estimated that Meteor lake is comparable to M2 and Lunar lake is more efficient than M3 at the same time.

[LL]

One question to Notebookcheck.Net, how was a accomplished the total wattage consumption for each CPU?

[Starjack]

It would have been wonderful if there had also been tests of 1215U with deactivated E cores to actually attempt approaching the fine line between efficiency and insufficient performance. However, given that 1255U at 15W nominal TDP and 2.5W maintained surfing TDP is 2.5x faster in single thread than I need for office work, that fine line could be approached only by also underclocking and maybe undervolting.

I don't think is possible to disable E-cores on a laptop processor, not even in the BIOS. I have a Lenovo laptop with the Core i3-1215U in it. The only thing you could disable is the hyper-threading of the P-cores but that would cost the performance of the CPU. The CPU could undervolt on its own depending on how long it was use. But when you use it for games, it will give you lower framerates, unless you reboot your device. Rebooting it will allow the CPU to reset itself to operate at higher performance, about up to 28W.

[Anonymousgg]

I don't think is possible to disable E-cores on a laptop processor, not even in the BIOS. I have a Lenovo laptop with the Core i3-1215U in it. The only thing you could disable is the hyper-threading of the P-cores but that would cost the performance of the CPU. The CPU could undervolt on its own depending on how long it was use. But when you use it for games, it will give you lower framerates, unless you reboot your device. Rebooting it will allow the CPU to reset itself to operate at higher performance, about up to 28W.

You should be able to disable cores in Windows and Linux at the OS level.