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
And yet it is not wrong. There has been really no movement in x86 in a long long while. Just boosting power and shrinking the CPU die to keep what they think is "on top". You are going to need a nuclear reactor to power an intel system sooner or later.
I wonder if you could overclock AS, I wonder what you would get. I'm betting it would blow the doors off of anything x86.
It takes time for a microarchitecture to mature. Look at all the years of incremental gains on the A-series chips that contributed to the launch of Apple Silicon. Intel’s E-cores will take time to get to a reasonable level of performance-per-watt, they are quite a few years behind. In the meantime Intel will keep pumping out hot and power-hungry chips. I predict their real problems will start when AMD takes some steps in the same direction as Apple.
Personally, I’m very happy with my M1 iMac. It’s plenty fast, runs cool and silent, turns on in about 4 seconds and doesn’t cost a lot in electricity. Those are the metrics I care about, how it performs on those counts. It matters a lot if the fans spin up and the thing sounds like a leafblower while I’m trying to watch the Red Sox. And what do I care about being the very fastest in multithreaded workloads when i see no noticeable difference in my use cases. As long as it doesn’t keep me waiting I’m good.
I think Apple has made an excellent choice in going with Apple Silicon. It allows them to focus custom resources on those scenarios which do keep people waiting, such as video export.
These power comparisons aren't all that meaningful, given that power consumption grows super-linearly with clock rate and voltage. The desktop CPUs are tuned for absolutely highest performance, because they wanted to clearly beat AMD. But there are some indications that you can reduce the power consumption very significantly without losing much performance, see e.g.:
Also, while the i9-12900K guzzles power when all cores are fully loaded, it actually seems more power efficient than AMD when running more realistic workloads, probably because of the E-cores. See e.g. here (scroll down to the power consumption section near the end):
Intel Core i9-12900K review: Intel. Is. Back.
Intel achieves a stunning come back for the ages with Alder Lake and the Core i9-12900K.www.pcworld.com
Considering that Intel is still at a disadvantage on the manufacturing process I think those are very good results. If they manage to catch up to TSMC, they'll probably take a clear lead.
Anyway, it'll be interesting to see what the upcoming laptop versions of the CPU can do.
That’s true … up to a point. You’ll notice in that tweet that between 150 and 125 you start to get much bigger drop offs. They’re starting to hit the area of the power curve where performance drops more precipitously. Getting down to M1 Max levels while maintaining any kind of performance will require more than just lowering power. However given that the laptop variants are delayed they may have different silicon characteristics (my guess is probably will) so we’ll see.
Also important to note that their manufacturing deficit is only wrt Apple and that still doesn’t close the gap. When comparing with AMD, they are at least on an equivalent node and very probably on a better one - Intel claims equivalent to TSMC 7nm+ or better on their latest node. And that’s probably fair. AMD is still on TSMC 7nm.
So in terms of perf/$ AMD will have to respond. Perf/W is still probably pretty good and raw perf AMD should have the answer with Zen3+. Should be fun times ahead.
Aren't the E-cored just updated Atoms? And Intel put them in because they didn't have good enough yields for 16 big cores like AMD, not because they are so great.
How much does the M1 Max actually consume under persistent load (i.e. when the "high power mode" on the 16" MBP kicks in)?
You probably meant to say exponentially, which is true. Power usage scales linearly with frequency (i.e. all else being equal, Power(2GHz) = 2*Power(1GHz)) while it scales quadratically with the voltage (i.e. power scales by V^2). In practice though, modern CPUs have dynamic power tables with varying voltages along their frequency range so the scaling is thrown even more out of balance. All of that is to say that a CPU at 5W might only be twice as fast at 20W (purely fictional example to illustrate a point), depending on the architecture, node and supporting power components.
Around 40 watts at most. Around 35W in Cinebench R23
High power mode appears to be mostly for the GPU and even then the effect seems … minimal. For the CPU it maxes out at 35-40W depending on the workload.
Intel 7 is really Intel 10nm+ … but they rebranded because “nm” is almost a nonsense no matter who is saying it and Intel 7 is probably around TSMC 7nm+, maybe a bit better if Intel is be believed. So they’re probably on a better node than AMD now.
Last edited:
Wow - M1 Pro and M1 Max hold up really well (in their laptop configuration) against the latest Intel desktop/workstation Alder Lake.
In my humble opinion, Alder lake looks to be a fantastic option for those who want an instant performance boost on existing x86 optimized workloads.
Back to the M1, M1 Pro and M1 Max, the comparison here is very impressive, especially when you consider the power and frequency delta to Alder Lake - M1 Pro Max per clock tick is both a more efficient and energy efficient design!
Appreciating Lemans’ opening comment that Alderlake does indeed blow past M1 In single core SpecINT - I also think that this only shares part of the performance consideration/perspective with how we should both acknowledge and perceive anandtechs’ results.
In real world applications - developed and optimized for Apples system on a chip approach to performance/energy efficient - we need to note that Apple is approaching performance through a combination of their own microarchitecture implementation of the ARM ISA and also using a combination of dedicated accelerators and co-processors for ISP, GPU, encryption, video encode, video decode etc…
In real world workloads where apps are optimized through Apples API stack and compiler optimizations for their Apple Silicon I still expect M1 Max to be neck and neck if not outright faster for the types of workloads that I am interested in which can leverage the machine learning accelerators and video encode/decode.
Of course I can compile synthetic tests that test only the ALU and FPU in isolation and compare with Alder Lake / Ryzen, but I feel that it kind of misses the point and potential of Apples architectural design a approach philosophy to both power efficiency and fast running applications.
I’m very interesting to see what the desktop comparison to Alder Lake is going to be from Apple…. This bodes very well for a future Mac Pro!
In my humble opinion, Alder lake looks to be a fantastic option for those who want an instant performance boost on existing x86 optimized workloads.
Back to the M1, M1 Pro and M1 Max, the comparison here is very impressive, especially when you consider the power and frequency delta to Alder Lake - M1 Pro Max per clock tick is both a more efficient and energy efficient design!
Appreciating Lemans’ opening comment that Alderlake does indeed blow past M1 In single core SpecINT - I also think that this only shares part of the performance consideration/perspective with how we should both acknowledge and perceive anandtechs’ results.
In real world applications - developed and optimized for Apples system on a chip approach to performance/energy efficient - we need to note that Apple is approaching performance through a combination of their own microarchitecture implementation of the ARM ISA and also using a combination of dedicated accelerators and co-processors for ISP, GPU, encryption, video encode, video decode etc…
In real world workloads where apps are optimized through Apples API stack and compiler optimizations for their Apple Silicon I still expect M1 Max to be neck and neck if not outright faster for the types of workloads that I am interested in which can leverage the machine learning accelerators and video encode/decode.
Of course I can compile synthetic tests that test only the ALU and FPU in isolation and compare with Alder Lake / Ryzen, but I feel that it kind of misses the point and potential of Apples architectural design a approach philosophy to both power efficiency and fast running applications.
I’m very interesting to see what the desktop comparison to Alder Lake is going to be from Apple…. This bodes very well for a future Mac Pro!
Oh I know why Intel put them in and yes they’re upgraded Atoms … however looking at them this is a pretty big upgrade for that line. If, and this is a big if, the core can be scaled then Intel can pull another Conroe. Atoms for years were the ignored, underfunded line of cores that even Intel never cared too much about. Going into their main product may change that even if they did that because their mainline cores are too big and hot. As @Bodhitree says this can take time and no one else will stand still. It depends on how aggressive Intel decides to be. This alone will not catch them up to Apple obviously, but it’s still big for Intel.
These power comparisons aren't all that meaningful, given that power consumption grows super-linearly with clock rate and voltage. The desktop CPUs are tuned for absolutely highest performance, because they wanted to clearly beat AMD. But there are some indications that you can reduce the power consumption very significantly without losing much performance, see e.g.:
Also, while the i9-12900K guzzles power when all cores are fully loaded, it actually seems more power efficient than AMD when running more realistic workloads, probably because of the E-cores. See e.g. here (scroll down to the power consumption section near the end):
Intel Core i9-12900K review: Intel. Is. Back.
Intel achieves a stunning come back for the ages with Alder Lake and the Core i9-12900K.
This is all very true, but then again, it only illustrates the issue further. There is no doubt that these CPUs will behave much more sanely at 20% lower performance, but then again, at 20% lower performance they are not as interesting to the enthusiast crowd. I started this thread mostly to reflect on the „come Alder Lake Apple will be sorry“ sentiment - and I don’t see what Apple should be sorry about. Looking at these results, and even taking the scaling issue into consideration, I have little doubt that M1 with its 40W sustained will effortlessly outperform Alder Lake at 60W in any useful workload (except Cinebench of course, Cinebench loves x86 and hyperthreading).
I hear this argument quite often and I have difficulty understanding it. Yes, right now Intel is approximately one node behind. But it’s architecture is 3-4 times less efficient even at its sweet spot. How would node parity help them? This would imply that overtaking TSMC would somehow allow Intel to reduce their power consumption by 70% - how do you imagine that happening? Their problem is not (just) the process, it’s the architecture. They’ve reached the limit of scalability and need something else. Pentium Pro architecture is amazing. It has been carrying Intel for the last 25 years. But they are running out of steam…
I am discussing it in the context of Apple vs. Intel offerings. So far the situation is that the 35W 10-thread M1 Pro is within 5% of 24-thread i9 Alder Lake 200+W in a GCC compile benchmarks.
It’s unlikely that Firestorm goes beyond 3.2ghz. If Apple could clock them higher they probably do for the Pro Macs. There were speculations that one of the reasons how Apple can a Hiebe it’s incredible power efficiency is by optimizing the silicon to run at a specific frequency, without much dynamic window.
I really should just move everything over to the M1 Mac Mini I bought as a second server and turn my Windows rig into just a NAS. Or better yet, throw the bugger out and buy a couple of 8 bay NAS enclosures and save myself a fortune on electricity
Do it! You won't regret it!
I’ve probably said it 10 times here before but I strongly believe that x86 is essentially dead. Big.little architecture with x86 is like putting a bandaid on a broken bone. Was pretty impressed with the benchmarks until I realized even the i5 has a 125 watt TDP. Wtf???
Sometimes I wonder about the environmental impact of gaming computers with 1 kilowatt power supplies ._.
Sometimes I wonder about the environmental impact of gaming computers with 1 kilowatt power supplies ._.
Last edited:
Nope, having 8 E-cores is actually better than having 2 or 3 more P-core (which would have taken about the same space) in lots of real world use cases. These aren't so slow as some might think and do add quite a lot to overall performance (when used correctly) and allow the chip to stay cool (and the fans quite) when the PC is under low to medium load (same is true for the M1 family).
Back to the top:
Intel is on top because they made a somewhat bigger than usual leap right now but AMD will answer soon with 3D-vcache (which should help with performance not powersaving) and Apple has already shown the slightly better A15 core which might or might not be the base for the M2 series next year.
Intel probably won't get into that range unless/until they catch up to TSMC 5nm. But I suspect the laptop Alder Lakes won't be far behind Apple's CPUs, and the laptops will cost significantly less than the MBPs.
In any case, the big.little architecture should give Intel lots of options to balance single-thread performance, multi-thread performance, and power consumption in different ways for different SKUs. I think this is a big step for x86 in general. To borrow a line from an Apple executive, "can't innovate, my ass".
It’s better for Intel, because their performance cores need a separate diesel generator to get to any meaningful levels of performance… but all in all, 8 E-cores barely manage to outperform a 2015 Skylake while consuming half the power. That’s the progress you get in 6 years : 50% power reduction…
Conversely, you could favorably compare the price-performance for the $300 i5 against the M1 Pro's, which remains exclusive to $2K+ setups. Intel trading blows with AMD once again only benefits consumers.
Technically, nothing prevents x86 from achieving reasonable parity with ARM, besides some minor overhead. Enterprise legacy baggage, on the other hand, remains a trickier problem involving Microsoft. To that end, Windows 11 looks like a decent first step.
Unlikely barring another 1-2 major architectural upgrades and Apple standing still in the meantime. Also while I’m sure there will be budget laptops the truth is that the premium brands - the space Apple actually competes in - prices stuff pretty similarly to Apple. Intel has cut prices, for the first time in what feels like forever, but it’s because of how behind they are in most other respects. They have to.
In any case, the big.little architecture should give Intel lots of options to balance single-thread performance, multi-thread performance, and power consumption in different ways for different SKUs. I think this is a big step for x86 in general. To borrow a line from an Apple executive, "can't innovate, my ass".
Maybe … of course there will be teething issues and these E cores ain’t exactly little … more like humongous.medium than big.little
Sure, one is a computer the other one is a paperweight that might just a bout good enough hold down 1 sheet of paper.
Turn the i5 into a computer and you are >1500$.......
To match you do have to buy 16GB of DDR4/5 RAM, a decent motherboard, a 3060, a decent quality monitor, etc … starts get a bit pricier than just that $300 for the CPU alone.
And that’s if you can find some of that for a reasonable price right now. True you can build it more slowly, but the total money is the total money and so price per $ not quite as good as it might seem … and that’s not even considering that it’s laptop vs desktop. Napkin math by CPU designers I’ve read put the intrinsic advantage of ARM v8 vs x86/64 at about 10% performance per watt. When comparing Apple and Intel, the rest is micro-architecture and fabrication with the former being the lion’s share. So yes theoretically Intel could catch up and Alder Lake is a big boost for them but that shows how far in micro-architecture design they have to go.