Got this slightly wrong. The SoC cores are 6 performance and 2 efficiency or 8 performance and 2 efficiency. So my guess was entirely accurate after all.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
User predictions for Apple Prosumer Silicon (M1X, M2 etc.)
- Thread starter leman
- Start date
- Sort by reaction score
Looking at the cache size on the slides, it’s probably Firestorm. LPDDR5 confirmed.
And the preliminary winner seems to be @deconstruct60 with prediction nr 8! Let’s wait a bit for architecture confirmation before congratulating him though 
I don’t think anyone has predicted the 512bit RAM bus in a laptop though…
I don’t think anyone has predicted the 512bit RAM bus in a laptop though…
Well got the the GPU binning right on the 14 and 16", didn't expect the CPU binning on the 14" though. I wonder how a 6+2 M1 Pro compares to a 4+4 M1, while I expect it to be faster I'm not sure how much faster it'll be.
The un-core aspects will put some differences between the two.
Single Threaded:
Faster Memory that fills a bigger system cache. ( even if kept the core clocks the same that would help incrementally).
A single P core probably won't get more than its "fair share" of bandwidth increase, but there is more to go around. This will allow some pipeline stalls to clear faster, so not a huge jump but probably measurable.
Multiple Threaded:
swap two E cores for two P cores. That will be a net increase. Then add two more P cores. ( Faster memory and bigger cache helps here also. ), The multiple threaded difference for "heavy" computational work will be more substantively different. "5" versus "6.5" . CPU wise not huge but graphics ( and GPGPU compute ) gap is much bigger.
Didn't expect the Max on the 14". They made improvements to the thermal capacity on a slightly bigger (and heavier) 14".
Given the base "upper" 13/14 went from $1,799 to $1,999 the "binning" on the CPU cores isn't surprising. Apple already 'slipped' on increase costs. This bumps the "full" M1 Pro system price higher and therefore the average system price higher also. Apple is counting on folks saying " '6 P' cores is just too close to the M1 , I better bump up to get my money's worth. "
I doubt there are enough faulty P cores to support that SKU. They are flipping some cores off so can charge higher prices for the fully enabled one. It is $200 more just to "turn on" two P cores. That is $100 per core. Another $100 to get two G cores. That's is just printing extra money.
I think I did pretty good with my predictions. My pricing was actually pretty close. For instance, I was bang on for the $3200 M1 Max except they made 32 gigs standard for it which bumped it to $3499. I was wrong about the 14 inch carrying the higher end GPUs. Also, I didn't expect the binning to include the processor core, thought it'd be limited to the GPU.
My most interesting prediction was about the 32 core GPU which I indicated would outperform the 3070 by 15 to 25 percent. Of course, in the charts Apple provided, they compared the performance to a 3080. It sounds like I was SLIGHTLY optimistic, but not by much. However, we'll need real world tests to see if that's true.
The 50% increase in memory bandwidth was because Apple went from LPDDR4x in the M1 to LPPDR5 in the M2, and the latter gives 50% more bandwidth than the former. But the M1 Pro/Max/Ultra already have LPDDR5. The Pro/Max/Ultra could increase in bandwidth by going to LPDDR5x, which would give another 33%. This will happen eventually, but not sure if it will be for the M2 models. Depends on the availability of LPDDR5x. Samsung has started mass production of LPDDR5x, and we should be seeing it in 2023 devices; not sure about late 2022.
M2 got two extra GPU cores and frequency seems to increase by roughly 10%. If the same ratio can be assumed for other variants (20% more cores, 10% higher frequency), that’s still 30% in total.
so you think there is just 30% and no extra 20-30% from the upcoming LpDDR5x?
Kind of sad, but at the end 30% from one generation to the other is still significant
i also wonder if the MBA with M2 will sustain that gpu at least for a mediocre period of time since Apple showed that the gpu is drawing 16W instead of 12W...maybe thats why they still keep the active cooling 13" MBP because this M2 will make a bigger difference between a fan-less design and the active one at the same price. Maybe im wrong
So the M2 family will be multiple of 24gb of Ram (instead of 16).
So the M2 family will be multiple of 24gb of Ram (instead of 16).
Apple said 35%, so I'd use that figure, which indicates an 8% increase in per-core performance, since the number of cores increased by 25%: 1.08 x 1.25 = 1.35. But yes, if you extrapolate that to the rest of the lineup, you'd expect the same percentage increase with the M2 Pro/Max.
What's the ballpark expected %increase in performance with NVIDIA's 40-series cards over the 30-series? [Yes, I know they're supposed to be extraordinarily power-hungry, with rumors of a 600W TDP for the 4090 desktop (because Alder Lake by itself doesn't generate quite enough heat), but we can discuss efficiency separately.]
Last edited:
Current UMA bandwidth for Mn-series SoCs...
LPDDR5X is 33% faster, uses 20% less power, and should have a maximum of 64GB per RAM chip (so 1TB RAM in a quad Mn Max SoC configuration); I would expect Apple to retain the LPDDR5X RAM for ASi Mac Pro usage only...?
SoC | RAM Type | Bandwidth | Minimum RAM | Maximum RAM | ||
|---|---|---|---|---|---|---|
M1 | LPDDR4X | 68GB/s | 8GB | 16GB | ||
M2 | LPDDR5 | 100GB/s | 8GB | 16GB | 24GB | |
M1 Pro | LPDDR5 | 200GB/s | 16GB | 32GB | ||
M1 Max | LPDDR5 | 400GB/s | 32GB | 64GB | ||
M1 Ultra | LPDDR5 | 800GB/s | 64GB | 128GB |
LPDDR5X is 33% faster, uses 20% less power, and should have a maximum of 64GB per RAM chip (so 1TB RAM in a quad Mn Max SoC configuration); I would expect Apple to retain the LPDDR5X RAM for ASi Mac Pro usage only...?
Last edited:
Is this 2-fold increase also expected to be seen with the laptop 4000 GPU's, where they can't increase the power as much as with the desktop cards?
50gb/s bandwidh for the M1? Apple said that 100gb/s is 50% and not double of the M1
M1 is around 70GBps i think . Edit around 68 it seems, but clearly not half of the M2
Attachments
That's right. Here's a nice table from Anandtech; though where they write LPDD4 for the M1, that should be LPDDR4x.
Apple Announces M2 SoC: Apple Silicon for Macs Updated For 2022
Last edited:
very nice scaling will beThat's right. Here's a nice table from Anadtech; though where they write LPDD4 for the M1, that should be LPDDR4x.
Apple Announces M2 SoC: Apple Silicon for Macs Updated For 2022
www.anandtech.com
View attachment 2015072
The M2 Max will have up to 96gb unified Ram and the Ultra 192 Ram
M1 Ultra had 20.8 TFLOPS , the M2 Ultra 28.8 TFLOPS
Last edited:
No. The multi-core performance of the M2 CPU is, according to Apple, 18% higher than the M1's. By contrast, the M1 Max's multi-core CPU performance is twice that of the M1. The M2 will exceed the M1 Max only in single-core performance. And the difference will likely be less than 18%, since a lot of the M2's increased multi-core performance is from faster efficiency cores.
Presuming tha the Pro/Max got the GPU core count increase. Maybe not if those dies are stuck on N5P. So wouldn't get the core count bump if Apple is trying to keep the die size about the same on the larger dies. Lots more cores so already going to grow just due to the cores/caches being incrementally larger. That could just blow out the transistor budget increase for the prescribed die size cost constraint.
M2 Pro and Max may not be "long lived" dies. Could be just 10-12 month stop-gaps. Incrementally better "gen 2" laptops that Apple can sell. Not desk CPU/GPU world beaters. If Apple doesn't put a lot of R&D effort into them then they don't have to last long time on the market to recoup costs.
Nice to see the M2 has arrived! Some interesting things I found:
First, Apple's disclaimer for the "up to 35%" faster GPU performance: "Final Cut Pro 10.6.2 tested using a complex 2-minute project with 4K ProRes 422 media." (Apple Store M2 MacBook Air page, footnote 3). The performance increase is likely due to the M2's hardware ProRes encoder/decoder. This is a very specific task so we can't generalize it.
Second, the "up to 18%" faster CPU claim is hidden behind a "Go inside M2" button and apparently only applies to the M2 MacBook Pro: "Testing conducted by Apple in May 2022 using preproduction 13-inch MacBook Pro systems with Apple M2" (Apple Store M2 MacBook Air page, footnote 4). The Air is likely too thermally constrained due to its thin and fan-less design.
I can't find the claims related to the disclaimers in footnotes 5-9.
We will have to wait for reports and benchmarks by users to get a more accurate picture of the performance of the M2.
First, Apple's disclaimer for the "up to 35%" faster GPU performance: "Final Cut Pro 10.6.2 tested using a complex 2-minute project with 4K ProRes 422 media." (Apple Store M2 MacBook Air page, footnote 3). The performance increase is likely due to the M2's hardware ProRes encoder/decoder. This is a very specific task so we can't generalize it.
Second, the "up to 18%" faster CPU claim is hidden behind a "Go inside M2" button and apparently only applies to the M2 MacBook Pro: "Testing conducted by Apple in May 2022 using preproduction 13-inch MacBook Pro systems with Apple M2" (Apple Store M2 MacBook Air page, footnote 4). The Air is likely too thermally constrained due to its thin and fan-less design.
I can't find the claims related to the disclaimers in footnotes 5-9.
We will have to wait for reports and benchmarks by users to get a more accurate picture of the performance of the M2.