Staggering M1X performance predictions place the upcoming 10-core Apple Silicon ahead of a 2019 16-core Mac Pro

[Redaktion]

New performance predictions for the Apple M1X SoC have been offered up, and they show the 10-core Apple Silicon surpassing the 16-core Intel Xeon W-3245 in the Mac Pro from Late 2019. It has also been reckoned that the M1X cores will be based on both M1 and A15 cores, presumably for increased efficiency.

https://www.notebookcheck.net/Staggering-M1X-performance-predictions-place-the-upcoming-10-core-Apple-Silicon-ahead-of-a-2019-16-core-Mac-Pro.565265.0.html

[roland_m]

I have to say I am a bit appalled that arbitrary claims by a popular YouTube who has proven himself again and again to have very little technical understanding about CPU technology and benchmarking methodology is being reposted as some sort of expert testimony. It doesn't matter whether Vadim is right or wrong — the numbers are probably realistic, as anyone can do a simple extrapolation from M1/A15 benchmarks — but there mere fact that these "findings" are being reposted as some sort of exiting new information is worrisome.

[Tridents]

I would say his CPU numbers are not at all realistic.
He is way underestimating the performance of the efficiency cores and its historic impact on the multicore score, which throws his calculations completely off-course. He also doesn't seem to realize that Geekbench doesn't scale very well with increasing number of cores. Geekbench M1X multicore score should hover around 12,500-13,000, maybe less, certainly not 15,000.

[8&8]

hypotethical predictions but nearer to reality than others, one bad point of view is the NPU (that is not considering), that twist completely the predictions. i really doubt that is a new arch, for only 6%-10% of gain perf.

I continue to think that the gain is thanks only to a new powerful NPU like SD 888 vs SD 888+ that in GB5 signs +20% of perf mantaining the same arch.

[Tridents]

hypotethical predictions but nearer to reality than others, one bad point of view is the NPU (that is not considering), that twist completely the predictions. i really doubt that is a new arch, for only 6%-10% of gain perf.

I continue to think that the gain is thanks only to a new powerful NPU like SD 888 vs SD 888+ that in GB5 signs +20% of perf mantaining the same arch.

The NPU has no part in these Geekbench tests. One is CPU bound and the other uses only the GPU.

[TKay]

So you're saying this new chip will be faster than a two year old chip AND it will be the fastest Apple laptop chip?
Um.. ya.
That is like saying the PS5 is going to faster than all models of PS4. 

[Ish]

Why waste time on hypotheticals from random YouTube nitwits when there are actual useful things to report on? Do these clowns pay NBC to post this kind of garbage so they get more viewers and ad revenue? I love NBC for reviews and benchmarks, but just about everything else is simply trash.

[Bojan]

These interpolations are making a big assumption by going from the A15.

It's also possible that Apple has started with Cortex-X instead of Cortex-A core for M1X.

[Ish]

So you're saying this new chip will be faster than a two year old chip AND it will be the fastest Apple laptop chip?
Um.. ya.
That is like saying the PS5 is going to faster than all models of PS4.

This "article" is horseshit, but that aside... you do realize a Mac Pro and a MacBook Pro are two entirely different things, right? That one is a laptop and one is a desktop? That desktop and laptop CPUs are entirely different things having different power requirements? It really doesn't sound like you get the distinction because the PS4 to a PS5 comparison doesn't even make sense... at all.

[8&8]

yeah i wrote a bullshit! However I wait for anandtech review,.

6-10% of more improv don't come from a new arch.
Higher clock? LPDDR5 instead LPDDR4?

the GPU arch maybe yes because is around 15%. 

[Whytho]

A14 and A15 use the same core design, dogshit article.

[Tridents]

A14 and A15 use the same core design, dogshit article.

Why do you say something like that?
Apple actually mentioned they were new cores. Doesn't mean they changed them too much, but there are differences.

[CraigC]

I would say his CPU numbers are not at all realistic.
He is way underestimating the performance of the efficiency cores and its historic impact on the multicore score, which throws his calculations completely off-course. He also doesn't seem to realize that Geekbench doesn't scale very well with increasing number of cores. Geekbench M1X multicore score should hover around 12,500-13,000, maybe less, certainly not 15,000.

I agree with you that many of the youtube celebrities and this article are extrapolating as if the only cores operating during the multicore score are the p-cores and no performance is coming out of the e-cores.   I did a rough calculation using e-cores providing about 70% the performance as the p-cores (which I have heard elsewhere of what they are capable of) and without taking any degradation of just adding more cores the maximum likely would have to be less than I think 14,750ish if I remember right...  I don't think there would be much of a degradation, so I would expect somewhere around 14,000 as my best guesstimate... which still is a really really good performance number for a laptop chip (especially in a 14" MacBook Pro).

[CraigC]

These interpolations are making a big assumption by going from the A15.

It's also possible that Apple has started with Cortex-X instead of Cortex-A core for M1X.

Apple does not use any cores from ARM...  If you are talking of architecture, the Cortex-X is still v8+ ISA (instruction set architecture)... the M2 series is the earliest you might see Apple use v9 ISA... but it would still be Apple cores.

[Tridents]

I would say his CPU numbers are not at all realistic.
He is way underestimating the performance of the efficiency cores and its historic impact on the multicore score, which throws his calculations completely off-course. He also doesn't seem to realize that Geekbench doesn't scale very well with increasing number of cores. Geekbench M1X multicore score should hover around 12,500-13,000, maybe less, certainly not 15,000.

I agree with you that many of the youtube celebrities and this article are extrapolating as if the only cores operating during the multicore score are the p-cores and no performance is coming out of the e-cores.   I did a rough calculation using e-cores providing about 70% the performance as the p-cores (which I have heard elsewhere of what they are capable of) and without taking any degradation of just adding more cores the maximum likely would have to be less than I think 14,750ish if I remember right...  I don't think there would be much of a degradation, so I would expect somewhere around 14,000 as my best guesstimate... which still is a really really good performance number for a laptop chip (especially in a 14" MacBook Pro).

It has already been observed in other benchmarks that the efficiency cores can represent around 20-25% of the CPU multicore performance on a M1.
You can also easily see, going through other CPU results, that Geekbench does not scale very well with the number of cores. With increasing number of cores the scaling "efficiency" decreases quite a lot!
Using this information, if you do the math for the expected 8+2 M1X cores, you would not even get to 13,000, even assuming the same scaling efficiency achieved by the M1.