Hacker News new | ask | show | jobs
by retskrad 1648 days ago
Alder Lake, Intel's flagship dekstop class CPU, is only 10-15% faster than the M1 Pro/M1 Max in CPU performance but it requires an insane amount of cooling and the power consumption is frankly ridiculous.

Seems like Apple Silion will be 5 years ahead of the competition - similar to how long it took for others to catch up with the first iPhone.

4 comments

Performance per Watt only matters to Intel when they are leading. Otherwise they brute force to be "fastest". I think that the ARM in Macs is on a different trajectory than Intel. The battery life along in an ARM based Mac is second to none. That's more important to most folks, IMO.
> ...the ARM in Macs is on a different trajectory than Intel

Agreed. Yes and:

I'd like more analysis and punditry (predictions) comparing SoC offerings from Apple, Intel, AMD, etc.

Especially wrt Apple's anticipated consumer features and markets, like AR/VR, video chat, integrated Apple Pay, whatever.

For instance, I want to hear more about Apple Silicon's codecs. More about biometrics UX (Face ID, Touch ID).

I think a server optimized SoC from Intel is right and proper. Ditto for Apple Silicon's strategy for mobile and media. And I don't anticipate Apple caring so much about server use cases. Maybe in-house stuff, like server farms for Siri. And I mean "market focus", vs "technical focus". While Apple Silicon is probably fine for server, I don't anticipate Apple caring, leaving those segments to existing cloud providers, eg Amazon's Graviton.

Now that I think about it, I'd like to see an Apple Silicon vs Graviton matchup. Just to get a sense of the landscape. For instance, I'd guess Apple Silicon's TPU (Neural Engine) is optimized for their software stack, whereas Amazon's doing something more general purpose, as befits their respective markets.

Sorry for stream of conscious, blathering for too long. I'm just writing out loud.

> While Apple Silicon is probably fine for server, I don't anticipate Apple caring, leaving those segments to existing cloud providers, eg Amazon's Graviton.

It mostly comes down to the supply chain, from what I can see. Apple is already pushing their luck with the 5nm node, their current strategy is pretty much just keeping everyone else off the cutting edge technology so they can monopolize TSMC's manufacturing chops. They could pivot to servers, but that would cannibalize their plans for the Mac and probably force them back onto the 7nm node, which would probably throw away the majority of their bragging points right now.

Plus, if history has shown us anything, people simply don't want Macs in the datacenter. xServe gave people the option, but that quickly went the way of Itanium after a year or two. Apple's history of "it's not a bug, it's a feature" and "you're holding it wrong" doesn't exactly appeal to corporate buyers who want to set-and-forget a system and have a number to call when it breaks.

Ya. I know this is true. Monospony. Constant jockeying around the supply chain for advantage.

Piqued, I briefly poked around trying to find absolute number of chips everyone produces, fab capacity, forecasts. Much like The Limiting Factor and others do for Li-ion batteries. No joy.

Raw performance has never been an issue for me on laptops anyway. I care more about the laptop not sounding like an airport because the fans have to spin at full speed to cope with web browsing. Or having to carry a charger around with me everywhere because the laptop gets 4 hours battery despite having a huge battery pack.

These new macbooks are like a dream. By far the best laptops I have ever used.

Memory bandwidth is still 76 GB/s. One reason I’m moving science workloads to M1 is for the cache line size and memory bandwidth, which bump by factor 2-3x the speed.
"science workloads". What kind of science workloads?
Computational neuroscience, but since your username is earth science I would mention one of the core algorithms which is quite a bit faster is the spherical harmonic transform.
Not the parent commenter. But, I know a few scientific tools in the genomics field that had the exact demand. The bottleneck was always memory bandwidth. 64bit wide floating point and memory bandwidth would make a huge difference for such tools.
Maybe not 5 but it is years ahead in terms of process.

Also Alder Lake is actually pretty normal in terms of power usage on actual workloads, almost no one is running AVX stress tests. The power draw is basically a symptom of intel massively pushing it past the efficiency frontier, undervolting is the new overclock thanks to massive stock OC

In most gaming benchmarks alder Lake is basically the same or less than Zen 3 in terms of power, last time I looked.

Apple will only be ahead of the competition for as long as they can buy out the whole TSMC's latest and greatest node. After that day, it won't be.
It's not only that. They have the knowhow, and a top-notch engineering team. The iPhone processor were developed over the course of 15 years and grew organically to become the M1. Even if you consider other mobile processors on par with Apple's (they are not) - it will take years before they can be adapted to PCs. And as latest Intel efforts show us processors that were not designed with efficiency as a key target from the beginning cannot be easily modified to challenge the M1 either.

Even without TSMC Apple has all the cards - experience in state of the art chips, phones, tablets and PCs, from the silicon to the software, all in one company. They should do some serious slipups before someone has even a chance to challenge them.

Don't discount the other engineering teams. AMD is currently not too far behind (even ahead in a couple benchmarks) despite being on TSMC's previous node. It's all gonna be pretty close.
And Intel seems to be storming back, albeit with pretty incredible power draw. Either way consumers win.
I think there is good chance, they remain at the top, for a while, but a low chance there will be a huge gap to the competitors, when they have access to the same tech stack.

The M1's speed and efficiency are mostly not the result of something ground-breakingly smart, but 5nm TSMC and a non-modular design, integrated chip platform. In other words: They played monopoly and made a huge compromise on the design front. I don't expect they could pull off another leap like that soon. The M1 isn't magic, as far as I can tell.

There is good reason to believe that’s not the case. Oh having access to TSMC 5nm to other designs would definitely help reduce the gap, but we can see from like-for-like node implementations of previous A series chips and other ARM designs that Apple has a significant lead in architecture too. M1 has some really impressive new architectural features of its own that it’s hard to imagine their competitors being able to replicate any time soon.
According to AMD you get >1.25 performance at 1/2 the power usage when chips go from 7nm to 5nm. Some of that might be architecture but when x86 gets to 5nm the difference in performance might be fairly negligible.

https://www.tomshardware.com/news/amd-unveils-zen-4-cpu-road...

The competition isn’t other ARM players, it’s Intel and AMD. Both are on larger nodes for now.