Apple’s Sept. 7 Event

Status
The first post of this thread is a WikiPost and can be edited by anyone with the appropiate permissions. Your edits will be public.

exoticspice1

Site Champ
Posts
307
Reaction score
108
Could be. I guess the question is whether it makes any difference on a phone.
Guys, this means M3 if based on A16 won't see ANY meaningful GPU impovement.

The M1 GPU cores were based on A14 mostly and M2 based on A15. Both of these A series chips had GPU increases but the A16 nada.
 

exoticspice1

Site Champ
Posts
307
Reaction score
108
Who really cares about breaking 2000 when the difference it makes will be imperceptible to, honestly, anyone.
oh please try using a i7 3700K and a i7 12700K big difference. If we don't need CPU improvements we would be stuck on a 4th gen i5 as that was enough for web, email and video.
 

Cmaier

Site Master
Staff Member
Site Donor
Top Poster Of Month
Posts
5,450
Reaction score
8,814
Guys, this means M3 if based on A16 won't see ANY meaningful GPU impovement.

The M1 GPU cores were based on A14 mostly and M2 based on A15. Both of these A series chips had GPU increases but the A16 nada.
It doesn’t mean that at all. That’s the whole point of SOC design methods. The gpu is independent of the cpu.
 

Cmaier

Site Master
Staff Member
Site Donor
Top Poster Of Month
Posts
5,450
Reaction score
8,814
M3 if based on A16 will show <1% GPU improvement as previous M1/2 chips GPU was based on A14/A15 and these had good GPU improvements
M3 won’t have the same gpu as a16. I can guarantee you that.
 

exoticspice1

Site Champ
Posts
307
Reaction score
108
No thank you, I have had a deep and metastasizing dislike for Anthill CPUs for more than 40 years. Comparing old junk to less-old junk is pretty weak sauce. But, you are speaking of 10 years ago, well before we got close to this wall.
ok then try using a Bulldozer chip and then a Zen 4 chip(when it launches, can't say Zen4 is junk ;) ). Of course this is desktop but yeah I get what you mean mobile chips are enough cause apps and games are meh on mobile.

Desktop and laptops are the game now. I can't wait to see how Zen 4 and M2 Pro/Max perform on laptop and desktop as games and Pro tools definitely can show the improvement.
 

Cmaier

Site Master
Staff Member
Site Donor
Top Poster Of Month
Posts
5,450
Reaction score
8,814
hopefully it won't but past M chips have always been based on A chips.
No, they’ve never been based on A chips. The way it works is there are teams working on CPU cores and teams working on GPU cores (under the direction of system architects who figure out how the overall SoC architecture should work for each product). Once a CPU core is designed, it is plugged into various chips, sort of like a lego brick. Same with GPU. Same with NPU, etc. The same core may end up on an A-series chip and a M-series chip (and by “same” I mean same pre-physical netlist. The gdsii won’t necessarily be the same). But saying “based on A chips” implies that they design the A-series chip, then say “ok, let’s now move this stuff over to M.” It doesn’t work like that.

If you meant to say that the M-series has similar GPU cores to the A-series in the past, that’s a different thing.
 

exoticspice1

Site Champ
Posts
307
Reaction score
108
If you meant to say that the M-series has similar GPU cores to the A-series in the past, that’s a different thing.
Yeah I mean this. I know you can't just copy paste IP but the arch for the CPU and GPU was very mostly the same.
 

Cmaier

Site Master
Staff Member
Site Donor
Top Poster Of Month
Posts
5,450
Reaction score
8,814

It wouldn’t be surprising. After all, how high can IPC go? The maximum IPC is fixed with respect to any given ISA. As you get closer and closer to that maximum, it takes more and more transistors to eke out less and less improvement. You can add more ALU pipelines, but the instruction stream may not be able to fill them. Or you can add fancier branch prediction, but maybe that gets you from 95% to 97% correct predictions at the cost of twice the die area/power for that block. Etc.

The good news is that the maximum IPC for Arm is much higher than for x86-64, so IPC will continue to rise for awhile and at a steeper rate than what Intel and AMD can manager. Apple hasn’t even had to resort to things like multi threading, etc., yet.

And if we believe the benchmarks (I don’t yet), keep in mind that Apple claims a 20% reduction in power. So rather than optimizing for IPC they may have chosen this time to optimize for power. If that’s what happened, then in the M-series you might see a much higher max clock rate.

That said, I won’t believe the benchmarks until the devices are out in the public and we see more data. The MP scores make no sense, which makes me question the SP scores.
 

Andropov

Site Champ
Posts
644
Reaction score
834
Location
Spain
What do you guys think about this?
I still think that Geekbench result is weird (due to the multicore performance, that the tweet ignores). I believe that at least that part is not accurate (unless Apple is doing something new and weird now). If the single core result is accurate, the CPU single core speed bump is smaller than previous years. Maybe Apple's chip designers spent their time in improving power efficiency instead of performance. Maybe they spent their time in other parts of the SoC, there's a new Neural Engine, which is a huge part of the SoC, and the new Photonic Something. Last year, for example, the E cores got the biggest overhaul while the P cores got smaller changes. Or maybe they just hit a wall and couldn't improve it further.
 

Cmaier

Site Master
Staff Member
Site Donor
Top Poster Of Month
Posts
5,450
Reaction score
8,814
This year ordering was very smooth, and both my iphone 14 pro max and my wife’s iphone 14 pro are expected next friday.
 
Top Bottom
1 2