That makes sense. To be fair, those game demos looked pretty damn good. Very much looking forward to seeing the design decisions made for the M3 and its brethren.
Thread: iPhone 15 / Apple Watch 9 Event
- Thread starter exoticspice1
- Start date
Jimmyjames
Elite Member
- Joined
- Jul 13, 2022
- Posts
- 1,085
10% improvement for Perf cores seems disappointing? That’s less than the improvement of going to N3. Hmmm
Indeed. The quoted 4x RT performance improvement also sounds very good.
I am wondering about the CPU improvements. They mentioned a wider architecture, curious to see if SVE is included.
Also as per discussion with @Jimmyjames in the other thread, doubling the Neural Engine and specifically talking up how the Neural Engine can collaborate with the GPU. Maybe a hint of things to come on that front as well?
If I had to guess they poured it all into efficiency on the iPhone, but we’ll see. Did they give code names for the new cores?
I'm surprised that they're not talking about crazy battery life, yet they keep mentioning that it's the lightest ever. Hmm
Jimmyjames
Elite Member
- Joined
- Jul 13, 2022
- Posts
- 1,085
Could be. I hope that doesn’t make it’s way to the M3 and it scales higher.
From what I understand, MetalFX used the NPU for upscaling from the start. That’s why the latency is ridiculously high.
- Joined
- Aug 10, 2020
- Posts
- 12,842
- Solutions
- 18
- Main Camera
- Sony
You called it.
- Joined
- Sep 26, 2021
- Posts
- 6,151
- Main Camera
- Sony
Well, it’s 120mm effective, with a much smaller sensor. Won’t be the same as 120mm on a 35mm sensor. That said, it’s a nice focal length and at this point they cover anything I’d likely carry with me for a real camera, other than the 300mm zoom I use for sports.
I may have missed it but do we have new code names for the cores?
That’s unfortunate, how does it compare AMD/Nvidia in that regard?
Jimmyjames
Elite Member
- Joined
- Jul 13, 2022
- Posts
- 1,085
Not that I’ve seen.
Yeah I am curious as they didn’t explicitly mention if it was ARMv9 - also wondering about the matrix accelerator if it’s now ARM-standard or not.
- Joined
- Sep 26, 2021
- Posts
- 6,151
- Main Camera
- Sony
i don't believe they revealed that.
Yeah I’m wondering if there is going to be permanent divergence between the lines and what that’ll mean.
- Joined
- Sep 26, 2021
- Posts
- 6,151
- Main Camera
- Sony
i would have been shocked if they mentioned ARM v anything. Users don't care.
I am led to believe that these support ARMv9 (or, more accurately, that SOMETHING does - I assume this is it).
Andropov
Site Champ
- Joined
- Nov 8, 2021
- Posts
- 712
My guess would be that hardware raytracing takes a lot of space that can't be used for other things in the GPU, but I know very little about GPU hardware architecture.
What are your thoughts on the 10% performance improvement on the P cores? It's not the usual 20% but I'm under the impression that the last couple years Apple is focusing more on efficiency than peak performance for the iPhone chips, as the peak was already stellar.
I found this very interesting as well. They mentioned "Wider decode & execution engines". My understanding is that Apple's cores were already extremely wide at both, I'm curious to see what they've achieved. I thought they were talking about increased instruction level parallelism rather than something like SVE though. I would find the latter particularly interesting because my guess is that increased instruction level parallelism is a more efficient way of increasing performance (if you can feed those wider execution units) than increasing core clock.
Yup. I had a crash using MetalFX on iOS 16.0 (fixed on 16.0.1 or 16.1, can't remember) that clearly stated that the processing was happening in the ANE.
Just some napkin math with blender open data (v 3.6.0): M2 Ultra scores 3447.15; naively extrapolating 20% GPU compute improvement with 4x raytracing speed results in a score of ~16546.32. RTX 4090 scores 13100.33. Even without any exciting M3 surprises, that's looking pretty good.
