Intel Core i5 12600K / Core i9 12900K "Alder Lake" Linux Performance

[drakonas777]

So, having read quite a lot of reviews here are some pertinent and important conclusions:

• Intel has pushed their P-cores for MT scenarios to extremes to be the absolute performance king and at least be faster than 5900X and rival 5950X in many cases. This results in an insane power consumption out of the box but only for heavy MT tasks, e.g. video encoding, rendering, software compilation, math calculations - not something average people do daily.
• This extreme power consumption does not translate into every day scenarios like modestly threaded applications or games - in fact many reviewers show that ADL CPUs are the most power efficient in games. Igor's Lab, AnandTech and computerbase.de have shown that limiting their TDP to 125W or even lower does not meaningfully affect frame rates.
• It seems very likely that if P-cores maximum frequency is decreased by just 200-300MHz their efficiency will be incredible.
• Factory OC'ing is not new and NVIDIA, AMD, Apple have been doing that for at least a couple of years. No one is crying foul because of that.

TLDR: Overall ADL CPUs are great sans an extreme factory OC for heavy MT scenarios which can be easily mitigated by limiting their power consumption by setting the PL1 limit in BIOS. At the moment the only issue is the price of the platform because even though the CPUs are competitively priced, you need to purchase a quite expensive motherboard, DDR5 RAM (the faster the better) and a decent cooling solution (preferably AIO).

Too many reviewers are fishing for views and ad revenue, so having loud and disparaging headlines which aren't necessarily representative of the real world is their way of achieving that which is quite sad.

Average people are not buying i9s and i7s (especially K ones) anyway, so your conclusion regarding them makes no sense. Furthermore, the fact that current everyday scenarios and games are not utilizing all the CPU cores to their max and as a consequence power draw does not peak all the time sounds like a lame excuse for a poor MT efficiency of stock i7/i9 K models, even though 125W is "enough" to maintain high FPS. I mean, 240W, 240 f****ing watts for a mainstream desktop CPU sounds a bit ridiculous to me, considering that this is a HEDT power territory and the whole PR around that hybrid AL architecture was kind of better power and surface efficiency.

The actual conclusion is that K models are insanely "brute forced" out of the factory and tailored for those, who care only about absolute performance regardless of the power draw. This was the only way to kill ZEN3 in MT, because ZEN3 MT efficiency is still far superior. Also, non-K and T models will be more reasonable choice most likely: cheaper and more efficient.

[sdack]

You are either too young to remember, or too ignorant to acknowledge AMD's history with regards to x86.

You are still only trying to compare Intel to Monsanto. You never "discussed" it. You only insinuated it.

[blackshard]

Well... probably because M1 has Turbo, Performance Cores, Energy Efficient Cores and a whole lot of specialized hardware acceleration. They are also extremely wide when compared to other designs (8 vs. ADL's 6 or Zen's 4). Second generation 5nm process also helps with fitting 57bn transistors (RTX 3090 has 28,3bn for comparison) in a small package.

Yet looking at more standardized benchmarks like SPEC M1 loses in absolute performance to both ADL and Zen 3. The energy efficiency is extremely good, however:

The Apple M1 has only a BIG.little core design, with 4 BIG cores and 4 little cores. No hyperthread and no turbo gimmicks. Here is the anandtech article where the confrontation against x86 is incredible. I'm not an Apple fan at all, but looking at those benchmarks it looks like x86 days seems over.
And look at the power usage: 25 Watts, 1/10 (one tenth) of the power draw of this Intel crap!

[numacross]

The Apple M1 has only a BIG.little core design, with 4 BIG cores and 4 little cores. No hyperthread and no turbo gimmicks. Here is the anandtech article where the confrontation against x86 is incredible. I'm not an Apple fan at all, but looking at those benchmarks it looks like x86 days seems over.
And look at the power usage: 25 Watts, 1/10 (one tenth) of the power draw of this Intel crap!

The M1 Pro and Max do have Turbo for the big cores. I do not say they are not impressive, far from it, but they have a node advantage over AMD and Intel, because of the second generation 5nm TSMC process (vs. 7nm TSMC for AMD and Intel 7, formerly 10ESF).
As for x86 days being over, a recent interview on AnandTech makes some great points about the relevancy of ISA wars to core design. Basically Zen was designed in tandem with an ARM core and the ISA has far lesser impact on microarchitectual design than people give it credit for.

[bridgman]

<performance chart for SPECint2017_r1 / SPECfp2017_r1>

The Apple M1 has only a BIG.little core design, with 4 BIG cores and 4 little cores. No hyperthread and no turbo gimmicks. Here is the anandtech article where the confrontation against x86 is incredible. I'm not an Apple fan at all, but looking at those benchmarks it looks like x86 days seems over.
And look at the power usage: 25 Watts, 1/10 (one tenth) of the power draw of this Intel crap!

At the risk of asking a dumb question, isn't this a single-thread performance comparison ? How is the number of cores/threads relevant ?

The next page of the article you linked is probably more relevant - multi-core numbers against x86 parts designed to operate in a similar power range:

https://www.anandtech.com/show/16252...le-m1-tested/5

The M1 ends up between a 15W x86 and a 35W x86 (AMD 4800U and 4900HS respectively) although the M1's FP numbers were actually better than the 35W part.

[nikslor]

Speaking of p- vs e-cores, here's in depth coverage from computerbase.de:

https://www.computerbase.de/2021-11/intel-core-i9-12900k-i7-12700k-i5-12600k-test/2/

It's quite dense German text but Google Translate probably can manage it.

deepl.com is the new cool kid on the block regarding translations - give it a try if you have a few minutes; for me it works better in most cases.

[sophisticles]

I don't see the source of excitement. Yes, it is ultimately a good thing that intel finally has something competitive, as amd offerings are diminishing in purchase value due to the increase of their margins... due to lack of competition.

But it is actually a tad sad that intel needs DOUBLE the power and DOUBLE the ram bandwidth to catch up to zen, and in light workloads only, while still falling behind in the heavy and time consuming ones, which I dare say are far more important than some intangible fps improvement.

Intel had to push this sucka well beyond its energy sweetspot, butchering its efficiency in the process, all to come out a bit on top, making the whole thing seem but a pyrrhic victory. But it still counts to all those demented fanbois, they don't care about reason after all.

All in all, amd will most likely handle that with the vcache refresh of ryzen 3, and will still have the ddr5 card to play next year with zen 4. So intel better have a lot more improvements in store, because it will need them. Especially in the far more lucrative enterprise market, where power efficiency matters.

I don't know what you're on about but if you leave Phoronix and look at other benchmarks, you will see that you are very mistaken. Check out Puget Systems' reviews to see what Alder Lake is capable of,

[smitty3268]

The M1 Pro and Max do have Turbo for the big cores.

Yep, depending on the definition of "turbo". They certainly do a lot less than x86 machines do.

If I recall correctly, they can only hit their "max" speed if a single big core is running per 4-core complex.

Once 2 of them are going, it drops by 100 or 200 Mhz, and then drops again when all 4 are busy.

[smitty3268]

At the risk of asking a dumb question, isn't this a single-thread performance comparison ? How is the number of cores/threads relevant ?

Yes, the M1's single-core performance is very impressive but Alder Lake is in the same ballpark and not even Intel's chips require 250W on a single core.

The M1 ends up between a 15W x86 and a 35W x86 (AMD 4800U and 4900HS respectively) although the M1's FP numbers were actually better than the 35W part.

I believe the FP tests are highly bandwidth sensitive, and the M1 is known for having a lot of memory bandwidth and LPDDR5 which probably explains most of that score.

Unfortunately i was never able to find anywhere that did a good job of estimating how much power the M1 actually uses and providing a good comparison against x86 laptop chips. Just about everything was just claiming it was better, without any numbers, or else comparing versus 5 year old Intel chips from previous Mac lineups.

[blackshard]

At the risk of asking a dumb question, isn't this a single-thread performance comparison ? How is the number of cores/threads relevant ?

The next page of the article you linked is probably more relevant - multi-core numbers against x86 parts designed to operate in a similar power range:

https://www.anandtech.com/show/16252...le-m1-tested/5

The M1 ends up between a 15W x86 and a 35W x86 (AMD 4800U and 4900HS respectively) although the M1's FP numbers were actually better than the 35W part.

True. Can't find back the article with the real-world benchmarks. There was a much better placement for the M1 against x86 cpus, in particular the memory bandwidth was pretty impressive. But I understand that without the article these statements can't be validated.