Sidebar Sidebar
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

J

JMacHack

Suspended
Mar 16, 2017
1,965
2,424
Arguments dropped:
6. EEPROM programming
7. User-programmable fpga
8. Intel artificially blocking off certain circuitry in software and selling it back for more money (this was supposed to beat Apple how?)
9. Intel blowing up their power requirements for a statistically irrelevant “win” is good
 
  • Like
Reactions: Stratus Fear and thedocbwarren
mi7chy

mi7chy

macrumors G4
Oct 24, 2014
10,625
11,296
  • Like
Reactions: huge_apple_fangirl
B

bombardier10

macrumors member
Nov 20, 2020
62
45
JMacHack said:
Intel seems to think it’s important, why else would the 12900k have 8 efficiency cores instead of performance cores?
Click to expand...
We are getting to much into technology. For me as a user important is the performance/price only.I have no need to know how the processor works. Intel i9-12900K is faster than 28 cores MacPro (with Xeon 28 cores)...
 
cmaier

cmaier

Suspended
Jul 25, 2007
25,405
33,474
California
bombardier10 said:
We are getting to much into technology. For me as a user important is the performance/price only.I have no need to know how the processor works. Intel i9-12900K is faster than 28 cores MacPro (with Xeon 28 cores)...
Click to expand...
So what? Apple hasn’t released their solution for that kind of workload yet. Why not wait until they do instead of ludicrously comparing jet fighters to jumbo jets.
 
  • Like
Reactions: JMacHack
diamond.g

diamond.g

macrumors G4
Mar 20, 2007
11,438
2,665
OBX
JMacHack said:
Well there must be some benefit to the big.little design if AMD follows through with this patent https://www.hardwaretimes.com/detai...and-process-node-for-different-core-clusters/

Perhaps the power/frequency scaling isn’t the best method, which I believe is the case. As it pertains to mitchy, he only claimed it’s the best method because Apple doesn’t use it. Somehow Apple and Intels chip designers should’ve consulted him.


I’m aware of this argument too, but since Intel claims it’s for power efficiency as well , I’m going with power efficiency being a factor. Likewise the AMD patent above.
Click to expand...
🤷🏽‍♂️ I was just offering a potential reason. Not saying it is the right one, lol.
 
F

Fomalhaut

macrumors 68000
Oct 6, 2020
1,993
1,724
JouniS said:
That would be an improvement to the current state of things. Today, if you don't buy the most expensive top-tier chip, you are likely to get a crippled chip with no option to re-enable the disabled cores. Intel, AMD, and Apple are all guilty of that.
Click to expand...
One would assume that the binned chips are mostly sourced from parent chips that have at least one fault, rather than "crippled chips". And there are at least 3 tiers of Apple Silicon (M1, Pro & Max) which are entirely different products so you can hardly called the M1 Pro a crippled version of the M1 Max - it's a completely different product.

If yields were so good that the only way to offer the binned variants was to deliberately disable functional cores, then there would be a case for trying to make this user configurable. It seems reasonable to make this a paid option. I'm sure the same kind of thing happens in the automotive industry.

Plenty of items in a market economy have their prices deliberately set by manipulating supply.
 
F

Fomalhaut

macrumors 68000
Oct 6, 2020
1,993
1,724
bombardier10 said:
We are getting to much into technology. For me as a user important is the performance/price only.I have no need to know how the processor works. Intel i9-12900K is faster than 28 cores MacPro (with Xeon 28 cores)...
Click to expand...
Faster in what metrics? Geekbench? The 12900K has fast single core performance, but appears to be slower than the 28-core Xeon in multi-core.

In real-world apps, there may be quite a difference in how an 8 performance-core machine compares to a 28-core machine. Let's see some application workload tests first before jumping to conclusion over what is better. The Alder Lake may well be better value for money, however.
 
mi7chy

mi7chy

macrumors G4
Oct 24, 2014
10,625
11,296
Fomalhaut said:
Faster in what metrics? Geekbench? The 12900K has fast single core performance, but appears to be slower than the 28-core Xeon in multi-core.
Click to expand...

1644654473559.png


1644654516126.png
 
  • Like
Reactions: bobcomer
M

mr_roboto

macrumors 6502a
Sep 30, 2020
856
1,866
mi7chy said:
Different companies have different approaches. My impression is...

Apple efficiency core is very low performance relative to performance core but benefits from low <1W power consumption at idle.

Intel's hybrid E-core is mid to 3/4 performance of P-core and more like essential core leaving out extraneous rather than efficiency core. Idle power consumption unknown.
Click to expand...
Your impression of Apple's core properties is... incomplete.

If you load up all 10 cores of a M1 Pro or Max, the entire E cluster (2 cores + L2 cache) uses about 650mW and each P cluster (4 cores + L2 cache) about 19000mW. I checked these numbers just now with powermetrics.

So yes, Apple's E core is far under 1W... at full blast, not idle. When close to idle, it's down in the 10s of milliwatts.

While Apple's E core is not as fast as Intel's, it's not horrible either - about 1/3 of Apple's P core. Quite good given how little die area and power they use.

mi7chy said:
AMD uses power/frequency scaling instead of E-core that doesn't take up die space but still manages to have low ~<1W idle power consumption.

On the surface, AMD's decision seems the most sound.
Click to expand...
On the surface, you have no clue about engineering decisions.

Apple's done such good work that they're in a completely unique position. Their P core uses only about 6W at peak single thread clock speed, yet it is highly competitive with Intel and AMD cores using 3x-5x as much power at their peak ST clocks. Unlike Intel's P core (IDK about AMD's), it doesn't use a ton of die area either. They add to this a hyper-efficient (both area and power) E core, with an OS scheduler designed to mostly utilize it as a place to run low-priority background tasks.

Apple and Intel both use voltage and frequency scaling. Not sure why you think that's unique to AMD.

If Alder Lake had P cores as good as Apple's, there would be no need for it to have Gracemont-like E cores. If Intel had the ability to design a true E core as good as Apple's, they'd throw a few of those in too, especially in SKUs designed for ultra low power laptops.

As for AMD, I'm sure they'd also love to have an E core as good as Apple's. Instead, they don't have any.
 
  • Like
Reactions: ahurst, huge_apple_fangirl, joeblough and 3 others
D

deconstruct60

macrumors G5
Mar 10, 2009
12,493
4,053
huge_apple_fangirl said:
View attachment 1957575
No, they aren’t power efficient compared to P-cores. Look at chart- at almost every power level, P-core offers greater performance. E-cores use lower power, sure, but also much lower performance. But they are area efficient “under given physical constraints”.
Click to expand...

Look at the chart? What are you looking at in the chart. That chart doesn't say way you are expressing. It says what I said.

Power efficient is going to show up chart as both the curve show up farther on the left of the power axis and/or with a higher slope that the other. Either the core can turn on and do productive work sooner ( with lower power while the other is off because can't even sufficiently coherently run ) or it is getting more "bang for the buck" the performance goes up per watt faster.

I think you are caught up in how long the "tail" of the curve is. That isn't power efficiency at all. All of these processors are going to hit a "knee" in their curve where it bends over and start loosing slope (and trend more toward being horizontal, "flat". ) That flatter part of the curve is far more about how far you can overclock the core , not power efficiency. Overclocking is chasing diminishing returns of applying more power to get relatively smaller increments of performance. you have largely left "efficiently".

Generally things that are tuned high efficiency tend to overclock worse ( a flatter tail once hit the 'knee' . ) . This is in part driven that there is a catch-22 of being "too good". The higher the performance the more likely going to "hit the wall" on some part of the rest of the system. Fetch too fast on too much and start to get cache misses. Load more out of "farther" parts of the memory hierarchy. Form completed data and have to store onto the storage system interim results before moving on . Things like that.

So back to the graph. The graph on the left... The P core curve doesn't even start before the E (gracemount) core does. Between where the E core starts the P core starts the the division of the (e core score ) / ( p core score ) is infinity. You dividing by zero. There is no question the E is doing a better job. Second, there is only a high correlation between Power and clocks . Just because two cores implementation are at the same power doesn't mean they are at the same frequency.

The graph on the left seems more constructed to give an implementation that there is a smooth hand off between running mostly on E cores to running on P cores. The likelihood that the slopes match exactly at the handoff point might have been a design objective (or wishlist) or this is some aggregated or 'cherry picked' benchmark workload. That the E core goes flat after the curves intersect doesn't mean have to run the E cores in that "too overclocked" mode. However, may "have to" in some circumstances for "better than nothing" results.

The graph on the right is one of those "better than nothing" situations. They can flog the E cores to get a 50% boost on MT results over just 4 P cores by themselves. If running 4 P cores at high workload then the frequency ( power ) is going to drop because have to share out of the capped power budget. E are better at sharing power than P cores are. That is mostly how they squeeze that extra increment out. The P cores don't have to drop as far to share and the E cores also contribute to filling what they can get done.

If try to clock synchronize the P and E cores closer to the P core Max then yes will soak up gobs of power. However, most of that is going to be P cores at the root of that problem. When aggregate them all together the E cores just let them go further down the diminishing returns portion of the curve. That extension is mostly flat so it isn't about power efficiency. In that narrow usage zone the E cores aren't highly energy efficient. That isn't the point. Relatively they are better than the P can be with the same area budget. So area effective there but trying to run them in "overlocked" isn't clear test of their power efficient. Run them where they are designed to run to measure that.


Intel and AMD run their cores over a wider operating range. Apple cores probably do worse when run them past their diesign operating range. Apple's comparison graphs chop the tails off of their curves because they don't try to sell that range as a 'feature' . ( and probably not that impressive either ... hits the wall and goes mostly flat. )
 
D

deconstruct60

macrumors G5
Mar 10, 2009
12,493
4,053
mi7chy said:
This article does seem to confirm that.

https://chipsandcheese.com/2022/01/28/alder-lakes-power-efficiency-a-complicated-picture/
Click to expand...

Chuckle. This is like bizzaro world . Posting from another dimension? The article doesn't really confirm that . I suspect you are latching onto just this sentence....

" ... This looks bad for the Gracemont based E-Cores. According to Intel, they can’t beat the P-Cores at any power level, meaning the E-Cores are only efficient in terms of area. ..."

The article goes on to disprove this. And frankly as I have already outlined the Intel graph doesn't say that all. In the range before the P core can even turn on the E core is better. The author tries to spin that as just being "low power" as opposed to power efficient. If getting no user work done then that is a loss.

So then the author then goes into talking about data .... which doesn't support his stated hypothesis.

First graph at points close to the cross over point outlined.
4 E cores 3.04FPS at 5.5W ---> 0.55 FPS/W
4 P core s 2.65FPS at 5.35W ---> 050 FPS/W

only area efficent? Not. It is operating at a better "bang for buck". Before the cross over of the curves the E cores is obviously better.

Author's off in the weeds commentary

"... With a vectorized workload, Gracemont only beats Golden Cove when running at ultrabook-throttlefest speeds and drawing under 6W. ..."

Well if the SoC is actually deployed in a ultrabook has is the E cores running at that frequency a "bad thing". This is not objective analysis... In the context of chasing desktop , overclocking, tech porn benchmarks "ultrabook" is bad but that has little to do with being efficient or not.

Second graph
4 E cores 25.24MB/sec at 13.93W --> 1.81 MB/sec/W
4 P cores 21.41MB/sec at 14.95W ---> 1.43 MB/sec/W

Author's comments

"... Below 15 watts, Gracemont achieves higher performance while consuming less power than Golden Cove. ... "

Area efficient only? Not really.

"... Looking through the entire power range, Gracemont struggles to scale well past 3-4 watts per cor ..."

Scaling up to a maximum overclocking range isn't about power efficiency... It is about max power consumption can get away with without destabilizing the core(s). Why do you want to go higher than 3-4 W per core? That is a root issue.

So then get to his "Energy Efficiency" section of the article.

"... At stock speeds, Gracemont cores are more efficient. Although slower, they draw so little power that they end up consuming less energy to finish the job. Golden Cove can be efficient too – just not at stock. Between 3 and 4 GHz, these P-Cores can give the E-Cores a run for their money. ..."

What a shocker. Absolute shocker. If you use the E core as it was designed to be used it performs better. But is that the evaluation criteria here? Nope.

"... That means running Gracemont above 3.2 GHz is pointless if energy efficiency is your primary concern. Running the E-Cores at 3.8 GHz basically makes them worse P-Cores. But that’s exactly what Alder Lake does by default. ... "

Default? Hmm, Intel really that dense to run their own design outside the limits?

i7-12700 mainstream Core i7 gen 12 offering at 10 cores.
https://ark.intel.com/content/www/u...12700-processor-25m-cache-up-to-4-90-ghz.html

"...
  • Efficient-core Max Turbo Frequency3.60 GHz
  • Performance-core Base Frequency2.10 GHz
  • Efficient-core Base Frequency1.60 GHz

..."

3.8GHz where????? The Max Turbo is set to 3.60GHz.

The i7-12700K? Yes the E cores max out at 3.8GHz (and base at 2.7GHz) . Are E cores cranked up too far really worth the additional money Intel is charging? Maybe ( will probably need to pay substantively more to over cool the package substantially past what Intel's default requirements are. ). However, they are generally not an "high value" overclocking feature.

If 'K' and 'KF' and higher (i9) Intel is selling conflicted packages. That doesn't make the E (gracemount) cores not energy efficient... that is more sales marketing parameters. Highly likely 85+% of what Intel sells ( laptops and non extremist desktop ) isn't being sold that way. [ decent chance that it is 95+ % isn't this overclocked focused stuff. ]

For that "overcloking" skewed market more than a few folks there just buy on GHz numbers. Higher number , means they pay more. Most likely this will work to crank the E cores into a more less efficient zone because this crowd isn't looking for efficiency. In fact, higher power consumption is a "feature" many of them are actually seeking.


[ Windows 10 being sold (or used ) with these is probably a small contributing factor. If the OS doesn't "know" not to ask E cores to Turbo to "infinity and beyond" then can get quirky behavior. For Windows 11 and effective use of Thread Director the turbo of 3.8GHz isn't bad if the OS never 'asks' for that. If find P core suitable threads move them there and generally don't overload the E cores if the specs allow it anyway. ]


edit
P.S.

image-23-1.png



AMD has a steeper slope ( better perf/w ) for portion of the range and then hits the wall and goes almost flat ( or flatter than the alternatives) . It is a common trade off.
 
Last edited:
  • Like
Reactions: JMacHack
M

mr_roboto

macrumors 6502a
Sep 30, 2020
856
1,866
deconstruct60 said:
Look at the chart? What are you looking at in the chart. That chart doesn't say way you are expressing. It says what I said.
Click to expand...
No, it doesn't.

deconstruct60 said:
So back to the graph. The graph on the left... The P core curve doesn't even start before the E (gracemount) core does. Between where the E core starts the P core starts the the division of the (e core score ) / ( p core score ) is infinity. You dividing by zero. There is no question the E is doing a better job. Second, there is only a high correlation between Power and clocks . Just because two cores implementation are at the same power doesn't mean they are at the same frequency.
Click to expand...
The post you're replying to made no mention of frequency at all. I also have experienced you responding to one of my posts with a torrent of paragraphs about stuff I simply didn't say.

deconstruct60 said:
The graph on the left seems more constructed to give an implementation that there is a smooth hand off between running mostly on E cores to running on P cores. The likelihood that the slopes match exactly at the handoff point might have been a design objective (or wishlist) or this is some aggregated or 'cherry picked' benchmark workload. That the E core goes flat after the curves intersect doesn't mean have to run the E cores in that "too overclocked" mode. However, may "have to" in some circumstances for "better than nothing" results.
Click to expand...
You need to actually read what Intel says the left hand graph is about instead of going off on what you think it's about. It's right there in the graph's caption: "P-Core delivers higher Performance on single and lightly threaded scalable apps." And the captions for the curves in that graph: "1 E-core", "1 P-Core".

The reason there's a spot where the two curves almost touch is just that this is often what happens when you plot perf/power curves for two different CPU architectures. It doesn't mean Intel's trying to imply handoff.

It's not hard to simply accept Intel at their word and use that as the basis for further reasoning: "E-core provide higher computational density under given physical constraints". That's the caption for the right hand graph, where they contrast the MT performance of a 4P configuration against a 2P+8E configuration. If you go look up annotated die photos of Alder Lake family processors, you'll find that a rectangle drawn around eight E cores is not much larger than a rectangle drawn around two P cores. With that context, the message of the slide can be understood as: "By substituting eight E cores for two P cores, we gain 50% MT performance over a 4P config".

So Intel's definition of "E"fficiency for Alder Lake E cores is mostly about area, not power. Golden Cove's high ST performance comes at an extreme area cost, and they weren't going to look competitive against high core count AMD processors without doing something about that.
 
  • Like
Reactions: huge_apple_fangirl
R

Romain_H

macrumors 6502a
Sep 20, 2021
520
438
mi7chy said:
A lot of workloads don't run on upsized iPad SoC so it's irrelevant.
Click to expand...
Oh please. Are you trying to belittling the M1 by calling it „iPad SoC“? Tell you what: That upscaled toy beats Intels best in all but one metric. Easily

If a workload runs on a given platform is not a matter of hardware. Of course M1 runs any workload Intels run. There is no reason why it wouldn‘t
 
Last edited:
  • Like
  • Haha
Reactions: JMacHack, huge_apple_fangirl and bobcomer
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom