Emulated 12900K for mobile has better performance with low power consumption like M1 Max?

[sunny5]

알더레이크 모바일 엄청 섹시하네요 - 미코

6+8코어가 35w만 먹고 M1Pro/Max만큼 성능이 나온다니 인텔이 이렇게 분발할줄은 상상도 못했습니다

meeco.kr

Just an emulated result but according to China, 12900K with 6 big cores(3.0ghz) and 8 small cores(2.4ghz) used only 35W of power usage and the Cinebench result was around 14300 which is similar to M1 Max's result.

If this is true, then Intel is able to make very efficient CPU with x86 architecture and 7nm which might be very threatening result to Apple Silicon. But still, it's just an emulated result and Intel has more small cores but if it's true, I dont think x86 is the main bottleneck for having ARM's efficient power consumption anymore.

Yes, we need to wait and see how Alder Lake mobile CPU performs but it seems Intel might be able to make M1 and M1 Pro/Max grade CPU with similar power consumption.

https://www.reddit.com/r/hardware/comments/qo41ss
Thoughts?

 

[bill-p]

Thoughts?

Well, the site is Korean.

Also note the CPU package power being 43.83W in their screenshots. Possibly higher. It is not 35W.

Also we know 12900K will use up to 240W. This site is saying they can use 1/5th the power but lose only 30% the peak performance? That's... wishful.

It's honestly hopeless for Intel. They are using up to 7x more power just to beat the M1 Pro/Max.

 

[Kazgarth]

Many things to digest here.

The Intel machine is running 8 efficiency cores, while the M1 Max has only 2 (It's designed for high performance after all).

The M1 Max power usage includes 32 GPU cores and 32GB of memory.

That Intel power draw reading excludes the GPU and I suspect even the memory.


Edit: Videocardz.com just posted leaked score for the high-end laptop variant of the i9-12900

 

[sunny5]

Well, the site is Korean.

Also note the CPU package power being 43.83W in their screenshots. Possibly higher. It is not 35W.

Also we know 12900K will use up to 240W. This site is saying they can use 1/5th the power but lose only 30% the peak performance? That's... wishful.

It's honestly hopeless for Intel. They are using up to 7x more power just to beat the M1 Pro/Max.

The image shows that 12900K at 35~45W beats M1 Pro/Max. Yes, it's an emulated result but 12900H might be able to have M1 Pro/Max's performance with 35~45W power consumption.

 

[leman]

I do not doubt the results, but I don’t buy the rhetorics. First, M1 Pro/Mac running tust test uses 34W package power as reported by Anandtech (this includes GPU and RAM). The ADL uses 35W on CPU clusters, 44W package power plus an unknown on RAM power. No idea why the iGPU usage is so high and we don’t have detailed breakdown on CPU vs. GPU power for M1.

Second, cinebench is one benchmark where Apple Silicon does not perform very well, and it has been discussed a lot. The M1 is hopelessly behind the full i9 ADL in Cinebench. But it does very well in a number of SPEC tests.

In the end, yes, this experiment demonstrates that ADL can have similar efficiency as M1 on Cinebench, which is an amazing result for Intel. It is definitely competitive with Zen3 here. But we need to see more benchmarks of different types to get a clear picture.

 

[ADGrant]

Second, cinebench is one benchmark where Apple Silicon does not perform very well, and it has been discussed a lot. The M1 is hopelessly behind the full i9 ADL in Cinebench. But it does very well in a number of SPEC tests.

Someone with a 27" 10 core i9 iMac and a 16" Mi Mac MBP ran Cinebench on both and that i9 was slightly faster. However, it was the only benchmark where the i9 iMac beat the M1 Max MBP.

There are many theories for why cinebench seems to favor the i9. The suggestion is the software has just been better optimized for the Intel architecture. The i9s do have either more cores or more full power cores. They also support two execution threads per full power core.

 

[crazy dave]

I do not doubt the results, but I don’t buy the rhetorics. First, M1 Pro/Mac running tust test uses 34W package power as reported by Anandtech (this includes GPU and RAM). The ADL uses 35W on CPU clusters, 44W package power plus an unknown on RAM power. No idea why the iGPU usage is so high and we don’t have detailed breakdown on CPU vs. GPU power for M1.

Second, cinebench is one benchmark where Apple Silicon does not perform very well, and it has been discussed a lot. The M1 is hopelessly behind the full i9 ADL in Cinebench. But it does very well in a number of SPEC tests.

In the end, yes, this experiment demonstrates that ADL can have similar efficiency as M1 on Cinebench, which is an amazing result for Intel. It is definitely competitive with Zen3 here. But we need to see more benchmarks of different types to get a clear picture.

Actually the package power on the M1 *might not include dram (but does include memory controller) and while it does include GPU, Apple is very good about keeping power off to accelerators which aren’t in use. Plus I believe you can get powermetrics to report just cpu core so if the reviewer knows what he’s doing that *should* have been a fair comparison (and given Anandtech’s numbers I suspect it is on that front).

My issue is a little different: Mostly that software reporting of hardware power can get a little janky when that hardware is operating waaaay outside its normal operating range. We even see this with powermetrics under less extreme conditions where powermetrics reports more power being used than being sucked up by the wall (and not enough to justify being worried it was using battery + wall if the computer even had a battery).

Plus other reviewers have tested reducing power in the i9 against perf in CB23. While no one else has run it that low while simultaneously disabling two P-cores, the relatively abbreviated pattern is one of sane reductions in performance for the reductions in power that you would expect from a desktop power curve. So I’d like to see this replicated to make sure it actually is the behavior and I’d like to see wall measurements as a sanity check on the software reporting.

I mean I think Intel’s E-cores look really promising but hmmm … even for me this is in the realm of “extraordinary claims require extraordinary evidence”.

 

[bill-p]

The image shows that 12900K at 35~45W beats M1 Pro/Max. Yes, it's an emulated result but 12900H might be able to have M1 Pro/Max's performance with 35~45W power consumption.

It won't.

Again, you're "wishing" for a CPU to only drop 30% of its peak performance by running at 1/5 its max power consumption? That's physically impossible.

Also, the M1 Pro/Max is 30W max for the CPU.

 

[JamieLannister]

INTC firing on all cylinders to promote their crapware yet again. Iteration after iteration is playing catchup to AMD. Please stay within your bounds and not look at Apple because there's no intel laptop that is near the quality and performance of the new M1 Pro/max variants. So please GTFO with this rhetorical bull.

Intel is desperate as usual and will pursue 110% marketing budgets to overcome their stigma. Worse company on the planet.

 

[Gnattu]

We have multiple factors here:

- The clock and the voltage is extremely low compared to the normal operation frequency/voltage of an 12900K. The P-cores are even running at a frequency lower than its base clock, and the E cores runs just at base. I think the voltages has been manually undervolted as well, but I may be wrong.
- Cinebench has an extremely low CPU(pipeline) utilization on Apple's CPUs. Although CPU is reported to be used at 100%, that is only 100% CPU Time, not the whole pipeline. The power consumption(for CPU, not the whole package) on Cinebench is ~20%-25% lower than other CPU heavy tasks at full load on M1 as a result. Don't know M1 Max numbers because I'm still waiting for mine.

Then I can hardly mark this as a good comparison because such "emulation" looks like that the emulated 12900H will perform slower in almost anything other than Cinebench. It does not have the high frequency to be faster in single thread, and "similar" to M1 Max in Cinebench usually means slower in almost anywhere else.

 

[satcomer]

Cinebench is nott even a silicon tester, only Intel! So your comparison failed!

 

[bombardier10]

Cinebench is nott even a silicon tester, only Intel! So your comparison failed!

Cinebench R23 is made for Apple M1 too.

 

[satcomer]

Cinebench R23 is made for Apple M1 too.w

Yea running in Rosetta though, I said natively running like it doing in Windows 11!

 

[crazy dave]

Yea running in Rosetta though, I said natively running like it doing in Windows 11!

CB23 is native arm64 … optimized for M1 is a different story

 

[satcomer]

CB23 is native arm64 … optimized for M1 is a different story

That's exactly my point! The tester they used doesn't have a native tester (unlike other native testers) that runs native on Windows but has an extra layer on Macs!

 

[crazy dave]

That's exactly my point! The tester they used doesn't have a native tester (unlike other native testers) that runs native on Windows but has an extra layer on Macs!

It’s not extra layer in the M1 … CB23 is native to the M1’s architecture aarch64 so there’s no translation layer. But rather it’s more like on Intel, CB23 is written in a way that drives the processor to its fullest extent but has difficulty doing the same when running on the M1. It may also have this trouble on Zen 3 for some reason. So it isn’t even necessarily an x86 vs ARM problem though that may be part of it.

I’m not disputing that there’s a problem with CB23, just slightly correcting what the actual problem is.

 

[senttoschool]

CB23 is native arm64 … optimized for M1 is a different story

Cinebench is a really poor benchmark for the M1 even if it's ARM64.

 

[sunny5]

Apple moved to ARM for better power efficiency but Intel with 12th gen and x86 already achieve M1 Pro/Max's power efficiency which is very disappointed. I hope it's not true tho.

 

[Gnattu]

Apple moved to ARM for better power efficiency but Intel with 12th gen and x86 already achieve M1 Pro/Max's power efficiency which is very disappointed. I hope it's not true tho.

Again, this simulation emulated an '12900H'

- slower than M1 Max in almost everything other than CineBench
- consumes more power with CPU cluster only than the whole package of M1 Max

How can your draw such conclusion?

 

[sunny5]

Again, this simulation emulated an '12900H'

- slower than M1 Max in almost everything other than CineBench
- consumes more power with CPU cluster only than the whole package of M1 Max

How can your draw such conclusion?

- slower than M1 Max in almost everything other than CineBench
Source please?

- consumes more power with CPU cluster only than the whole package of M1 Max
We are talking about emulated 12900H at 35W, not desktop version.

With Cinebench, 12900K at 35W was still faster than M1 Max's CPU in terms of multi core. This is what I've been talking about: Can Intel achieve ARM/Apple Silicon's power efficiency with 7nm and x86?

Yes, it's an emulated or simulated result but 12th gen or alder lake uses modular system so the actual result might be same.

I'm talking about a legit topic with Alder lake. What if Intel can achieve ARM's power efficiency with 7nm and x86 just like that?

 

[Ceed]

I'm talking about a legit topic with Alder lake. What if Intel can achieve ARM's power efficiency with 7nm and x86 just like that?

Sounds like great news for everyone.

 

[bill-p]

With Cinebench, 12900K at 35W was still faster than M1 Max's CPU in terms of multi core. This is what I've been talking about: Can Intel achieve ARM/Apple Silicon's power efficiency with 7nm and x86?

Look again. CPU Package Power is 43.83W. Not 35W.

 

[sunny5]

Sounds like great news for everyone.

Well, technically bad news for Apple if Intel really achieve M1 Pro/Max's power efficiency.

 

[Gnattu]

- slower than M1 Max in almost everything other than CineBench
Source please?

From whatever source we have that shows M1 Max is slower in Cinebench.

 

[sunny5]

Look again. CPU Package Power is 43.83W. Not 35W.

Still not a big difference. Intel tends to use way more power than that like 60~100W.