M5 Pro and Max unveiled

[JRLMustang]

Just lurking here and not intending to get all “Battlefronty” but I think what some of the folks here are pointing out wrt: M series outclassing Intel is the fact that folks are even having this conversation comparing Apple silicon to any of the incumbents.

To be clear, Intel and the others have been iterating for decades. Apple is the relative newcomer here, and with each iteration Apple is closing the gap quickly. Apple already wins big on efficiency and to my mind, that’s the real story here —delivering comparable real-world speed at much lower power. That’s the message.

The M series is punching well above its weight class. Apple’s efficiency edge (fewer cores matching Intel’s total output). It’s like Intel is hauling it flat out down the Mulsanne Straight redlining it, sucking down the fuel and Apple is half-throttle sipping slowly and basically “right there”. If this is 24 hours at Le Mans, Apple takes the checkered here. (I know, I know… car analogies.)

 

[cbum]

speed at much lower power. That’s the message.

Absolutely. Just consider the brewing nightmare of IT centers/ AI eating up all available electricity.

 

[JRLMustang]

Yeah, it’s like the other companies got a wake up call. They’re probably pushing the envelope harder than they had planned and in doing so are bound to either wind-out or trip up. We haven’t even gotten an M5 Ultra yet, so who knows what that looks like? M6 could be really interesting. Apple likes to play a long game, and they continue iterating their M series processors rather quickly.

 

[Yoused]

To be clear, Intel and the others have been iterating for decades. Apple is the relative newcomer here, and with each iteration Apple is closing the gap quickly.

Apple is not really "closing the gap", they are opening it up. Intel is running a jalopy cinched up with bailing wire and cellophane tape (not even duct tape) trying to catch up. Nvidia has Grace-Hopper that is poised to murder the server space, and IBM could, if they were a little less greedy, squeeze x86 out from the other side.

Apple is not exactly a newcomer, really. They have been working with ARM for decades and helped them work on developing AArch64 (probably putting their Intrinsity people on the effort). They released the first 64-bit phone nearly 13 years ago and have been refining it year by year. For the past 6 years or so, Apple chips have been at least as fast as x86 alternatives and slowly creeping ahead, widenting the gap. ARMv8+ appears to be just that much easier to improve.

 

[JRLMustang]

Good points.

 

[Cmaier]

Apple is not exactly a newcomer, really. They have been working with ARM for decades and helped them work on developing AArch64 (probably putting their Intrinsity people on the effort).

Most of the intrinsity folks weren’t really architects, though one of the co-founders had some architecture experience (x86/powerpc). He sat in the cubicle next to me at Exponential and cursed a lot. (Intrinsity was essentially our Austin design team - as we started to go out of business, they changed their name to EVSX (“everything else sucks”) and later to intrinsity. Anyway, they started as a design services outfit, then my former cubicle-wall-sharing-neighbor got some patents on 1-of-n circuits and they were off to the races.

 

[throAU]

Apple is not exactly a newcomer, really. They have been working with ARM for decades and helped them work on developing AArch64 (probably putting their Intrinsity people on the effort). They released the first 64-bit phone nearly 13 years ago and have been refining it year by year. For the past 6 years or so, Apple chips have been at least as fast as x86 alternatives and slowly creeping ahead, widenting the gap. ARMv8+ appears to be just that much easier to improve.

Pretty sure Apple were an early investor in ARM way back as soon as it became a seperate company spun off from Acorn.

 

[Yoused]

Pretty sure Apple were an early investor in ARM way back as soon as it became a seperate company spun off from Acorn.

Yes, they used early ARM processors in the Newton, and later in iPod. Steve wanted to use an Intel processor in iPhone, but Intel were not interested in that space, so they ported Darwin to ARM and apparently started exploring how to transition to 64-bit.

 

[KingOfPain]

Apple is not exactly a newcomer, really. They have been working with ARM for decades and helped them work on developing AArch64 (probably putting their Intrinsity people on the effort).

A lot of people forget (or never knew) that Apple is the reason that ARM exists as it is now.
ARM started as the Acorn RISC Machine, developed by roughly 12 people working for Acorn Computers, because they wanted to replace the 6502 in their current computers with something better than a 68000 or 80286. They made the ARM2 and ARM3.
Then Apple was looking for a CPU for the Newton, but didn’t want to buy from a competitor in the computer market, thus Advanced RISC Machines was founded by Acorn, Apple, and VLSI. The first chip was the ARM6.
I guess without this move ARM would have been just as dead years ago as Acorn is.
I thought Apple were the first ones to implement AArch64, because of their architectural license (AArch64 had been announced years earlier), but later I read that Apple had been the driving force behind its development.

I always found it laughable when people said that ARM will never be ready for desktop use, because these people simply didn’t know that ARM was designed for desktop computers first, and later it turned out that they can be useful for the embedded market.
When Apple released the first passively cooled MacBook, I had two gripes about it: The single USB-C port that was also used for charging, and the fact that it had an Intel i3. Even back then I was of the impression that an iPhone CPU would not only have been fast enough, but the far better choice. Now the MacBook Neo does exactly that.

 

[Yoused]

I remember about 14 years ago, some of us were imagining Apple going to ARM from Intel, and there were x86 fans on MR who were saying no way ARM could compete with the power of Intel. The early versions in the late '80s were beating 68Ks and even outperforming 386s that were running twice the clock. Apple should have gone all out with ARM instead of PPC and saved themselves a couple transitions.

 

[Cmaier]

I remember about 14 years ago, some of us were imagining Apple going to ARM from Intel, and there were x86 fans on MR who were saying no way ARM could compete with the power of Intel. The early versions in the late '80s were beating 68Ks and even outperforming 386s that were running twice the clock. Apple should have gone all out with ARM instead of PPC and saved themselves a couple transitions.

ugh, don’t remind me. Do you know how many times I had to explain over there that the ISA makes almost zero difference in suitability for “desktop.” Then being told why I was wrong by people who never designed an x86 or a RISC chip.

BTW, for the same reason, PPC would have worked just as well as Arm (from a technical sense, anyway). RISC is RISC* (even though PPC drove me crazy with its ass-backwards bit numbering), and I am quite confident that if M5 was PPC it would have essentially identical power/performance characteristics. There are a lot of non-technical reasons that Arm was a better choice, of course.

*That said, I have some hesitation about Sparc, just because of the register window scheme that I think might complicate things in bad ways.

 

[Yoused]

RISC is RISC

What about this wonderful "RISC-V"? Presumably it is good enough, but how good is it? Seems like the CPU version of Linux Salad.

(even though PPC drove me crazy with its ass-backwards bit numbering)

Which may be one thing that attracted Apple to the PPC transition. Both 68K and PPC are BE, which made the move a little easier at the time. Both x86 and ARM are LE, which probably made less of a difference in that move.

 

[Cmaier]

What about this wonderful "RISC-V"? Presumably it is good enough, but how good is it? Seems like the CPU version of Linux Salad.



Which may be one thing that attracted Apple to the PPC transition. Both 68K and PPC are BE, which made the move a little easier at the time. Both x86 and ARM are LE, which probably made less of a difference in that move.

RISC-V would be fine. It has some weird quirks that I don’t like, but it would work.

As for PPC, the issue wasn’t even endianness. The msb is bit 0, and the lsb is 32 or 64. I am unaware of any other architecture that does this (other than IBM’s other architectures). Very different issue than endianness. If I want the least significant bit from a register, I need to mask off bits other than 32 (or 64 if 64-bit architecture). It’s very confusing.

This caused me all sorts of grief when I was designing floating point units, among other things.