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.
Apple's Next-Gen Chip Technology on Track for 2025 Debut
- Thread starter MacRumors
- Start date
- Sort by reaction score
I have come to a similar conclusion but some of your reasons are based on wildly false premises.
Your claim that they maximize cash per transaction is unsupported. There are items that could support this theory, but also others that support conflicting theories. On balance it seems quite clear that what they chase, extremely effectively, is maximum profit.
As for scheduling... others also bring out their most powerful chips last - Intel for example did that for many years, bringing out Xeons with CPU cores that had first been introduced the previous year in their consumer processors.
That said, I agree that this is clearly suboptimal, and if Apple can manufacture the highest-end chips first (which they definitely could not, for example, when N3B started shipping), then it seems very likely that it's in their best interest to ship top-end systems first, and that they will think so too.
Partially true, partially false. The size of the M3 Max is nearly 4x that of the M3. At a defect rate of 0 defects per cm^2, the M3 Max would therefore cost nearly 4x as much as the M3. But that's not possible. Depending on the actual defect rates, and on how much binning Apple can do, the actual M3 Max cost may be 5-6x... or it might be 10-15x. At some point it is simply too expensive even for top of the line systems, and that only changes as the process matures and the defect rates go down.
Two things govern this:
1) The larger the chip, the larger the loss of silicon per defect.
2) The more you can afford to lose parts of the chip (by binning), the lower the loss of silicon per defect.
That's why nVidia can eat more defects per cm^2 than Apple can, without driving costs insanely high. Most of the area of an nVidia chip is noncritical- that is, if it contains a defect, it can be fused off without making the chip unusable. On the other hand, that's less true for Apple chips: less of their chip area is noncritical because less of it is taken up by replicated units (GPU cores and/or CPU cores).
LOL! This is where you ran entirely off the rails. Even if it were "unfixable" (which, as it turns out, it isn't, even in the M1 and M2), there would be no legal action. You know how I know? It's because Intel and AMD processors have had a pile of worse problems, none of which they've ever had to pay out for. Tons of speculative-execution-related issues, but sometimes even worse. For example, the hilarious failures of Intel SGX.
Also LOL. The not-yet-released SXE is miserably underpowered even compared to the base M3 in single-core, which is still what matters most to most users. The M4 just crushes it more. In multicore, assuming you're talking about a desktop PC (because this is NOT true for a laptop), the SXE top-end chip running at its highest power settings will indeed beat the base M3. It doesn't beat the M3 Pro, and it gets utterly crushed by the Max.
That said, pricing does matter. Qualcomm won't be competitive with Apple any time soon on a technical level, but they are pricing the SXE aggressively enough that it does make sense, from a consumer point of view, to at least consider them compared to M3 Macs (Airs and low-end Pros using base M3 chips). I'd love to see that push Apple to improve their pricing (though I doubt it will, at least in the next year). But even if it does, that will be pricing for their base units. Maxes, Ultras, and maybe even Pros are untouchably fast compared to the SXE, which is unlikely to catch up much in the next few generations.
And in fact they don't. Those measurements are fantasies. Actual feature sizes are in the several tens of nm range. That's true for all leading-edge processes, not just TSMC's.
Last edited:
Oh, sorry, perhaps you are right. There I was thinking that they were desperately outsourcing all things AI at the last minute to anyone they could find, but perhaps I was wrong and they have themselves secretly developed the technology behind all the wonderful AI development (/s) we will see in a few months' time with Siri able to say good afternoon in the correct time zone and make a simple diary entry without screwing it up. Please feel free to message us back later in 2024 to point out to us just how much Apple had kept up with AI compared with what the media (this media) has been wildly reporting.
Apple hasn’t cared about their desktop line in about 15 years. I don’t see them changing. They will always prioritize their MacBook pros as those are more profitable. I fully expect we will be stuck with M2 for a while.
I came back to Apple with Apple Silicon. But if they don’t change their stance soon I’m out. I don’t want to use just a laptop. But right now the M3 Max does things a lot better than an M2 Ultra and that is just ridiculous.
I don't know what they've been developing in secret, but publicly they've been publishing ML & AI research and models for awhile. Like the stuff here: https://huggingface.co/apple There's actually quite a bit of Apple ML/AI development published on HuggingFace, Github and other places. Facts just don't play as well in the media as sensational claims.
HobeSoundDarryl
macrumors G5
MB are "more profitable" on VOLUME but not on per unit or per customer sold. For example, I can buy MANY configs of MB for much less than the basic config of Mac Studio ULTRA and- I think- just about ALL Configs vs. Mac Pro. Yes, Apple needs volume of profit for "another record quarter" but flipping the release schedule doesn't necessarily have to alter when PRO & MAX releases. For example, M4 Ultra about NOW, M4 PRO & MAX in October, M4 Base thereafter (and yes, I recognize base is already out in iPad, so go ahead and roll out other M4 base as they are ready). Volume sales of laptops feel no impact... unless some buyers can't resist maximum power and thus spend MORE on a Studio or PRO instead (for which Apple then makes more money in a higher profit-per-customer-sold substitution).
And then there's this "longing" effect of coveting both "latest & greatest" and "most powerful" Mac which is always shortchanged by rolling out ULTRA Macs last because they are King of those hills for only a relatively short amount of time before a next generation M Mac is released and then actually lose the "most powerful" race in many measures only a few months after release by next gen MAX chip release.
But if ULTRA LEADS the launch cycle of a new generation- for example if M4 Ultra is launched at- say- WWDC, the Macs that have it are king of the hill until M5 Ultra is launched. They will have "latest & greatest" Silicon AND retain "most powerful" crown all the way to M5 Ultra. Everyone who only wants a MB can still get a MB on the very same schedule.
As has been, even those who burn for "most powerful" are somewhat trained to know that if they can control themselves for only about 5-6 months they can spend about HALF the money to get "most powerful" and "latest gen." Flip it and they have to spend 2X or more to feed that need/want. Modern Apple Inc. LOVES more money.
Last edited:
You seem to want to fight, so I am walking away.
But let my parting message to you be that you seem unaware of the difference between "AI" and the specific technology of LLMs. This suggests that you're not especially well informed in this area, and you might want to fix that by interacting with some primary sources (like Apple papers, models on HuggingFace, and talks at WWDC over the years) rather than outsourcing your thinking to others.