arm’s biggest obstacle is still its inability to scale up. it can only scale out. intel and amd builds a cpu core that works at anywhere from 1ghz to 6ghz, in a variety of consumer devices or enterprise servers, and a variety of power and cooling profiles. arm builds cpu cores that target low-medium power applications and only that.
cpu core that works at anywhere from 1ghz to 6ghz,
3rd gen Apple Silicon hits 4.05 GHz.
1st gen Qualcomm Spandragon X Elite hits 4.2GHz.
Both likely does more per clock speed than x86.
enterprise servers,
Everything you said applies to ARM more than a decade ago. R&D money from over hundreds of billions of battery-limited devices has changed that. Of which Apple has sold over 2.32 billion $429-1599 iPhones
cpu core that works at anywhere from 1ghz to 6ghz,
AMD dispelled the megahertz myth, now the gigahertz myth, in the 90s. I should know, I read the online forum posts about it in real time.
-
3rd gen Apple Silicon hits 4.05 GHz at under 45W.
-
1st gen Qualcomm Spandragon X Elite hits 4.2GHz at under 80W.
-
1st 6GHz Intel chip used 250W.
When ~80% of PCs annually shipped globally are battery-powered laptops then a desktop achievement of that magnitude is very relevant to more people.
Both likely does more per clock cycle than x86. As both have larger/more complex cores.
enterprise servers,
Fugaku became the fastest supercomputer in the world in the June 2020 TOP500 list as well as becoming the first ARM architecture-based computer to achieve this.
What x86 has an advantage of is sheer number of software it has over ARM. But that is also changing as more people buy into it. It will accelerate as soon as Windows on Snapdragon 2016-2024 exclusivity expires
-
Everything you said applies to ARM more than a decade ago. R&D money from over hundreds of billions of battery-limited devices has changed that. Of which Apple has sold over 2.32 billion $429-1599 iPhones
cpu core that works at anywhere from 1ghz to 6ghz,
AMD dispelled the megahertz myth, now the gigahertz myth, in the 90s. I should know, I read the online forum posts about it in real time.
-
3rd gen Apple Silicon hits 4.05 GHz at under 45W.
-
1st gen Qualcomm Spandragon X Elite hits 4.2GHz at under 80W.
-
1st 6GHz Intel chip over 250W.
Both ARM chips likely does more per clock cycle than x86. As both have larger/more complex cores.
When ~80% of PCs annually shipped globally are battery-powered laptops then a desktop achievement of that magnitude is very relevant to more people.
enterprise servers,
Fugaku became the fastest supercomputer in the world in the June 2020 TOP500 list as well as becoming the first ARM architecture-based computer to achieve this.
What x86 has an advantage of is sheer number of software it has over ARM. But that is also changing as more people buy into it. It will accelerate as soon as Windows on Snapdragon 2016-2024 exclusivity expires
quantity of chips sold is irrelevant to my argument. it doesnt change the fact that arm cores cant scale up.
mhz isnt the absolute indicator of speed. it demonstrates how well your architecture can scale up with power consumption, and whether or not the design is capable of something like that. try giving an 8 core arm chip a power budget of 150W and see whether or not it can scale up to draw that much power. a 13900k or 7950x can be given a power budget of anywhere from 45W to 200W. the chip will adjust its clock speeds to fit within your given budget, and its reliable enough that every unit sold will be stable and function at that power budget.
that 2021 article about laptops was riding on a covid sales surge. if you look at more recent data the market has completely corrected itself, and even rebounded in the opposite direction.
supercomputers are a different breed. they tend to work best on highly paralellizable jobs where individual core performance is less important than the sheer number of cores in a cluster, and the quality of interconnects. once again, demonstrates arm’s ability to scale out but still fails to scale up.
-