5%? What happened to the bs +30% over Cortex-A78 claims?

By: --- (---.delete@this.redheron.com), July 1, 2022 10:27 am
Room: Moderated Discussions
dmcq (dmcq.delete@this.fano.co.uk) on July 1, 2022 6:27 am wrote:
> Anon4 (no.delete@this.example.com) on June 29, 2022 12:20 pm wrote:
> > Anon4 (no.delete@this.example.com) on June 29, 2022 12:10 pm wrote:
> > > It's unclear what the speed path limitations are and if you can trade any of that power
> > > advantage for higher clock at the same power increasing the performance uplift.
> >
> > Ah it is clear, the 5% is at isopower with the A710.
>
> Well 5% beter performance with everything else except the power staying the same. THhe
> power goes down 20% which is very impressive - but I'm fairly sure the 5% won't go up
> to 20% better performance at the same power. They seem to have the X3 as the performnce
> core and the A715 is pratically turning into a new level of efficiency core.

Is that in fact what's happening?
Look at a picture like
https://fuse.wikichip.org/wp-content/uploads/2022/06/arm-a510-graph.png

On the one hand you can say each of the three chips has its optimal performance range, and so three gives you a wider "efficient dynamic range".
On the other hand, it's non trivial to figure out exactly which code to run where, even with just two different chips (each with their own DVFS).

Apple's solution clearly consists of
- high performance for their little curve. (1/4 to 1/3 of big, which looks like around the "efficient" part of the A715 curve relative to X3)
- a methodology for what runs on which CPU.

To what extent does this match Android? In that now (as opposed to, say 8 years ago) is it generally felt that Android does a good job of moving between big and little cores and DVFS levels in a way that never feels laggy, or uses too much energy? And does that methodology extend to three cores?

Another way to look at it, perhaps, is to imagine two classes of devices, high end that run X3+715 and low end that run 715+510?
And while vendors may want to boost their core counts by adding extra 510s to a device that already sports an X3, that's basically just dumb noise, to be ignored?
Do we have a real clue how well Android today handles 3-tiers? Is it basically to ignore one tier?
< Previous Post in ThreadNext Post in Thread >
TopicPosted ByDate
New ARM Cortex cores launchRayla2022/06/28 08:49 AM
  Solid upgrade, I'm not disappointed (NT)Kara2022/06/28 08:56 AM
  5%? What happened to the bs +30% over Cortex-A78 claims? (NT)none2022/06/28 08:56 AM
    5%? What happened to the bs +30% over Cortex-A78 claims?none2022/06/28 10:13 AM
      5%? What happened to the bs +30% over Cortex-A78 claims?none2022/06/28 12:38 PM
        5%? What happened to the bs +30% over Cortex-A78 claims?none2022/06/28 02:43 PM
          5%? What happened to the bs +30% over Cortex-A78 claims?Anon42022/06/29 12:10 PM
            5%? What happened to the bs +30% over Cortex-A78 claims?Anon42022/06/29 12:20 PM
              5%? What happened to the bs +30% over Cortex-A78 claims?dmcq2022/07/01 06:27 AM
                5%? What happened to the bs +30% over Cortex-A78 claims?---2022/07/01 10:27 AM
                  5%? What happened to the bs +30% over Cortex-A78 claims?Anon2022/07/03 12:25 AM
  Better articleRayla2022/06/28 08:58 AM
    Better articleAdrian2022/06/28 10:31 AM
  New ARM Cortex cores launchGroo2022/06/28 01:26 PM
Reply to this Topic
Name:
Email:
Topic:
Body: No Text
How do you spell avocado?