Apple M4 SoC analysis - AMD, Intel and Qualcomm currently don't stand a chance

[Redaktion]

Apple has presented its brand-new M4 chip alongside the new iPad Pro. It is a further development of the M3 SoC and is also manufactured using a 3 nm process. We took a closer look at both of the new M4 chips and compared them with the latest processors from AMD and Intel.

https://www.notebookcheck.net/Apple-M4-SoC-analysis-AMD-Intel-and-Qualcomm-currently-don-t-stand-a-chance.839332.0.html

[paviko]

As far as I know, M4 introduced new instruction set aka AVX and new version of Geekbench simply is using it. There is no true 20% better single thread performance.
Some years ago when Geekbench started using new instruction set from Intel to do encryption, Intel was ahead of other CPU in this benchmark, but the support was removed, because it was not fair. So now with M4 is it fair?
I'm not sure, but probably there are some result of Geekbench 6.x with AVX512 enabled on Intel and it scores more than 4600 in signle thread...

[AverageUserJoe]

This performance review is pretty vague and deviates heavily from the high quality journalism notebookcheck is known for. Geekbench with its AVX instructions kicks M3 out with Intel touching nearly 5K on single thread with same benchmark.

[LifePo7]

There isn't a single benchmark that scales with real-world performance. I'd be ashamed to post that.

[Neenyah]

There isn't a single benchmark that scales with real-world performance. I'd be ashamed to post that.

There is - Cinebench. But yes, you are correct.

[David.M]

GB is such a terrible benchmark that it makes the 96-core TR almost on par with the M3 max. The company behind this benchmark should be investigated, I'm sure there is something very rotten hidden there

[ksc]

Why do ppl compare Apple and PC CPUs? Those are two completely different ecosystems. Nobody is going to go to Mac to run PC software just bc it is 20% faster. In the current state, we have an oversupply of CPU performance. Nobody cares if a CPU runs a little faster anymore. Very fast CPUs are useful only for special applications like games, AI, etc., so it's pointless to evaluate fast CPU without running it in its intended applications.

[Mate]

Do you have guys any links to those sub 5k single-core scores of intel cpu?

[RobertJasiek]

Very fast CPUs are useful only for special applications like [...] AI [...]

Typical AI runs on dGPUs (or sometimes TPUs). The CPU is only a secondary factor. Higher single thread speed and more cores have slight impacts but the dGPU is the bottleneck, as I experience daily running AI. Only if one uses several dGPUs, many CPU cores in Threadripper and huge amounts of RAM can be useful. On servers with many dGPUs, as Huang tells us, also the network speeds for inter-dGPU communication matters. (Some kinds of AI, especially those with huge models, profit from fast inter-dGPU communication already for 2 dGPUs. For other kinds, it is immaterial for a small number of dGPUs. It can depend on how an AI is programmed.)

IMX, two dGPUs would double the speed. Higher CPU single thread speed might raise dGPU load by just a few percents and much more only ca. 5% of the time. More CPU cores would cope with even less frequent 100% dGPU load for several seconds, rarely for a longer time.

[George]

Forgetting for the moment that the majority of the Charts included in the article DO NOT support the article title...

Ok, silly Apply dropped their new M4 chip(s) into iPad's.

How many MORE HOURS of run time do these have over my current M1 11" iPad-Pro?

What EXACTLY can these new M4 powered devices DO that my M1 powered device can't?

As others mentioned, the applications and use patterns are different across the PC, Mac and YES iPAD computing environments.

However only time will tell if between Qualcomm & Apple we can FINALLY ditch the gawd awful x86 crap we've been stuck with for +40yrs!

[Neenyah]

The point is that the whole benchmark here is irrelevant when a 3 y.o. Ryzen 7 H can push 4000+ single core as soon as you benchmark it with Android booted from live USB stick. And on Windows it's not getting past 1700.

[Neenyah]

Very fast CPUs are useful only for special applications like [...] AI [...]

Typical AI runs on dGPUs (or sometimes TPUs). The CPU is only a secondary factor.

ksc was reacting to Apple's claims about SME2 and enhanced next-gen ML accelerators which are in the CPU, not part of the NPU.

[Trews]

dont stand a chance seems like an overstatement....

[D]

Could you double check whether Geekbench Single-core test score is version 6.2 or version 6.3? Since version 6.3 has SME support in Object detection and background blurring workload.   

[Paolo Gaolo]

Impressive numbers put out of context, mashing up apples & oranges along the way.

If there were any political undertone to this, it'd get dismissed as fake news.