Kangal, 30 Jun 2022What is the best chipset, depends on the best compromises, and also the way your software beha… moreThe designs I'm mostly excited about are 0+5+0, 0+3+5, and 1+2+5. Here's why:
0+5+0: Has lowest latency!
5x Cortex-A73 cores which can be miniaturised on the modern TSMC-6nm will run very well. With the cache at a somewhat low (128KB each) amount, they will function similar to Cortex-A55 at low-end but can ramp up performance instantly. We're talking clockspeeds between 2.1GHz all-core, or potentially a 2.6GHz boost. Symmetry is the motto here. Multithreaded software will run fastest in this design, which dictates if this gets the best performance here or the lowest, but it's efficiency should be the higher than other options regardless.
0+3+5: Optimal balance!
Here we will use the DynamIQ / BIG.little setup, which is a net gain. However, since the software platform on AndroidOS is not fully optimised we have landed on this strange setup. We avoid the inefficient Cortex-X1 processor, and invest all that thermal potential into 3x Cortex-A78 cores (512KiB each) which can work together and CONSISTENTLY hit (2.7GHz) high frequencies. The small cores (64KiB) will also be given breathing room, at 2.0GHz all-core, so they can help if possible. There will be slight latency, but better than the industry standard. All-hands on deck is the motto. Efficiency should be great overall, no surprises.
1+2+5: Best Compromise!
While the Cortex X1 in general is inefficient and not very useful, there are certain uses for it. Very intensive calculations, particularly with Emulation (Xbox +), or Intensive Games, or Professional Applications may benefit greatly with the pure grunt (1MiB) it delivers. AndroidOS and the platform is somewhat optimised, however, peculiarities will cause delays in the above two options, and not-here. This design allows the X1 to closer to 3.1GHz potential, instead of throttling like other designs. To make the best use of this core, matching it with 2x Cortex-A73 that are small (128KiB) but work efficiently 2.2GHz will prevent from suffering too many stutters in the middle. The five Cortex-A55 small cores (32KiB), can handle idle tasks. They will barely hit the standard 1.8GHz frequency, and are needed to help boost performance in the midrange tasks. Latency begins to become an issue. Race-to-Idle is the motto here. Good efficiency overall, with sacrifices made to provide the option of bonus performance if/when necessary.
seems that next gen won't change anything , let's hope tsmc new FinFlex proves me wrong.
OhNom, 30 Jun 2022Of course, the TDP of the 8xA77 would be 15W and that of the 8 gen 1 is 11W, "super low c… moreBig thanks, this infographic again proved my theory.
dbjungle, 30 Jun 2022Battery life??? 🤔rework A715 to be more efficient, by controlling it's voltage dynamically and no need to reach theoretical maximum performance at the cost of efficiency. maybe put an actual "customization" in it's "semi-custom" name.
Anonymous, 30 Jun 2022"So useless" . What a coincidence that so many phones with octa A53 have giant batt… morethat's not a flagship, kiddo. read again my comment, slowly, words by words
Face, 30 Jun 2022Since many are ignorant of arm core's failure in recent years, you all should go check th… moreOf course, the TDP of the 8xA77 would be 15W and that of the 8 gen 1 is 11W, "super low consumption" and could not even match SD 888 consuming 16% more
Take the image as a reference and add all the cores it said and see the performance and consumption
Anonymous , 30 Jun 2022Geekbench is one of the least reliable benchmark toward actual tech reviewers for years now ex… morePerhaps, Benchmarks are always a questionable method of discerning actual performance metrics. Take Android fans are their love of AnTuTu for instance, specifically how they can't help themselves in lording the 'higher' scores over their Apple counterparts. This despite AnTuTu themselves stating the numbers are not comparable between the two platforms.
And as to your reply to about Apple chips falling behind, yeah your dreaming if you think Apple haven't retained all relevant license's relating to ARM IP, and thus would be fully capable of implementing any of ARM's advances as and when they felt necessary.
Anonymous, 29 Jun 2022For those who didn't get what this guy suggesting is basically same approach with Apple. … moreWhat is the best chipset, depends on the best compromises, and also the way your software behaves. Believe it or not, AndroidOS but more importantly it's Apps have evolved a fair amount in its day. Some new processor designs were thirsty, so using them sparingly was the better option, whilst others were more balanced in efficiency. The Cortex-A53 had "decent" performance back in 2013, and it is still comparable to the likes of the Cortex-A55, Cortex-A510, and Cortex-A515. However these are all deemed as "atrocious" performance now for 2023.
2009, 48nm, 1+0+0 design, Cortex A8
2011, 48nm, 2+0+0 design, Cortex A9
2012, 32nm, 4+0+0 design, Cortex A9+
2013, 32nm, 4+0+0 design, Cortex A17
2014, 24nm, 0+2+6 design, Cortex A15-A7
2015, 24nm, 1+0+7 design, Cortex A57-A53 (not much improvement since 2013)
2016, 16nm, 2+0+2 design, Cortex A72-A53
2017, 16nm, 0+4+0 design, Cortex A73 (smallest "big" core!)
2018, 12nm, 2+0+6 design, Cortex A75-A55
2019, 12nm, 0+3+5 design, Cortex A76-A55 (DynamIQ at work)
2020, 8nm, 0+3+5 design, Cortex A77-A55
2021, 8nm, 0+3+5 design, Cortex A78-A55 (the last best ARMv8 option)
*2022, 6nm, 2+4+0 design, Apple a15P-a15E (example of iOS)*
2022, 6nm, 1+5+2 design, Cortex X2-A710-A510 (new Lithography helping)
2023, 6nm, 1+5+2 design, Cortex X3-A715-A515 (the X3 should only be used when single-core performance is prioritised, otherwise an Overclocked A715 should compensate satisfactorily whilst using less power, or using several A715 cores instead of a single X3 core).
2024, 4nm, ??? design, Cortex ???
So we should have a little volcano in our hands next year
22niro, 30 Jun 2022The Chinese have already swallowed the company from the inside.. guess where the innovation ca… moreAre you talking about Arm China? That's just a subsidiary.
I wish that Huawei Hisilicon was still in the game. That would have helped to move things forward.
At least Qualcomm is back with TSMC. Unfortunately prices are going to go up.
SShock, 30 Jun 2022That's a big presumption that Apple will just stagnate and be doing nothing in next 2 yea… moreNot my word, it's Nuvia promise, and 2000 is just estimation this is their graph.
Fulljack, 29 Jun 2022Mediatek or Qualcomm shoupd step up the game and build a flagship SoC with 2xCortex-X3 and 4xC… moreBattery life??? 🤔
Anonymous, 29 Jun 202225% more Performance for X3 is nice, but I doubt that it can beat Apple Avalanche. I hope man… moreShould be ahead of Firestorm though. I like the 12-core 8+4 designs. I think this would be cool to see in tablets or laptops.
alcatraz, 30 Jun 2022I get the feeling that ARM engineers have run out of ideas. The improvements are coming mainly… moreThe Chinese have already swallowed the company from the inside.. guess where the innovation came from as a rule?
SShock, 30 Jun 2022"compute number between different chip design even on same architecture type" Thi… moreCompilers and optimisations in such pretty much do and Geakbench is pretty much prime example of this (the X86 part is horrible outdated regarding this).
You use industrial benchmarks for a sake of this and you compile them yourself if you want a real performance assessment.
Anonymous , 30 Jun 2022Geekbench is one of the least reliable benchmark toward actual tech reviewers for years now ex… more"compute number between different chip design even on same architecture type"
This sentence literally makes absolutely NO sense what so ever and it makes my brain hurt as I'm trying to decypher it. Benchmarks do not care about design or architecture and are not meant to differentiate. It just calculates how much time it takes to perform it (or it's based on number of operations it can perform within a fixed timeframe).
I get the feeling that ARM engineers have run out of ideas. The improvements are coming mainly from the switch back to TSMC.
865 + 870 were the last flagship chipsets made by TSMC (apart from SD8G1plus).
There we go again….x times more powerful, x times more efficient. You people don't get tired of this crap you're serving us all the time? And it's about everybody in this business, chip makers, phone makers, even those who make Gorilla glass and give IP rating. That type of advertising is running thin already, most of people can see through the smoke screen, change a little bit the narrative. You can full the buyers for a limited period after which they'll never trust you again, no matter haw honest you'll be.
SShock, 30 Jun 2022What "Apple fans"? It's a benchmark that gives compute tasks to a chip and it s… moreGeekbench is one of the least reliable benchmark toward actual tech reviewers for years now except apple users. And one of the worst thing about geekbench is it fails to measure the compute number between different chip design even on same architecture type.
And about Apple chips, I already replied a guy about it.
Fulljack, 29 Jun 2022having 4 of it are literally useless, as Apple manage to create Icestorm cores that both twice… moreMentioning apple chips here nothing but sheer ignorance from Apple fans. Apple design their SoC only for their specific devices, thus they have full control over their circuit design and performance with total optimization. Yet their SoCs are lack behind in many things. While ARM have to design chips considering all OEMs yet better in some ways than Apple chips.
All brands Rumor mill
Home News Reviews Compare Coverage Glossary FAQ RSS feed