[AT] Intel trying to release 7 nm in 2021

[PotatoWithEarsOnSide]

The 9900K when functioning according to spec is supposed to turbo to 4.7GHz all core, then clock down to 4.2GHz indefinitely (under load) if all thermal headroom has been exhausted. Performing to spec it is outstanding in terms of efficiency, and it in performance it pretty much still beats the competition at gaming, so why the need to go all unscrupulous is beyond me. Sure, it has to beat the 8700/8086K, but that too operates by default outside of spec.
Pushing the process to it's limits is not helping Intel long term.

 

[Topweasel]

It looks like it, doesn't it? See below.

The fact that you want to highlight the 9th gen's departure from the usual 1:1.25 PL1L2 ratio is in fact the answer that seems to be eluding you, and which I've already argued. What number is Intel to aim against when AMD has no fixed power consumption number for PBO? Is it 105w? 150w? 210w? Okay, you tell us.

The 9900k was not released in a vacuum. It had the 8700k and the 2700x to contend with. It handled the 8700k easily due to superior binning, resulting in higher turbo profiles, and higher PL2 watts to support the higher turbos and higher core count - naturally.



I know the idea that Intel went with 210w PL2 due to the fact that a Ryzen 2700x system can consume that much running multithreaded tasks may sound ridiculous. But as has been pointed out, AMD got clever by not exposing XFR2 and PBO consumption numbers as a fixed number. It simply leaves the overclocking up to the system based on thermal headroom. The result is that based on the cooler one opts to go with, these chips can push voltages up to 1.45+v, and it shows. But what AMD doesn't say, can still be measured at the system level and that's important because at the end of the day it's platform against platform. So if you're Intel labs, what do you do about a platform that can pull 210w? You adjust accordingly. You don't pick 210w because the 2700x consumes 105w, you adjust to the levels you know the chip can consume given the right conditions, all based on your own simulations in the lab.

@Topweasel says 210w recommendation is to allow the 9900k to run at 4.7GHz all the time. But that's simply a case of the 9900k running into motherboard maker's TAU=28 Seconds TAU=Unlimited. When properly configured, a 9900k should still turbo to 4.7GHz ACT (All Core Turbo), the only difference is that it'll be for not longer than 28 seconds (iirc), if operating conditions allow it to turbo that long in the first place. In any case, whether the 210w recomended PL2 review setting was to allow the 9900k to turbo all day or not, the aim is the same - to dominate the 2700x, which it does even with PL1 @ 95w.

Tau gets set to unlimited because of turning on another feature Intel recommends which is the I believe its called MCE (Multi-core enhancement).

The issue is in comparison to AMD. AMD if forthcoming about power usage and clock speeds across its rated specs at each of the max settings its under its rated clock speed. It then uses PBO1/2 and XFR1/2 to offer additional performance on top of its rated performance for a small time when needed if using a 105w cooler or longer or infinite if using better cooling. Intel doesn't make that information available. They no longer tell people what the inspec 95w 2-4-6-8 8 core clocks and base turbo are. Partially because anything they try to run under its rated spec is going to give AMD a lot of room to match them. Coffeelake has one great advantage and that is clock speed. 95w neuters that. So again they hide pertinent information, they hide true power requirements (which is probably more of a OEM thing, lots less OEM purchases if they go above x W TDP), so no one knows what they should expect. It will keep the cpu slammed against its thermal limit under load unless someone gets a cooler that is at least X good but no one knows what. This is just going to get worse under Comet Lake.

 

[fleshconsumed]

Wasn't there some research conducted that showed 80% of K CPU owners did not overclock? There's a large bunch of folk out there that will buy the most expensive simply because they assume price = quality; that the product is not targeted at them is a minor irrelevance.

I didn't overclock my K CPUs because I value stability over speed. I just purchased the K chip because they are higher clocked than non-K chips and because they are easier to resell when I'm ready to move on to something else.

 

[Ajay]

Speaking of Apple, the 10 nm fiasco almost certainly is the final straw for Apple to dump Intel for their own processors. It's only matter of time now, and 7 nm isn't going to arrive in time.

Apple's A series arm SoCs just don’t have enough throughput to effectively run x86 emulation. The situation was a bit more favorable in the PPC to Intel transition. The caveat is that this time Apple could design a laptop/desktop A series SoC that includes some custom logic to improve the emulation/translation performance. All IMHO.

 

[moinmoin]

Speaking of Apple, the 10 nm fiasco almost certainly is the final straw for Apple to dump Intel for their own processors. It's only matter of time now, and 7 nm isn't going to arrive in time.

I can only guess Apple's Mac line is not important enough anymore for any more swift actions. But Intel at least makes for a good scapegoat for declining Mac sales.

 

[Dayman1225]

Even Samsung will have left them in the dust by that point.

Based on what Samsung just announced, no not really, but I get what you're saying

 

[TheELF]

Everything. You can't actually tell what kind of power draw to expect from the chip looking at Intel's TDP rating for the 9900k (or the 8700k, though it was somewhat less egregious). You have no idea what kind of cooling to spec for the chip. At least not without reading several reviews first, and figuring out how to configure your board's power limits.

How can you not tell?Intel is telling you right out
"Thermal Design Power (TDP) represents the average power, in watts, the processor dissipates when operating at Base Frequency with all cores active under an Intel-defined, high-complexity workload. Refer to Datasheet for thermal solution requirements. "
so the 9900k will run with 95W all day long at 3.6Ghz 100% as long as you don't run a power virus,it will probably even boost up a bit once in a while.
The cooling you choose will determine the power draw,the CPU will just throttle down if it hits the cooling ceiling.

 

[Markfw]

How can you not tell?Intel is telling you right out
"Thermal Design Power (TDP) represents the average power, in watts, the processor dissipates when operating at Base Frequency with all cores active under an Intel-defined, high-complexity workload. Refer to Datasheet for thermal solution requirements. "
so the 9900k will run with 95W all day long at 3.6Ghz 100% as long as you don't run a power virus,it will probably even boost up a bit once in a while.
The cooling you choose will determine the power draw,the CPU will just throttle down if it hits the cooling ceiling.

So you actually believe that, even after multiple websites have said "its based on the motherboard, and the settings you use, it can be anywhere from 95 watts to 210 watts, NOT OVERCLOCKED." Once overclocked, we have ween it as high as about 500 watts.

If you believe everything that Intel tells you, then I have a bridge I would like to sell you in Brooklyn.

 

[jpiniero]

So you actually believe that, even after multiple websites have said "its based on the motherboard, and the settings you use, it can be anywhere from 95 watts to 210 watts, NOT OVERCLOCKED."

Well, if the cooler can't do the job, it's not going to use 210 watts. But there shouldn't be any problem using a 95W rated cooler, you will just get the base clock of 3.6 instead of the ACT which is 4.7.

 

[PotatoWithEarsOnSide]

Base is 3.6GHz not because that's it's 24/7 95w spec, rather to allow thermal headroom in order to boost to high clocks for short periods; it runs 24/7 at 95w at 4.2GHz, though running that as base would not allow turbo to function at all when combined with a 95w TDP cooler.

 

[DrMrLordX]

Refer to Datasheet for thermal solution requirements.

Is that the same datasheet that shows PL2 being 120W?

edit: the real problem with "it boosts to match the cooling" is that unless you pick exactly a 95W cooler, you'll get more power draw than what you might expect. Put some rinky-dink cooler on there and it throttles. Put something bigger on there and it'll start pulling up to 160W. No way for the end user to know about that behavior without at least a small amount of expert knowledge.

Based on what Samsung just announced, no not really, but I get what you're saying

Eh? I missed that. Please tell me they didn't go GF on us.

 

[Ajay]

Didn’t we have this PL1/PL2 9900K debate already. What is going on??!

 

[DrMrLordX]

Didn’t we have this PL1/PL2 9900K debate already. What is going on??!

Apparently @TheELF missed it, and wants a reenactment.

 

[Dayman1225]

Quite annoying when every Intel thread is derailed by stupid and often repeated arguments that we have all heard before and have little to no relevance to the actual thread.

 

[Zucker2k]

edit: the real problem with "it boosts to match the cooling" is that unless you pick exactly a 95W cooler, you'll get more power draw than what you might expect. Put some rinky-dink cooler on there and it throttles. Put something bigger on there and it'll start pulling up to 160W. No way for the end user to know about that behavior without at least a small amount of expert knowledge.

You might as well have been talking about the 2700x; or any processor for that matter. Processors boost or throttle based on certain variables, among which is cooling headroom or lack thereof.

 

[TheELF]

So you actually believe that, even after multiple websites have said "its based on the motherboard, and the settings you use, it can be anywhere from 95 watts to 210 watts, NOT OVERCLOCKED." Once overclocked, we have ween it as high as about 500 watts.

If you believe everything that Intel tells you, then I have a bridge I would like to sell you in Brooklyn.

Oh I believe there are such MOBOs,I posted a link to an anandtech mobo review that showed 33% difference in power draw with stock settings after all.
They do exist and I do appreciate them exciting because you now what? You can just pop your CPU into the mobo and get the best performance short of manual overclock,no worrying about uefi about aesa about b-die rams no worrying about anything.
If you get into windows and don't like the temps or the fans are too loud or whatever else you can launch the intel extreme tuning tool and throttle to your heart's content and set up automatic profiles for things that you do want to boost.

 

[TheELF]

Is that the same datasheet that shows PL2 being 120W?

edit: the real problem with "it boosts to match the cooling" is that unless you pick exactly a 95W cooler, you'll get more power draw than what you might expect. Put some rinky-dink cooler on there and it throttles. Put something bigger on there and it'll start pulling up to 160W. No way for the end user to know about that behavior without at least a small amount of expert knowledge.



Eh? I missed that. Please tell me they didn't go GF on us.

Without at least a small amount of expert knowledge the user will not even be aware about the CPU using lot's of power,I doubt a lot of people even see or rather notice that 95W TDP anywhere during the whole process of buying the system.

 

[DrMrLordX]

Quite annoying when every Intel thread is derailed by stupid and often repeated arguments that we have all heard before and have little to no relevance to the actual thread.

How did we get on the subject of the 9900k anyway? Everything's getting hazy. All I know is that when people start posting misinformation, that's when the stupid happens.

I do hope that if/when Intel finally releases 7nm CPUs that they won't have this problem. Should be interesting what they can/will release to cope with Zen3/Zen4. 16c minimum, I expect. That doesn't necessarily have to chew up a lot of power on 7nm EUV though. It'll also be interesting to see what (if anything) changes in their core designs based on MDS and other vulnerabilities that apparently still plague their chips. I seriously would like to see them come through with 7nm EUV. The problem is that we have so little to go on right now outside of assurances from a company whose track record with new nodes has been spotty for awhile now. 14nm shaped up nicely, eventually.

You might as well have been talking about the 2700x

The 2700x comes boxed with a cooler that lets it boost normally. It stays within its listed TDP. The end-user has to enable XFR and presumably provide better cooling to get it to start boosting out of its "normal" range. There is no guesswork.

Without at least a small amount of expert knowledge the user will not even be aware about the CPU using lot's of power,I doubt a lot of people even see or rather notice that 95W TDP anywhere during the whole process of buying the system.

Well then. As long as they don't notice it, it's just fine. Case closed!

 

[Topweasel]

Is that the same datasheet that shows PL2 being 120W?

edit: the real problem with "it boosts to match the cooling" is that unless you pick exactly a 95W cooler, you'll get more power draw than what you might expect. Put some rinky-dink cooler on there and it throttles. Put something bigger on there and it'll start pulling up to 160W. No way for the end user to know about that behavior without at least a small amount of expert knowledge.

Eh? I missed that. Please tell me they didn't go GF on us.

This is my biggest issue, its not even about active power draw. While what they are doing is purposely deceptive, even that isn't even a big deal. My problem is that without full knowledge of the product most users are going to be running at the thermal limit of the CPU under-load pretty easily and it will just sit there at that top temp level unless you were prepared and got at least a 120w cooler, but probably a 140w or 160w would be more fitting.

 

[Ajay]

Argh! Please stop and make a new thread on the i9 9900k power limits!!

 

[Dayman1225]

Argh! Please stop and make a new thread on the i9 9900k power limits!!

Agreed. Intel 7nm thread turned into arguing about the 9900K TDP...

 

[DrMrLordX]

Argh! Please stop and make a new thread on the i9 9900k power limits!!

Thread seems to have gone off-track early in the last page.

So, I'll expand on a point I made earlier: Can Intel get back in the fab game if their 7nm EUV node isn't ready until 2022? That's assuming it works? I hadn't considered that, but I ignored the fact that the only 7nm EUV product they plan to launch in 2021 is their enterprise variant of Xe. So my assumption that Intel is officially dead in the fab world if/when they fail to launch desktop and/or server CPUs on 7nm EUV needs revision.

Intel also cryptically has Sapphire Rapids in 2021. Whatever that's gonna be. If that isn't going to be on 7nm, then . . . ?

 

[TheELF]

Thread seems to have gone off-track early in the last page.

So, I'll expand on a point I made earlier: Can Intel get back in the fab game if their 7nm EUV node isn't ready until 2022? That's assuming it works? I hadn't considered that, but I ignored the fact that the only 7nm EUV product they plan to launch in 2021 is their enterprise variant of Xe. So my assumption that Intel is officially dead in the fab world if/when they fail to launch desktop and/or server CPUs on 7nm EUV needs revision.

Intel also cryptically has Sapphire Rapids in 2021. Whatever that's gonna be. If that isn't going to be on 7nm, then . . . ?

Intel is making the most advanced CPU with the highest performance right now,how does that equate into having to get back into the fab game? The only thing someone could argue right now is that they don't have the most power efficient node.
Also what fab game?
Intel has their own fabs to make their own chips they don't need to have any fab game,they are their own customers,they just build a new fab because they needed more production.

Having a smaller node doesn't mean much if it's not performing any better.

 

[Dayman1225]

Thread seems to have gone off-track early in the last page.

So, I'll expand on a point I made earlier: Can Intel get back in the fab game if their 7nm EUV node isn't ready until 2022? That's assuming it works? I hadn't considered that, but I ignored the fact that the only 7nm EUV product they plan to launch in 2021 is their enterprise variant of Xe. So my assumption that Intel is officially dead in the fab world if/when they fail to launch desktop and/or server CPUs on 7nm EUV needs revision.

Intel also cryptically has Sapphire Rapids in 2021. Whatever that's gonna be. If that isn't going to be on 7nm, then . . . ?

Yes because Intel 7nm should still be competitive with TSMC 3nm which currently has no timeline for release, and Samsung “3nm” that was just announced is for 2021 is less dense than TSMCs 5nm but has a cool new GAA device architecture, also Sapphire Rapids is 10++, 2022 for 7nm Server CPUs (Granite Rapids) as per Navin

 

[moinmoin]

Yes because Intel 7nm should still be competitive with TSMC 3nm which currently has no timeline for release, and Samsung “3nm” that was just announced is for 2021 is less dense than TSMCs 5nm but has a cool new GAA device architecture, also Sapphire Rapids is 10++, 2022 for 7nm Server CPUs (Granite Rapids) as per Navin

The huge problem for Intel so far though is they don't do any in-between steps, so every node is completely make or break (more the latter in case of 10nm). This will be very interesting regarding the announcement of making the CPU designs node agnostic, that step doesn't make much sense when the amount of different nodes Intel can actually use stays as low as it is currently.