No there is not hyperthreading on phones. With how efficient ARM processors run it is cheaper to just to double the core count.
And from my experince HT does not work as well actual cores. if you look at something like cinebench which is a very multithreaded application still has a very low MP ratio. My 4C/8T laptop at home gets a MP ratio of 4.68. so HT only made 0.68 difference in speed vs just cores alone.
Nope there isnt as far as i know.
Hello. I’m from the future. 7 years in the future since the last reply, Huawei has made their own processor and I read somewhere that it has a hyperthread core or something. Not really sure what that is about but if this site is still working, and if you still check this page, do read up on it if you want
lol, I just saw your post. This graph really shows what is going on:
scroll to the picture half way down the page titled:
spec int 2017 estimated scores and efficiency.
Instead of hyper threading, tasks not on a user interactive process can run on one of the little processors. However as you can see from the graph, only apple’s little processors have better performance per watt than their big processors. So for the non-apple CPUs the little processors actually only end up being used if the entire load can be run on the little processors or the big processors are busy.
The way i understand it is that it has a lot to do with RISC vs x86. The instruction set is much smaller and this allows smaller cores and less “overhead”.
Single threaded x86 can be compared to going through a drive through at a fast-food place. You have to order, pay and then get the food, this can be complicated orders.
SMT/HT would be that instead of 1 ordering booth, there 2 order booths. you can handle taking 2 orders at a time. but there is still only 1 kitchen.
RISC would be fast food where you can only get a couple of dishes, but they can be made very quickly and you get the order in the same booth as where you order and pay.
Maybe not the best analogy but using hyperthreading on a CPU that is already very efficient on a small set of instructions would barely or not help.
Picture of Zen 4 execution pipeline
Picture of estimated pipeline of Apple A14 chip
Yes, this is because Huawei leveraged core design from it’s server cores for it’s phone SoC Taishan cores (think Kunpeng). I don’t think this is going to be a trend in the mobile market, it pushes performance up but at a disproportionate amount of energy usage.
I like that analogy.
Although RISC and x86 have been converging on core complexity for many years IMO, each architecture is getting more and more advanced instructions like SVE2 and AVX512 that create a lot of core logic overhead.
speculation cap engaged:
We might see a break in this paradigm of “extra” dark silicon with x86’s new paradigm of rentable units, Intel is rumored to be axing hyperthreading on it’s CPUs in the future to implement this. This idea has to have come from Intel’s SoftMachines acquisition and their focus on VISC architecture.