Apple's Silicon Magic Is Over!

It's hard to believe that just over 5-years-ago,  I was ripping into the internals of brand-new Macs  

looking to hodge-podge a fix together, such that  they wouldn’t overheat and cause benchmark scores  

to plummet. High TDP Intel chips paired with  inadequate cooling made for a deadly combo—one  

that created the worst Macs in decades—but  then this happened: [insert M1 unveiling]

Apple silicon was a revelation to  what had been deemed an ill-suited  

form factor. Not only did they *keep* things  thin and light, but they launched with the  

literal identical MacBook Air chassis just  to prove—HEY—not only is this 3.5x faster,  

but we don’t even need a fan. Oh, and by  the way, we didn't touch the battery size  

at all but it lasts 50% longer than before.  Price increase? Nah. Send me a G and we coo.

The redesign of the MacBook Pro and M2 MacBook  Air righted the wrongs caused by the butterfly  

keyboard-sporting laptops of yore bringing  back MagSafe and other I/O, improving displays,  

speakers, keyboards, and more. Meanwhile, the M1  iMac and iPad Pro proved that no form factor was  

too small for this master-class in efficiency.  It’s not a stretch to say that Apple’s current  

computer lineup is not just Apple’s best ever,  but perhaps one of the best lineups ever from  

*any* company. But be warned, because changes  need to happen—and fast—if Apple wants that to  

continue being true into the future. Let's  tackle why the M3 series doesn't measure up  

to the almost physics-defying standards the M1  set at launch; and, how for the first time ever,  

real competition is just months away  from almost every PC maker imaginable.

I’ve talked a lot about why Apple silicon  has absolutely dominated since launch,  

but there were really three main drivers behind  M1’s incredible performance: (1) Arm64 itself uses  

significantly more modern (ISAs) instruction set  architectures that don’t carry the legacy baggage  

of x86—nerdy crap like weakly-ordered memory  models, a larger number of general-purpose  

registers for parallelizable code, etc., (2)  Apple’s dedicated on-chip hardware blocks like the  

video engines, Neural Engine, matrix coprocessor,  and unified memory pool, handle specific tasks  

vastly more capably than a general-compute  CPU or GPU, and, (3) deep vertical control  

from hardware to kernel to OS to application  helped eliminate cruft and streamline efficiency.

All of the work had to be paved for M1.  Sure, M2 and M3 benefit from that work,  

but they’ve been iterative—and Apple is now  somewhat limited in the same way everybody  

else is: transistor density. M1 launched  on TSMC’s 5nm process and unlike the 90s  

and 2000s when transistor density and node  naming actually correlated with one another,  

process names today like “5nm” don’t really mean  anything. There’s more to chips than logic gates  

and hardly any of those features precisely measure  at the marketed process size anyways. Regardless,  

the M1—not to even mention the M1 Max—was a an  enormous die that was not just the biggest TSMC  

5nm chip to date, but one of the largest Arm  chips ever produced period. So what do you do  

to get a faster chip like M2? There’s really  three options: (1) shrink the size and power  

consumption of the transistors so you can add  more of them in the same envelope, (2) keep  

the transistor size the same but increase the  number of them which makes for a larger die with  

greater heat and power drain, or, (3) keep the  transistor size and count the same but increase  

the voltage to push up the chip’s clock-speed  which creates even more heat and power drain.

M2 did a combination of options 1 and 2. They  were able to move to TSMC’s refined N5P process  

which netted both a 7% performance improvement  over N5 while drawing about 15% lower power;  

and then to speed things up even more, they  increased the total number of transistors  

from 16 billion to 20 billion. But this  increase didn’t come free. Do some shotty,  

“not really the full story” napkin math, and the  data would suggest the M2 is more power-hungry  

than M1—running hotter and drawing more energy  as a total package over its predecessor. And  

that’s not theoretical: we proved back when the  M2 MacBook Air launched that the chip was more  

difficult to keep cool and experienced more  rapid and more severe performance throttling  

due to those thermals. So why wasn’t this more  widely reported? Well, because M2’s performance  

per watt increased as well—not just total package  consumption. Imagine a high-performance sports  

car. The car runs hotter and consumes more fuel  when it reaches its top speeds, much like the M2;  

however, because it's so fast and efficient, it  can complete a 'race' much quicker than a regular  

car, reducing the total time it is running at  its hottest most fuel-consumptive state. So,  

while yes, the M2 consumed more total energy  at its peak, that extra compute was able to  

get tasks done more quickly—reducing time spent  at peak and therefore maintaining lower energy  

consumption per task relative to M1. Sounds  like a win-win—so what’s the problem? Sand, man.

When M3 launched on TSMC’s N3E 3nm process,  it was the first chip to do so—and performance  

expectations from pundits were high. I mean, doing  napkin math anew would suggest a 2.8x increase in  

transistor density—HUGE performance gains! But  then M3 came out and we got… a slightly better  

jump than we did from M1 to M2—and those on the  same process! Huh? Well, I guess we learned our  

lesson – using napkins for calculations can be as  messy as using a lipstick for math. First of all,  

TSMC’s 3nm node uses transistors that are  much physically larger than 3nm—they’re  

closer to 3.5. Okayyy, but even that would  suggest a 2x density increase. Ah, but only  

the logic density comes close at 1.7x.  SRAM and IO density barely increases at  

all. And chips—even magical ones—contain  all of these components. Realistically,  

there’s only about a 1.3x shrink. The era of  massive improvements from one process shrink  

to the next are over. The shrinks themselves are  now YEARS apart. And even worse, each new node  

is orders of magnitude more expensive than the  prior, per unit area. It’s estimated that Apple’s  

cost on these N3E chips is greater—not lesser—than  just using a bigger area on an older node. Alas,  

that would not yield the same  efficiency gains we’ve come to  

expect from Apple. More transistors on a  bigger chip means more heat. So what do?

We’ve spent several minutes getting really  nerdy and into the weeds on a lot of stuff  

that normal people don’t care about—and  at the end of the day, normal people buy  

the vast majority of Apple’s products. Sure,  the jump from M2 to M3 wasn’t as massive as  

expected and the jump from M3 to M4 will likely  be even smaller, but might I suggest something  

heretical for a minute? That’s OK! The silicon  isn’t the problem in Apple’s lineup any longer.

Look, Qualcomm invited me out to San Diego  a few weeks ago and I got to check out their  

reference design laptops (which basically  means they’re not real—they’re prototypes)  

for the Snapdragon X Elite SoC. You  may recall, a year-and-a-half-ago,  

we bought Microsoft’s Arm-based Windows  Dev Kit. It utilized the same Microsoft SQ3  

chip (which was just a rebranded Snapdragon  8cx gen3) found in a few quirky low-power,  

low-performance Windows laptops. No offense to  the folks at Qualcomm or Microsoft, but this thing  

sucked. Its performance under ideal conditions  was mediocre and ideal conditions were hard to  

come by because so much of the Windows experience  was wildly unoptimized for Arm—even after a decade  

following the original release of Windows RT.  But that was then—we live in the now. Not only  

has every single native app for Windows made  the transition to Arm, but massive quantities  

of 3rd party apps have too—including big ones—like  Google Chrome. Graphics APIs like DirectX, Vulkan,  

and OpenGL are said to work through mapping  layers and both Microsoft and Qualcomm have made  

huge efforts to ensure a smooth transition—they  were quick to volunteer that Apple is better at  

this than anyone and they hope to be compared to  them this summer when the X Elite laptops ship.

So, what is the Snapdragon X Elite? Well, they  gave me one in a cute little acrylic trading  

card. It is a bespoke laptop chip—not based on  a mobile chip—built on TSMC’s 4nm process coming  

with a 12 high-performance core CPU, Adreno  GPU, and in-house Hexagon NPU. Additionally,  

the on-board sensing hub houses an additional ISP,  on-board WiFi 7 by default, and the capability to  

be paired with up to 64GB of LPDDR5 memory,  a Snapdragon X65 5G modem, and NVMe storage  

over PCIe. The specs suggest Qualcomm is not  messing around—and from the benchmarks I saw,  

it consistently placed itself in between  the M2 and M3. Not shabby at all. Now,  

Apple is still certainly going to have the  upper hand with their Pro, Max, and Ultra chips,  

but this doesn’t aim to compete with those. While  OEMs can push the X Elite to run up to 90W for an  

extra performance boost, its reference-design  consumes just 24W peak. Very, very close to M3.

Now do I think that Qualcomm’s going to come out  blazing with the best laptop chips within the next  

3-years? Not really, no. But they’re hungry,  they’ve got Microsoft behind them, and they  

alluded to the fact that using heavy-duty GPUs  from NVIDIA or AMD wouldn’t be off the table in  

the future—something Apple has zero  aspirations for. And just like Apple, Intel,  

and everybody else, they’re really leaning into  their NPU for tasks that can use software-defined  

hardware for maximum efficiency and speed.  It’ll be exciting to see what form factors  

the Snapdragon X Elite embodies given its massive  power envelope available to OEMs—from netbooks to  

power-hungry beasts. Layer in the fact that this  is just their first foray into the X Elite line  

and that higher-performance chips are on the  roadmap, and well, we’ve got competition, baby.

So what’s Apple to do? Rush TSMC to the next  process shrink? Pivot to developing hotter  

more consumptive chips in the name of speed?  No. And Apple knows that’s not their core  

competency. Apple Silicon has always been about  sufficient performance with extreme efficiency,  

but physics are a cruel mistress and many  watching this channel don’t realize they’re  

already pushing up against boundaries that didn’t  exist for the M1 series. I still see comments that  

Apple silicon laptops are dead silent and  that’s just… dead wrong. We edit videos for  

this YouTube channel on a 14” M3 Max MacBook  Pro and the fans run at full-tilt nearly all  

the time—not just when exporting. And even with  fans ablaze, our NLE struggles to get exports  

out even close to the time the 16” MacBook  Pro can. It’s throttling—hard. Now, does it  

throttle to the point that it’s no faster than  an M3 Pro 14” MacBook Pro? No, but its sometimes  

slower than an M1 Max Mac Studio—something that  benchmarks would very much suggest is impossible.

My point here is that for years, Apple had  the exact same Intel chip SKUs as other  

computer makers. The silicon was never  their selling point. That is until M1,  

when that formula got flipped on its head and  Mac owners were—for the first time ever—able to  

be braggadocios about their computer’s speed. But  Apple is pushing against technological limits and  

others are catching up—so lets sit the silicon  aside for a minute and focus again on Apple’s  

hardware design. As I see it, the entire  MacBook lineup is basically the same. Sure,  

the Air has a lower-quality display and worse  speakers than the 14” MacBook Pro and the 15”  

Air is cheaper than the 16” Pro, but I mean  come on… these machines are closer in design,  

size, footprint, weight, and feature-set than  ever before. There’s no laptop that truly takes  

advantage of the form-factor provided by Apple  silicon’s insane performance per watt. Imagine  

a laptop even thinner, smaller, and lighter than  the 2015 12” MacBook—a computer that still feels  

impossible today—nearly a decade after its  release. Only this time, it doesn’t have to be  

hamstrung by a crappy low-TDP Intel chip and lousy  I/O. Would it be slower than an M3 Air? Sure. But  

how many people own—heck—a base M1 MacBook Air  and have never even approached the limits of that  

chip? I’d venture to say MOST—and if your silicon  can enable those impressive form factors… do it!

On the other end of the spectrum, why not  put an M3 ULTRA in a 16” laptop that’s a  

bit on the hefty-side—the style that gamers and  creators buy all the time on team Windows/Linux?  

A chip that just absolutely screams when  needed with the thermal headroom to do it,  

but while maintaining the excellent idle  efficiency offered by Apple’s low-power  

cores. It could be the first “gaming” laptop  with a battery that doesn’t die in like 4 hours.

What I'm getting at is this - when  Apple decided to make their own silicon,  

it was a bold, risky move. It paid off massively.  The M1 series will be remembered as some of the  

greatest computers ever. But it also feels like  that was the last time Apple really took a risk,  

and I think its time they stop sitting on  their laurels and get back to work before  

the rest of the industry catches up. What do  you think? Let me know in the comments below,  

but most importantly—and as always—stay snazzy.