Intels Skylake-X and Coffee Lake CPUs Discussion

[Life-Binder]

https://www.intel.com/content/www/us.../overview.html


Ice Lake Platform

The Ice Lake processor family is a successor to the 8th generation Intel® Core™ processor family. These processors utilize Intel’s industry-leading 10 nm+ process technology.

[Blueobelisk]

Ayyy are you guys totally jelly about my brand new Intel hardware?



*winks at you and blows you a kiss while you stare in envy*

Lols. I'm surprised intel put their names on these things. The company that were giving these out wasn't even Intel.........

[Evildeffy]

https://videocardz.com/72112/intel-c...ter-than-7700k

Getting closer... seeing 3.7/4.3 still there for the 8700K .. whether that's All-Turbo or SC-Turbo .. we'll see.
But I doubt we're getting an IPC uplift of 15%+ even more than I believe power efficiency increase of 60%+

[Life-Binder]

4.3 has to be 6-core turbo or at least 4-core turbo

theres no way the single core turbo on 14++nm can be a measly 4.3, when even the 14+nm 7980XE has SC 4.4 and 7900X has 4.5


in fact the slide states 11% advantage in single-thread perf over the 4.5 7700K, which would very well fall in line with having a 4.7 single-core turbo on the 8700K - 200 mhz higher than 7700K .. some of the 11% would come from the 200 mhz, some from bigger cache and maybe a small part from IPC arch changes

liking what Im seeing a lot so far, as far as perspective pure gaming performance goes 6 cores with a high boost and an even higher single-core boost to have the least possible bottleneck on the main thread really looks like the gaming sweetspot for now and future

- - - Updated - - -

though I could see the final SC boost being 4.5, not 4.7

but not lower than 4.5

[Evildeffy]

4.3 has to be 6-core turbo or at least 4-core turbo

theres no way the single core turbo on 14++nm can be a measly 4.3, when even the 14+nm 7980XE has SC 4.4 and 7900X has 4.5


in fact the slide states 11% advantage in single-thread perf over the 4.5 7700K, which would very well fall in line with having a 4.7 single-core turbo on the 8700K - 200 mhz higher than 7700K .. some of the 11% would come from the 200 mhz, some from bigger cache and maybe a small part from IPC arch changes

liking what Im seeing a lot so far, as far as perspective pure gaming performance goes 6 cores with a high boost and an even higher single-core boost to have the least possible bottleneck on the main thread really looks like the gaming sweetspot for now and future

- - - Updated - - -

though I could see the final SC boost being 4.5, not 4.7

but not lower than 4.5

The problem with all this is that I still don't believe they advanced the process enough to allow for 50% more cores and be ~60% more power efficient whilst remaining in the 95W TDP range on the same uArch and lithography.

From an engineering PoV this is relatively ... "difficult" to do and I'm being very generous here.

I'm still waiting for final specs and reviews but honestly I'm still expecting they either have 4,3GHz Single Core/Dual Core turbo and 4,0GHz All-Core turbo or that they're straight up lying about the power envelope.

Both are possible.

[Life-Binder]

its probably more then 95W or they got "creative" with the TDP number scenario

fine by me, if the gaming performance is there


but yeah I dont see a SC boost below 4.5 no matter what .. even if the 6-core was 4.0 - they'd still make SC 4.5

[Thunderball]

The problem with all this is that I still don't believe they advanced the process enough to allow for 50% more cores and be ~60% more power efficient whilst remaining in the 95W TDP range on the same uArch and lithography.

From an engineering PoV this is relatively ... "difficult" to do and I'm being very generous here.

I'm still waiting for final specs and reviews but honestly I'm still expecting they either have 4,3GHz Single Core/Dual Core turbo and 4,0GHz All-Core turbo or that they're straight up lying about the power envelope.

Both are possible.

So again, you believe 65W TDP of a 8core but dont believe 95W TDP of a 6core? What?

[Evildeffy]

So again, you believe 65W TDP of a 8core but dont believe 95W TDP of a 6core? What?

So again:
Different uArchs, different lithography process and known variables.

I believe 95W can host an octa-core of Ryzen yes, I do not believe Skylake/Kaby Lake/Coffee Lake can do the same and offer these types of speed no.
I will ask you the question you've dodged in the other thread as well:
Do you believe that Intel, whilst remaining on the same uArch and lithography, has advanced their process enough with "just a refinement" to allow for a previous advancement of ~60% power efficiency?
Or do you truly believe that Samsung/GlobalFoundries' 14nm FinFET LPP (AMD) is the same as Intel's 14nm FinFET SOI along with architecture designs between Ryzen and Skylake/Kaby Lake/Coffee Lake?

[Life-Binder]

7820X can do 4.0 all 8 cores at ~1.0V, ryzens stock cores boost is below 4.0, so that would fit into 95W easily

[Evildeffy]

7820X can do 4.0 all 8 cores at ~1.0V, ryzens stock cores boost is below 4.0, so that would fit into 95W easily

Ah but you come again to the difference mentioned prior.

Does the 7820 @ 4,0GHz all cores stay under the 95W threshhold? Nope.
So knowing Skylake/Kaby Lake and moving up a "refinement" you think you can keep it within 95W even though it is Skylake/Kaby Lake essentially WITH an iGPU?

Ryzen also boosts stock above 4,0GHz if you have the right model for 8/6 cores (1600X and 1800X) etc, but not 8 cores all at the same time.

Like I said ... I call shenanigans, if they hold to the first leaked spec of all core turbo @ 4,0GHz I can potentially see that happening as margins are considerably smaller (even though it'd likely STILL surpass the TDP envelope) and within possibilities then but I don't see if happening with the 2nd leaked specs unless they flat out are incorrect/lying about the TDP.

[Thunderball]

So again:
Different uArchs, different lithography process and known variables.

I believe 95W can host an octa-core of Ryzen yes, I do not believe Skylake/Kaby Lake/Coffee Lake can do the same and offer these types of speed no.
I will ask you the question you've dodged in the other thread as well:
Do you believe that Intel, whilst remaining on the same uArch and lithography, has advanced their process enough with "just a refinement" to allow for a previous advancement of ~60% power efficiency?
Or do you truly believe that Samsung/GlobalFoundries' 14nm FinFET LPP (AMD) is the same as Intel's 14nm FinFET SOI along with architecture designs between Ryzen and Skylake/Kaby Lake/Coffee Lake?

1) Ryzen 7 1700 is 65W TDP. Do you believe that?
2) I've answered, Intel doesnt need to advance their architecture and litography by 60% to rate their 6 cores at 95W TDP.
3) Samsung doesnt have to do anything with Ryzen production, Glofo just got their tech (which is old for Samsung now, 10nm is current for their own mobile SoC and Qualcomm's Snapdragon 835, and we all know that their next step is developing 7nm fab for IBM). GlobalFoundries' 14nm litography is probably inferior to what Intel has right now, yes. I dont see much improvement in power consumption in Zen over Buldozer due to architecture changes, it's mostly litography, so pretty much what Intel needs to do is lower operating voltages without increasing clocks (and maybe tweak the way their boost works, Ryzen does save a lot of power due to that), which seems completely realistic.

[Evildeffy]

1) Ryzen 7 1700 is 65W TDP. Do you believe that?
2) I've answered, Intel doesnt need to advance their architecture and litography by 60% to rate their 6 cores at 95W TDP.
3) Samsung doesnt have to do anything with Ryzen production, Glofo just got their tech (which is old for Samsung now, 10nm is current for their own mobile SoC and Qualcomm's Snapdragon 835, and we all know that their next step is developing 7nm fab for IBM). GlobalFoundries' 14nm litography is probably inferior to what Intel has right now, yes. I dont see much improvement in power consumption in Zen over Buldozer due to architecture changes, it's mostly litography, so pretty much what Intel needs to do is lower operating voltages without increasing clocks (and maybe tweak the way their boost works, Ryzen does save a lot of power due to that), which seems completely realistic.

1: Is there something wrong with your ability to read when I wrote "Different uArchs, different lithography process and known variables"? Stop comparing these, they aren't comparable.
2: Yes it does, considering the speeds, uArch and process they do but no I'm sure that simply applying solder will reduce that, uArch has nothing to do with it(!).
3: Samsung/GloFo made the 14nm process (CPU process =! ARM SoC process) as well as licensing IBM's 7nm process (IBM is already there) and just to be sure here, since you're inferring it in your post above, you're stating that essentially Bulldozer and Ryzen are the same since the only power saving was because of Lithography?
I really hope you absolutely didn't mean to state that at all because if so it shows you having a massive lack of understanding underlying uArch technology.

Lowering operating voltages without increasing clocks (ZOMG! IS THIS THE MYTHICAL BEAST AND DEFINITION OF THE WORD "POWER EFFICIENCY?!?! NO WAI!) and flat out superior turbo boosts to the 7700K is nothing to do with increasing power efficiency?
From what exact Twilight Zone did you pull that out from?

Ryzen doesn't save a lot of power because of their boost technology but from their innate power efficiency and lack of iGPU.
The entire LPP process is built around building efficient silicon ... it doesn't matter if Intel's 14nm tech is more advanced (technically it is) if the efficiency of the CPU uArch is better of the "Lower Tech Lithography" than the competitor.

I will ask you once more:

Or do you truly believe that Samsung/GlobalFoundries' 14nm FinFET LPP (AMD) is the same as Intel's 14nm FinFET SOI along with architecture designs between Ryzen and Skylake/Kaby Lake/Coffee Lake?

Let me ask you one more thing which is very simple:
If the All-core Turbo is indeed 4,3GHz for 6 cores, which is "only" a 4,45% decrease in speed and 50% increase in core count (thus power envelope) vs. the 7700K's 4,5GHz on the exact same process of Kaby Lake would mean the CPU having a TDP of 130(,43)W.
Considering that's 35W over the 95W TDP rating or in %-ages: 37,29% above the rated TDP.

I'm not even considering the part about the die size increasing considerably as well and therefore considerable more power consumption tied with that as well as iGPU improvements, nor the higher speed turbo bins for lesser cores nor the fact that turbo-ing to a higher frequency on a lot of cores exponentially increases power draw and temperatures and therefore the ~60% power efficiency number.

How do you propose Intel does that whilst staying on the same uArch and Lithography?

Soldering alone will not even remotely cut it as temperatures alone do not dictate TDP.

Now ... if they base it off of the "Base Clock" speed being 3,7GHz instead of the "All Core Turbo" 4,3GHz they can do it.
But it'd be straight up lying because if "Base Clock" is 3,7GHz and "All Core Turbo" is 4,3GHz ... what exactly is the base clock for people?

The 8700K/8600K will not have a stock cooler and therefore will have people with an aftermarket cooler so we can rule out anemic coolers and we can easily assume people will have sufficient cooling for 4,3GHz Turbo right?
Which means that if people cannot reach 4,3GHz "All Core Turbo" it means they have insufficient cooling with aftermarket cooling and then switches to 3,7GHz Base Clock meaning once more that the 95W TDP is a load of crock since almost any aftermarket cooler in this class will be capable of cooling a 95W TDP CPU.

Simply put if they base it off the 3,7GHz Base Clock and the All Core Turbo is indeed 4,3GHz (I am hoping for this to be the case TBH) then it's a straight up "lie" as stock configuration of any board and BIOS should run 4,3GHz and not 3,7GHz.

[chronia]

What test to measure TDP are we looking at anyway. I mean load caused by gaming causes a different powerdraw then for example Prime 95 FFT's. If you look at for example Tom's Ryzen 1700 review http://www.tomshardware.co.uk/amd-ry...w-33854-8.html you can see that in all tests the cpus easily stay under their advertised TDP, except for Prime 95 FFT's where a far as i can see they all break their advertised TDP.

[Evildeffy]

What test to measure TDP are we looking at anyway. I mean load caused by gaming causes a different powerdraw then for example Prime 95 FFT's. If you look at for example Tom's Ryzen 1700 review http://www.tomshardware.co.uk/amd-ry...w-33854-8.html you can see that in all tests the cpus easily stay under their advertised TDP, except for Prime 95 FFT's where a far as i can see they all break their advertised TDP.

TDP is generally measured on full load but not on Power Viruses (Prime95 is a "power virus", especially when AVX gets involved).
I believe most use stuff like CineBench, Handbrake or Blender... things that can utilize the CPU fully without breaking the boundaries of torture testing.
Kinda like FurMark is a power virus for GPUs ... not really a load that's typical but stressing.

[Remilia]

Going by this http://www.anandtech.com/show/2807/2
Intel measures their TDP by average power + offset of consumer benchmarks (SPEC).
AMD measures TDP by maximum power that are thermally significant (basically not a tiny power spike) on commercially viable loads.

Granted this was 8 years ago, no idea if anything has changed.

[Thunderball]

1: Is there something wrong with your ability to read when I wrote "Different uArchs, different lithography process and known variables"? Stop comparing these, they aren't comparable.
2: Yes it does, considering the speeds, uArch and process they do but no I'm sure that simply applying solder will reduce that, uArch has nothing to do with it(!).
3: Samsung/GloFo made the 14nm process (CPU process =! ARM SoC process) as well as licensing IBM's 7nm process (IBM is already there) and just to be sure here, since you're inferring it in your post above, you're stating that essentially Bulldozer and Ryzen are the same since the only power saving was because of Lithography?
I really hope you absolutely didn't mean to state that at all because if so it shows you having a massive lack of understanding underlying uArch technology.

Lowering operating voltages without increasing clocks (ZOMG! IS THIS THE MYTHICAL BEAST AND DEFINITION OF THE WORD "POWER EFFICIENCY?!?! NO WAI!) and flat out superior turbo boosts to the 7700K is nothing to do with increasing power efficiency?
From what exact Twilight Zone did you pull that out from?

Ryzen doesn't save a lot of power because of their boost technology but from their innate power efficiency and lack of iGPU.
The entire LPP process is built around building efficient silicon ... it doesn't matter if Intel's 14nm tech is more advanced (technically it is) if the efficiency of the CPU uArch is better of the "Lower Tech Lithography" than the competitor.

1) Of course it's comparable, you just need to factor in the variables, and you're comparing them right now, without even considering those variables.
3) Samsung didnt develop different 14nm FinFET litography iterations for SoC and CPU (memory, NAND, etc), noone does that, of course it's the same thing. Yes, I'm saying that Ryzen and Bulldozer are very close in terms of power saving measures on architectural level (the slides are there, you can actually see that, there are some of course, but they wouldnt turn a most hot chip into something we see today with Ryzen), Ryzen of course got a big boost out of tremendously better software control and better litography, that's pretty clear.

Also, uArch is not a technology. It's a hardware development method no top3 company is using right now, so stop using it.

No, superior boosts clocks of 7700K have nothing to do with power efficiency, they have anything to do with chip ability to scale with higher voltages (at the cost of higher power consumption).

iGPUs doesnt affect power efficiency when not used, as they are powered separately and are simply not powered if disabled, more over, they can actually help heat dissipation by increasing die surface area.

Let me ask you one more thing which is very simple:
If the All-core Turbo is indeed 4,3GHz for 6 cores, which is "only" a 4,45% decrease in speed and 50% increase in core count (thus power envelope) vs. the 7700K's 4,5GHz on the exact same process of Kaby Lake would mean the CPU having a TDP of 130(,43)W.
Considering that's 35W over the 95W TDP rating or in %-ages: 37,29% above the rated TDP.

There is no point in addressing that as TDP rating is does not assume that all cores are boosting.

Now ... if they base it off of the "Base Clock" speed being 3,7GHz instead of the "All Core Turbo" 4,3GHz they can do it.
But it'd be straight up lying because if "Base Clock" is 3,7GHz and "All Core Turbo" is 4,3GHz ... what exactly is the base clock for people?

It's not lying, it's TDP, you just dont understand how it works, and that it's actually different from manufacturer to manufacturer.

- - - Updated - - -

TDP is generally measured on full load but not on Power Viruses (Prime95 is a "power virus", especially when AVX gets involved).
I believe most use stuff like CineBench, Handbrake or Blender... things that can utilize the CPU fully without breaking the boundaries of torture testing.
Kinda like FurMark is a power virus for GPUs ... not really a load that's typical but stressing.

TDP is never measured at full load. TDP is TYPICAL load.

[mrgreenthump]

TDP is never measured at full load. TDP is TYPICAL load.

TDP should be average heat generated by the chip at full load. Companies just tend to have their own definitions for whatever reason, making the term next to useless.

[Evildeffy]

1) Of course it's comparable, you just need to factor in the variables, and you're comparing them right now, without even considering those variables.

Wow... so you want to compare TDP of entirely different processes and brands and say they can be compared, I don't know what to tell you.

3) Samsung didnt develop different 14nm FinFET litography iterations for SoC and CPU (memory, NAND, etc), noone does that, of course it's the same thing. Yes, I'm saying that Ryzen and Bulldozer are very close in terms of power saving measures on architectural level (the slides are there, you can actually see that, there are some of course, but they wouldnt turn a most hot chip into something we see today with Ryzen), Ryzen of course got a big boost out of tremendously better software control and better litography, that's pretty clear.

Really so Samsung's ARM chips are the same as x86 CPUs in terms of development? Right ok.. Barring the fact that Samsung's 14nm LPP FinFET technology had to be adapted by GloFo (together with Samsung) to fit a different type of architecture they are identical Right? I'll leave it at that since going further is useless due to your lack of understanding.
You seriously cannot be stating that Ryzen is the same as Bulldozer with better lithography and software control, I REALLY hope you're not stating that.
As far as power saving is concerned .... C states and general power saving measures are used in a lot of processors, however comparing and stating Bulldozer and Ryzen are the same is ... I have no words for it really.

Also, uArch is not a technology. It's a hardware development method no top3 company is using right now, so stop using it.

What? Do you even know what uArch means/what it is? What is Skylake, Kaby Lake, Coffee Lake, Bulldozer and Ryzen?
This very statement is shows just how dangerously lacking your knowledge is.

No, superior boosts clocks of 7700K have nothing to do with power efficiency, they have anything to do with chip ability to scale with higher voltages (at the cost of higher power consumption).

*twitch* .. someone please explain this to the man before my brain short circuits.

iGPUs doesnt affect power efficiency when not used, as they are powered separately and are simply not powered if disabled, more over, they can actually help heat dissipation by increasing die surface area.

O'rly ... funny how that is so wrong considering the fact that the reason base clock overclocking is "impossible" nowadays is because of the still powered GPU if not used on CPUs.
They most assuredly do, regardless of use or not.

There is no point in addressing that as TDP rating is does not assume that all cores are boosting.

And here we go... the dodging begins.

It's not lying, it's TDP, you just dont understand how it works, and that it's actually different from manufacturer to manufacturer.

Nope, I have a very firm grasp of how TDP works and I hold every bit of tech to those standards, if you prefer to allow brands to "cheat" saying that Base Clock is what should be counted even though that will never be used to make brand A, B or C look better ... well then ... we should base TDP on (your favourite word) "Typical Load" for all users as an average which is idling on desktop... Look my Core i7-990X is now a 30W TDP CPU! ZOMG.

Tell me ... how is stock operation of a CPU dictated by Intel with "Turbo'ed" cores to be @ 4,3GHz not the reference point of TDP where the only way to reach base clock is having to underclock your CPU or have severely insufficient cooling, showing other issues?

TDP is never measured at full load. TDP is TYPICAL load.

So 20 - 30W ... got it... Typical Load is dependant upon what is used as a programme, if fully utilized it is "Typical Load" as well otherwise 99,99% of all CPUs on this planet will be rated at desktop idle loads.
TDP is rated at typical loads of programmes used to offer load unto the CPU and what the CPU needs in terms of cooling ability to dissipate.

Still funny how you dodged several questions and statements... picking out what you think you can respond to.

But if anything this latest response of yours clearly shows how much you're either trolling or know (better stated: do now know) of this technology.

[Thunderball]

Wow... so you want to compare TDP of entirely different processes and brands and say they can be compared, I don't know what to tell you.

Of course they can. It's not like the processors are fundamentally different, they are just solving some common problems with different means.

Really so Samsung's ARM chips are the same as x86 CPUs in terms of development? Right ok.. Barring the fact that Samsung's 14nm LPP FinFET technology had to be adapted by GloFo (together with Samsung) to fit a different type of architecture they are identical Right? I'll leave it at that since going further is useless due to your lack of understanding.
You seriously cannot be stating that Ryzen is the same as Bulldozer with better lithography and software control, I REALLY hope you're not stating that.
As far as power saving is concerned .... C states and general power saving measures are used in a lot of processors, however comparing and stating Bulldozer and Ryzen are the same is ... I have no words for it really.

Litography technology is identical (unless Samsung improved something after implementing their technology at GloFo, which is entirely possible). Manufacturing process of course is different (pretty sure even between Samsung and GloFo) but that shouldnt really mean any measurable differences. I never said Ryzen and Bulldozer are identical in terms of power saving, start reading already: those are very close in terms of architectural features that contribute to power savings.

What? Do you even know what uArch means/what it is? What is Skylake, Kaby Lake, Coffee Lake, Bulldozer and Ryzen? This very statement is shows just how dangerously lacking your knowledge is.

Skylake is the same uarch as Kaby Lake (although CPU core assembly part only, evolution of iGPU part is significant), Bulldozer and Ryzen are different (although are very close), Coffee Lake we have no idea (2 and 4 core parts probably use the same uarch as Skylake/Kaby Lake). The time of sweeping architecture changes is long gone. Most pieces of hardware created for a similar purpose can be successfully emulated to resemble each other through software means without much performance loss.

O'rly ... funny how that is so wrong considering the fact that the reason base clock overclocking is "impossible" nowadays is because of the still powered GPU if not used on CPUs.
They most assuredly do, regardless of use or not.

You could overclock locked Skylake processors with a special BIOS (and initially with an official BIOS) through BCLK, Intel just locked it after through microcode change, and on OS level starting with Kaby Lake.

Nope, I have a very firm grasp of how TDP works and I hold every bit of tech to those standards, if you prefer to allow brands to "cheat" saying that Base Clock is what should be counted even though that will never be used to make brand A, B or C look better ... well then ... we should base TDP on (your favourite word) "Typical Load" for all users as an average which is idling on desktop... Look my Core i7-990X is now a 30W TDP CPU! ZOMG.

Tell me ... how is stock operation of a CPU dictated by Intel with "Turbo'ed" cores to be @ 4,3GHz not the reference point of TDP where the only way to reach base clock is having to underclock your CPU or have severely insufficient cooling, showing other issues?

Brands cheat all the time because they themselves design TDP rating methodology and standards. It applies to all brands. I'm not saying it's good, I'm just saying that TDP numbers are arbitrary, and cannot be used to extrapolate CPU specs.

But if anything this latest response of yours clearly shows how much you're either trolling or know (better stated: do now know) of this technology.

Ok if I dont know of this "technology" (not a technology by the way) then enlighten me. We have a R7 1700 and a R7 1700X which are physically identical, and are within single digits of each other in heat dissipation under full load. One has 65W TDP, second one has 95W TDP.

[Evildeffy]

Of course they can. It's not like the processors are fundamentally different, they are just solving some common problems with different means.

Any electrical engineer would turn around in his grave reading this statement.

Litography technology is identical (unless Samsung improved something after implementing their technology at GloFo, which is entirely possible). Manufacturing process of course is different (pretty sure even between Samsung and GloFo) but that shouldnt really mean any measurable differences. I never said Ryzen and Bulldozer are identical in terms of power saving, start reading already: those are very close in terms of architectural features that contribute to power savings.

No, now you're altering things, you specifically stated that Ryzen and Bulldozer are very close to each other and that software control and better lithography attributed to it.
The uArch itself, and it is VASTLY different from Bulldozer to Ryzen, is the reason for power efficiency as well and stating that not to be the case is ... just WTF.

Skylake is the same uarch as Kaby Lake (although CPU core assembly part only, evolution of iGPU part is significant), Bulldozer and Ryzen are different (although are very close), Coffee Lake we have no idea (2 and 4 core parts probably use the same uarch as Skylake/Kaby Lake). The time of sweeping architecture changes is long gone. Most pieces of hardware created for a similar purpose can be successfully emulated to resemble each other through software means without much performance loss.

What happened to the "Also, uArch is not a technology. It's a hardware development method no top3 company is using right now" bit? ... You have not had a sweeping change from Skylake -> Kaby Lake and will not with Coffee Lake as it is yet again a refinement of Kaby Lake.
But comparing Bulldozer and Ryzen and saying they aren't drastically different ... I think with that statement every single Electrical Engineer that'd read that would lose IQ/Brain cells reading that.

That they are both based upon the x86 instruction set doesn't mean they can't go about it different ways.

The Internal Combustion Engine has the same principles and yet can drive on a multitude of different fuels provided they have a different design.
Diesel vs. Benzine (Petrol if you're British/US) vs. Water vs. Natural Oils.
Same principle ... entirely different execution.

You could overclock locked Skylake processors with a special BIOS (and initially with an official BIOS) through BCLK, Intel just locked it after through microcode change, and on OS level starting with Kaby Lake.

And how did they manage to do that? That's right they managed to separate the iGPU control entirely from the CPU.
That, which is impossible now, is why they could and Intel having fixed the "bug" by locking iGPU control with the rest of the CPU again means you'll never see this again, meaning ... in EVERY definition ... iGPU is most definitely present in CPU Power Efficiency and there's no way to dodge this.
Also Kaby Lake's "Intel Management Engine" is not OS based, it is most assuredly based in the firmware of the board, this isn't part of the OS otherwise it'd be 100% circumvented already.

Brands cheat all the time because they themselves design TDP rating methodology and standards. It applies to all brands. I'm not saying it's good, I'm just saying that TDP numbers are arbitrary, and cannot be used to extrapolate CPU specs.

Across different uArch and lithography I would agree with you.
Staying on the same uArch and Lithography it's process ... it's actually VERY viable to extrapolate some things such as power draw/envelope.

Ok if I dont know of this "technology" (not a technology by the way) then enlighten me. We have a R7 1700 and a R7 1700X which are physically identical, and are within single digits of each other in heat dissipation under full load. One has 65W TDP, second one has 95W TDP.

No 2 CPUs are identical, you have higher quality silicon (natural part of the process), speed and operating voltages at said speed.
1 may be extremely efficient requiring lower voltages than the other, center silicon wafer, and the other may come from the edges.
1 may be clocked considerably higher and require more operating voltage than the other.
You may have 1 which is inefficient and requires more operating voltage as well as being clocked higher with speeds, which is what AMD bases it's TDP upon the maximum voltage x electrical current required for operation and cooling to associate with that.
There's a plethora of reasons of why and how but let's get to that shall we as with your very example.

Weren't you the one who stated the following?

AMD does it, why wouldnt Intel do it aswell? Intel could absolutely fit a 6core into 100W TDP range with 7700K-like clocks by slapping some solder under the lid, but I guess it's not their way.

So what is it now then as you first say it is based upon heat dissipation and now not.

Even though this is a power virus so consumption goes beyond stated TDP as ordinary software cannot generate the same thing:
Click me for a picture!
Or are you saying these readings do not properly convey the differences since you don't believe there's a difference?
Also check out the 7700K there ... higher power draw than a 95W TDP CPU whilst itself is 91W ... strange ... the difference mustn't be in the uArch or anything, it must be because it's got no solder under the hood.

Do you think there would be a meaningful change to power draw and cooling requirements if you were to repeat the same test with a soldered/delidded CPU?

Also still dodging... yes you are very much lacking.