Neither is cancelled, just don't expect nothing like press leaks predicted, or something as hardcore fanboys anticipated.
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.
Could we see M3 before A17?
- Thread starter senttoschool
- Start date
- Sort by reaction score
So, A16-based 4nm M2 Ultra, or A17-based 3nm M3 Ultra & M3 Extreme...?
I like news of a new chassis, and it would help to deliver the message "this is not your dad's Intel Mac Pro"...
My speculation about the Mac pro, if it's sells at right price tag, it may render pointless the Mac studio, also given the Mac mini Mx-pro success the Mac studio lies Among two sales monster eating it's customers.
ASi Mac Pro is based on m2 max multi chip modules, as aimed at an workstation it should also come with higher clock, I'd expect the Mac pro extreme without extra peripherals to eat about 600W.
Guess we have to wait another year or so to see a M3 Extreme SoC with hardware ray-tracing...?
Maybe WWDC 2025 brings the M3 Extreme on the N3X process...?
Further, I'd expect only two m3 soc, both 2+8, different about GPU cores and cache size, m3, pro, m3 max ultra Extreme etc just to mix said two base m3 by means of different kinds of interposer solutions according Mobile or desktop, this 'new' approach should improve ASi yields and lower the number of defective chips as to not considered recycling it, even bring new uses not just ipad maybe an m3 console/Apple tv.
Inside a Mac Pro obviously to take much longer, but expect plenty ray tracing at MB,iMac and Mini E24 MBP E24 (or mini L23, MBP L23), of course don't discard coming from another GPU vendor.
M3 (single UF) & M3 Max (dual UF) as SoC building blocks...?
Single M3 is "base M3"
Dual M3 is "M3 Pro"
Single M3 Max is "base M3 Max"
Dual M3 Max is "M3 Ultra"
Quad M3 Max is "M3 Extreme"
I won't name it this way, as basically all them should be based on the same two base m3 soc, only different interposer structure.
So both SoCs are 10-core CPUs (8P/2E), but differ in GPU cores & cache size...?
"Larger" M3 SoC to be used in chiplet configuration with AMD GPU chiplets as well as iGPU...?
Any more insight towards these different interposer structures...?
Don't know, but A⇒M⇒M Pro⇒M Max⇒M Ultra is exactly what the release sequence has been thus far for both the M1 and M2 (except we haven't yet gotten to the M2 Ultra).
While there may have been some codevelopment at the start, the following release dates suggest to me Apple, for instance, wanted to thoroughly understand the M1 chip before finalizing the design of the M1 Pro/Max, and wanted to thoroughly understand the M1 Max before finalizing the design of the M1 Ultra.
M1 Series (based on A14 Firestorm/Icestorm)
A14 (iPhone 12): September 2020
M1: November 2020 (2 mos later)
M1 Pro/Max October 2021 (11 mos later)
M1 Ultra: March 2022 (5 mos later)
M2 Series (base on A15 Avalanche/Blizzard)
A15 (iPhone 13): October 2021
M2: June 2022 (9 mos later)
M2 Pro/Max: January 2023 (7 mos later)
Now maybe this was something only needed for the first generation of M-series chips, and for the subsequent generations they can indeed develop and release them in parallel. But that's not what the release schedule thus far has indicated.
Last edited:
I have no insights on m3, just my speculation following apple acquisitions and concessions on ASi design.
Just extrapolation on AMD Epyc Genoa design adjusted for Apple purposes, a.e. MB, iMac, mba and base mini should get a single SOC on both base and power configurations, while mac mini pro, MBP should get an mini-ultra m3 based on base/power m3, and Mac Pro having single x4 strip to x8 or double x4 + x4 in H with an extra fabric chip among them.
I have no idea what lithography Apple will use. Best guess can probably be made by digging deep into applicable public info from TSMC.
Given that a) there really aren't many options on the GPU side for ARM-based processors, and b) integrating third-party graphics solutions onto the die would basically mean Apple is partly giving up on Apple Silicon as a wholly standalone platform, this analysis seems to be lacking in reasoning and justification.
Since Apple licenses their their technologies, I would think the "iGPU" in the ASi SoCs would be using a variant of this in the M3-family of SoCs...
As for third-party graphics solutions, the latest rumors have AMD MI200-series derived MPX-format accelerator options for the forthcoming ASi Mac Pro...
If the AMD rumor is true, we will have to see how it is implemented, but there will still be an ASi "iGPU" in the ASi SoC...
The M1 is released two months after the A14 in 2020. Two months is not enough time to do anything in the SoC design world. Both M1 and A14 has to have shared the same core u-arch design from the get go. Different blocks (CPU, GPU, NPU, etc.) designed by different teams and integrated as and when the block designs are completed.
So whether Ax gets released first or Mx really is up to Apple.
I would think the reason the Pro/Max/Ultra comes laters is largely due to supply of the SoC from TSMC.
You've focused on the outlier. And when you have to base an argument on an outlier, that's not an indicator of a strong argument. It could be that the M1 was in development for some time, allowing it to be released right after the A14. The other gaps were 5—11 months. Do you have any evidence that, say, the 9 month delay between the release of the M1 and the M1 Pro/Max was because of chip availability from TSMC?
I don't have any evidence that it took Apple longer to design a more complex chip of a given generation than a simpler one, but it is at least plausible.
Last edited:
All my points are my thoughts on the issue, so no, I do not have any evidence whatsoever.
I'm just saying that a SoC design would have been decided way before it is released. Waiting for an Ax design to be finalised before re-working it for Mx would not be fast.
The reason why I said the larger SoC are due to TSMC is because the A14, being the largest volume based on new 5nm would take precedence, following by lower volume M1, and even lower volume M1 Pro/Max/Ultra. From a risks angle, it makes sense to stagger the rollout. I would think the M1 Pro/Max/Ultra is ready to go when the M1 is released. With larger dies, defect rate would likely be higher.
As for M2, I'm not sure if TSMC increase fab capacity for 5nm, but I think it is unlikely for them to increase by much? So the M2 is likely facing the same constraint as the M1.
I think the M3 variants are already done and dusted. Apple is likely already working on finalising design for M4 now for next year's release.
I can certainly imagine that bigger chips take longer to design and test, but the design of Pro/Max does not depend on the design of the base SoC. Rather, they use the same building block. Ultra is two combined Max chips, so unless you want to argue that the Max was changed after the initial release to make the UktraFusion functional, this narrative doesn’t strike me as likely.
Explaining the timeline in terms of production constraints and yields might be simpler and more reasonable.
Apple has their own RT solution, described in numerous patents.
As to using AMD accelerators… what would be the point? If they indeed decide to mess up their software ecosystem it would make sense to partner with Nvidia to benefit from popularity of CUDA.
Designs for M1, M1 Pro, M1 Max, M1 Ultra were likely finalized together at around the same time. They aren't developed in sequence.
Apple likely has leapfrogging teams. One team develops M1. Another team is developing M2. Once the M1 team finishes, they will develop M3 or M4 depending on how many teams they have. It takes roughly 4 years to develop a chip from ideation to shipping to customers according to interviews by Apple Silicon team. They aren't going to develop M1, "understand it" so they can make M1 Pro better within 9 months. The design for M1 Pro was finalized years ago.
The delayed Pro/Max launch is likely due to economics and resources in validating chips.
As the global health crisis began, Apple was in the stages of validating the M1 chip, the company's first Apple silicon chip for the Mac. The validation process includes engineers carefully inspecting the chips, their transistors, and every component. Due to the circumstances deeming this impossible, Srouji's team had to set up cameras throughout labs that they would then use to remotely inspect each and every chip.
Apple's Johny Srouji Gives Rare Media Interview, Discusses Apple Silicon For the Mac
In a rare media interview, Apple's senior vice president of hardware technologies, Johny Srouji, discussed Apple's transition to Apple...www.macrumors.com
The same team that validated the M1 probably also validated M1 Pro/Max. They probably couldn't validate all the chips at the same time. It made sense to validate M1 first.
Last edited:
Yes, plenty of people have dug into public info from TSMC. That's how the consensus/rumors/reports all point to M3 using 3nm. And since it makes no sense to backport A16 to 3nm, which could take years, we can all assume that Apple will likely use A17 as the base for M3.
But they continue to license from IMG, one might think they also build on knowledge gained from said licensing...?
I dunno man, I was cool with my thoughts towards asymmetric multi-dies SoCs & ASi (GP)GPUs, then I jumped on this latest wagon...
I just want to see the danged ASi Mac Pro, then I can start speculative reasoning on what the M3 variant will be...! ;^p