That's how I see as well. Many people point to Alder Lake as a proof that x86 can still improve, but I don't really see this. Golden Cove was a massive core redesign with a radically wider frontend and OOE depth, and yet the IPC improvements remained fairly mediocre. I am not confident they can deliver systematic improvements on this path. And sure, Alder Lake has a massively better parallel throughput, but that is achieved by side-stepping the issue and throwing many low-performance, low-power cores at the problem rather than real baseline improvements. It is an engineering achievement, no questions, I am not just confident that it is scalable.
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.
Intel Alder Lake vs. Apple M1
- Thread starter leman
- Start date
- Sort by reaction score
That's how I see as well. Many people point to Alder Lake as a proof that x86 can still improve, but I don't really see this. Golden Cove was a massive core redesign with a radically wider frontend and OOE depth, and yet the IPC improvements remained fairly mediocre. I am not confident they can deliver systematic improvements on this path. And sure, Alder Lake has a massively better parallel throughput, but that is achieved by side-stepping the issue and throwing many low-performance, low-power cores at the problem rather than real baseline improvements. It is an engineering achievement, no questions, I am not just confident that it is scalable.
Well, the problem is that the best showcases for ARM architecture are closed proprietary market offerings like Apple's M-series and Amazon's Gravition which is also only used for their own AWS instances. In the open server market, ARM still faces significant uphill battle vs x86 looking at Altera, etc.
For now. There are plenty of companies willing to address the wider market (nvidia, qualcomm, samsung, etc.) and they will keep improving faster than x86 can.
I've probably said it already in the 20 some pages of this thread, but the M1 stands as the existence proof. It has now been demonstrated that Arm can compete with x86 and win on key metrics. Power/performance sounds like a smartphone concern, but the biggest cost in running a datacenter is cooling-- power performance sounds good in that environment. So I agree the established players will be looking at their ability to move quickly into this space.
And the success of the M1 has just chummed the waters for every little startup out there looking for venture funding to take a piece of Intel's market. Where as a few years ago the idea of challenging Intel in their core markets would have gotten you laughed out of the room, I'd imagine there are more ears open to opportunities today.
What is this "open server market" of which you speak? Amazon, Google, Microsoft and Facebook are all reported to be working on custom datacenter chips. Apple is clearly capable of it. How much of a server market is left without those, and how excited are the smaller players going to be using x86 when they see the big players have all abandoned it?
Ok, but then we're talking about Intel becoming Lockheed. Essentially with a single customer and no longer really relevant to this conversation.
Its remarkable that we can hear Intel fans over the loud Intel fans.
As I see it alder lake got all this at massive peak power draw - if you limit it to say 100 watts max the improvement over skylake looks far less impressive.
Never mind that there is supposedly (intel marketing aside) at least 2-4 process improvements there as well.
Which is it intel? No node improvements or does your “improved” design really suck that bad?
You can’t have it both ways.
It looks even worse if you go back to say, sandy bridge which was on 32nm. Sure there are improvements but on that many process node improvements, for how much performance per watt change??
Re the only performant arm designs being closed… show me where I can say, obtain a reference design for alder lake…
The only reason there aren’t more performant arm designs so far is that nobody has attempted to take on the pc market with one before because the barrier to entry is so high.
Apple is literally the only player who could do it because they have an existing software stack with enough customers and a massive bank balance.
Even AMD were nearly killed trying to take on intel - without having to shift potential customers to a new ISA (largely due to intels illegal business practices) and this is their core market.
Last edited:
Apple could do it because they sell more processors than Intel does, and they sell them at high margins. I’d have to look up the numbers again, but Apple sells more iOS devices than Intel sells processors. It’s not just their bank balance— they have a revenue stream tied to the processors they develop that is bigger than Intel’s and thus can devote a comparable R&D budget to them.
Other phone makers are high volume, but the bulk of that volume is typically lower margin parts. For all the people arguing that “market share is more important than profit share”, think about that statement again in this context.
And Apple never let themselves fall into the “smartphones are limited devices” mindset.
Yup.
Could you imagine a brand new player trying to build and sell a high performance ARM variant?
No existing market - it would be dead on arrival unfortunately. Doesn't matter how fast it is if it has no OS platform or customer base to take advantage of it.
Even the original ARM computers - the Archimedes were killed by the PC, even though they were more performant.
I suspect part of the reason why Intel is facing the problems they do today is because they both design and manufacture their own chips for sale, meaning they have 2 business models which are fundamentally at odds with each other. Chip design is costly, so you want to milk it as long as you can in order to recoup the R&D money, but the market is constantly demanding something better and faster. Stop innovating and eventually, someone will do it better than you.
Apple technically doesn't sell processors, and that's their greatest strength. They use their custom chips to differentiate their products from the competition, and use that differentiation to justify their higher prices. And they can do it year after year after year because for them, their processor is but one of several ingredients that enables the user experience behind the iPhone. The money they make by shipping the latest processor in their latest iPhone or iPad is more than whatever money they could have saved by sticking with an older chip design.
Apple is also better able to customise the design of their processors to better meet the needs of their own hardware, in ways that Intel would never be able or willing to do. Take Alder Lake for example. Yes, it boasts amazing performance, but you are also never going to find it in any Intel Mac, because it focuses on all the wrong metrics that Apple has zero interest in competing in. Apple wants their MacBooks to have great performance, solid battery life, with an emphasis on sustained performance when your laptop is not plugged in to an external power source (ie: power efficiency and low heat generation).
All the areas that Intel has notoriously dropped the ball in.
There may be some selection bias here-- Apple is "winning" so we think Apple has a winning strategy. If Intel was still kicking ass, people would say it's because they have complete control of the design down to the process in the same way we say Apple succeeds because they control the system and the CPU. That model worked for Intel incredibly well but then they stumbled and, I believe, their corporate arrogance made them slow to protect their business by backfilling their manufacturing.
If Apple fails to deliver on a few generations of CPU, we'll say their competitors succeeded because they didn't rely on in house CPU designs. If TSMC chokes and Samsung pulls ahead, we'll say the Galaxy devices are succeeding because Samsung controls the process.
This, I completely agree with. Taking a system level approach to design give Apple an advantage. Intel has to make generic stuff because they're marketing to a generic customer-- and they made a name for themselves by never leaving a legacy customer stranded. Apple controls the whole HW/SW stack and can benefit from a heterogenous computing model that they can adapt overtime transparently to the upper layers of the software stack. If Intel adds an instruction to a processor, it is never, ever, deprecated. I don't think Apple will have that constraint. If Intel adds a coprocessor, they're kind of stuck with it for life-- if Apple decides to rip out the Neural Engine and replace with something else they can just abstract that hardware away and eventually just restrict OS versions to CPU generations.
Graviton3 already beats the pants off Intel and AMD in terms of cost/performance in the cloud. Even Ampere ARM chips are better. That's what ARM is really good at even right now.
ARM in the server does one thing really well, which is to give you many cores for cheap. Most cloud services do not require super high single core performance. Rather, they require many slow cores to handle user connections.
In terms of pure performance, I don't see it within the next 2 years. Nothing in ARM's roadmap suggests basic ARM designs will beat Intel's Meteor Lake in performance or AMD's Zen4/5.
I think Qualcomm with Nuvia might stand a chance. Mediatek is also making a lot of noise with moves into the laptop market.
Apple Silicon is excluded from this conversion of course.
Last edited:
Intel are losing imho because of a few things:
- inability to retain top talent (rumors of toxic workplace culture)
- un-willingess or inability to migrate to a cleaner design (to be fair, they'd need microsoft to be on board otherwise people would just buy AMD for better performance with legacy code compatibility - apple control the entire stack and have an advantage there)
- complacency after being so far ahead of AMD between say 2011 and 2016 - 5 years is a long time
- fab problems, perhaps caused by #1
It's not just about cost/performance. It's also about performance per se. We still don't have benchmark numbers for Graviton3 of course, but if improvements claimed by Amazon are true, it should compete well in single-core performance with modern x86 designs.
They might just surprise you
Strange things can happen if you start optimizing for performance (like recent ARM designs do) and not for die area (like ARM traditionally used to). It's an attitude shift, really. If you expect your products to run only on the phones, well, that's what you build. Once your tech starts to make its appearance in high-performance products, well, that's what you try to build. And there is no indication ARM's improvements are about to stop.
Also, the M1 is used in a freaking tablet. Never mind a 150-200+ watt notebook.
For the applications these machines are aimed at (not chess) they outperform desktop machines consuming upwards of 6-8x the power, costing upward of 3-10x the price.
Meanwhile, my AC is busted, its currently 35 degrees C and...
edit:
it's literally running right now with 15 other apps running in the background.
will update with the finished score when it completes. the machine is still totally responsive doing stuff while its running in the background.
Meanwhile, my AC is busted, its currently 35 degrees C and...
edit:
it's literally running right now with 15 other apps running in the background.
will update with the finished score when it completes. the machine is still totally responsive doing stuff while its running in the background.
Attachments
Last edited:
That's really not true. I have a middle of the road desktop that outperforms the M1, and it was cheaper than my M1 MBA before I added a couple hundred for 128G of RAM. Yes, it definitely uses more power, but more expensive and slower, definitely not.
You do know that's quite an old chip? And that's i9-9880H actually.
Ah, didn't realise it was mobile. Was definitely aware it isn't current. The mobile suffix doesn't mean much anyway, given OEMs are running them over 100 watts turbo.
Still... 10 series isn't much faster (5%), 11 series was a joke and slower than 10 series.
And as with any performance comparison: the question is "at what"?
Still wrong unless you're only counting mobile chips. My desktop is a 10 and it's about a 3rd again faster than the 9880H. (i9-10900)
I agree with you on mobile chips, at least the one's I have experience with, but not desktop chips, and you did say "desktop". Cooling is a major bottleneck with intel high performance chips.
Multicore performance.
How does it perform on respected benchmarks? Ones like spec or geekbench. The kind that a reputable reviewer would use. Anandtech etc.
Did you just compare the top end AMD to the lower end M1?
Try this for a move even example
Apple M1 Pro vs AMD Ryzen 9 5900HX: performance comparison
We compared Apple M1 Pro (3.2 GHz) against AMD Ryzen 9 5900HX (3.3 GHz) in games and benchmarks. Find out which CPU has better performance.
nanoreview.net