Intel's Future Chips: News, Rumours & Reviews

[gamerk316]

If true, a very interesting turn of events. There was speculation by several people that Intel should go fabless, so it can use the most competitive process available.

I read this another way; Intel is desperately trying to free up space on it's current node to meet demand for it's higher performing chips, and the easiest way to do that (short of getting 10nm working) is to outsource production of it's lower tier chips to free up fab space.

 

[jimmysmitty]

If true, a very interesting turn of events. There was speculation by several people that Intel should go fabless, so it can use the most competitive process available.

I read this another way; Intel is desperately trying to free up space on it's current node to meet demand for it's higher performing chips, and the easiest way to do that (short of getting 10nm working) is to outsource production of it's lower tier chips to free up fab space.

Lower tier like the U series or even just chipsets and NIC chips in general. Their chipsets have generally not been problematic so using a different FAB wouldn't be a bad idea to free up space for higher margin parts.

That said, TSMC might get a bit overloaded with everything they have. AMD and nVidia GPUs, AMD CPUs and whatever else comes their way. Might end up in a bidding war for priority.

 

[gamerk316]

If true, a very interesting turn of events. There was speculation by several people that Intel should go fabless, so it can use the most competitive process available.

I read this another way; Intel is desperately trying to free up space on it's current node to meet demand for it's higher performing chips, and the easiest way to do that (short of getting 10nm working) is to outsource production of it's lower tier chips to free up fab space.

Lower tier like the U series or even just chipsets and NIC chips in general. Their chipsets have generally not been problematic so using a different FAB wouldn't be a bad idea to free up space for higher margin parts.

That said, TSMC might get a bit overloaded with everything they have. AMD and nVidia GPUs, AMD CPUs and whatever else comes their way. Might end up in a bidding war for priority.

Boy, if only there were some OTHER foundry that could bid for this type of thing. Maybe one located in upstate NY?

...nah.

 

[goldstone77]

Boy, if only there were some OTHER foundry that could bid for this type of thing. Maybe one located in upstate NY?

...nah.

Ha ha ha! That tickled me!

 

[goldstone77]

Yeah, I agree they are trying to make space so they can meet data center demands. I also, think this could be a stepping stone as well. Going fabless doesn't mean Intel has to stop all manufacturing. It can keep on spinning out products, but also have a products being produced with the most cutting edge process as well.

 

[audiospecaccts]

I think they should make them lid-less considering the thermal hysteresis issues that lead people to de-lid the processors.

 

[jimmysmitty]

I think they should make them lid-less considering the thermal hysteresis issues that lead people to de-lid the processors.

And risk the die to be cracked by someone who puts too much pressure on the cooler? Sure. Have a whole new set of problems.

 

[8350rocks]

For all we know they could have a viable working 7nm process they are going to ramp up quickly after they push their 10nm out the door.

Jimmy, correct me if I'm wrong, but it doesn't appear that you understand the lithography process, and the limitations of current tooling. It's impossible to accurately make the features for what Intel is calling 7nm at this time. So, for all we know, they absolutely do not have a viable 7nm working. And without EUV it's a extremely hard task to make a single 10nm chip at 100.76 MTr/mm² without it being riddled with defects. The tools just are not capable of measuring that small accurately. So, I'll go back to what I originally been trying to say, Intel has been lying about 10nm for a very very long time!

Edit: Just to add to the tooling, the new tools are expected in 2021, and they are 2 stories tall! It is going to be a monumental feat into even integrate those tools to a clean room environment.
2nd Edit:https://twitter.com/lasserith/status/1042750214757646336
https://thechipcollective.com/posts/moshedolejsi/what-is-up-with-euv/

Again not going to argue. We don't know what Intel has or don't have.

Also, they have been working on FAB42 quite a bit recently, it is currently being retooled for 7nm, and they have multiple large cranes. I could take a few pictures as I pass by if you would like.

As for viable, they could have a viable 7nm process. I didn't say something they could ramp up right away Ias they do have tools that they would have used in the research phase for small scale testing.

Samsung as well announced that they have a EUV "7nm" process and are beginning to mass manufacture it. So there are tools just not the ones required for what Intel would probably want to push density wise. Still doesn't mean they might not have something similar to Samsungs that is viable.

PS while I don't know the innate details of lithography I do understand it. I know there are tools required to be developed. At the same time they have to do research to develop the tools and process for use with those tools. From my understanding they typically do small scale tests trying to get the results desired then retool the FABs to suit it. My assumption is that they did research in small scale and now are retooling FAB42 for 7nm, be it EUV or a different design, as now 5nm is in research.

So, let me step in here and somewhat explain how lithography works so that we are on the same page.

Without EUV this is how lithography works:

The smallest design you can use with standard lithography is using a light that is 200nm in wavelength. Yes, that is correct.

The way we have managed to overcome that is by using multiple patterns. Essentially, how that works is that you lay down your first 200nm wavelength pattern, and then move it 100nm in the direction you want to pattern, and now you eliminate the outer layers through a process of revision and the inner lines of your double pattern are now 100nm. To get to sizes like 10nm, you are talking about patterning stages that may require something on the order of 20 or more revisions due to the required precision needed to ensure you do not have any faulty connections. The fact that we have been able to pull this off at all is an outstanding credit to the ingenuity of the engineers who pulled it off.

EUV has a wavelength of 13.5 nm, and I am pretty sure at this point I do not have to explain why 10nm or smaller is basically going to benefit immensely from EUV lithography. Instead of 20+ revisions, you are now talking about stages back down to 1-2 revisions like 90nm was years ago.

Do keep in mind that this is a *very* limited explanation in layman's terms...if you want more detail, I can discuss it in greater detail; however, I felt an ELI5 version was probably most appropriate here.

 

[audiospecaccts]

I think they should make them lid-less considering the thermal hysteresis issues that lead people to de-lid the processors.

And risk the die to be cracked by someone who puts too much pressure on the cooler? Sure. Have a whole new set of problems.

I never had that issue with the naked chip die type processors of the past.

 

[jimmysmitty]

I think they should make them lid-less considering the thermal hysteresis issues that lead people to de-lid the processors.

And risk the die to be cracked by someone who puts too much pressure on the cooler? Sure. Have a whole new set of problems.

I never had that issue with the naked chip die type processors of the past.

You may not have. Doesn't mean others have not. Then there is the issue of TIM. People can use the wrong type that has conductive material that can short a CPU out or cause damage etc. It may not be the majority of people but its enough that even AMD went the route of an IHS with their K8 series CPUs and have not looked back.

 

[goldstone77]

Why Intel Processors Draw More Power Than Expected: TDP and Turbo Explained
by Ian Cutress on November 9, 2018 12:00 PM EST
https://www.anandtech.com/show/13544/why-intel-processors-draw-more-power-than-expected-tdp-turbo

 

[volkgren]

Why Intel Processors Draw More Power Than Expected: TDP and Turbo Explained
by Ian Cutress on November 9, 2018 12:00 PM EST
https://www.anandtech.com/show/13544/why-intel-processors-draw-more-power-than-expected-tdp-turbo

I think Intel should go back to having the K-series with lower frequencies and leaving the overclocking to the enthusiasts. They've gone down this road of having higher and higher frequencies, and then having to out-perform themselves with every new CPU release.

 

[Supernova1138]

Why Intel Processors Draw More Power Than Expected: TDP and Turbo Explained
by Ian Cutress on November 9, 2018 12:00 PM EST
https://www.anandtech.com/show/13544/why-intel-processors-draw-more-power-than-expected-tdp-turbo

I think Intel should go back to having the K-series with lower frequencies and leaving the overclocking to the enthusiasts. They've gone down this road of having higher and higher frequencies, and then having to out-perform themselves with every new CPU release.

Intel's problem is that because they can't get 10nm working, all they can do is ramp up the clockspeed, throw on more cores and hope the chip doesn't melt in order to get more performance. With the 9900k, we are effectively at the limit of 14nm, and that's a chip that despite the official 95 Watt TDP, tends to draw 150+ Watts under full load unless you specifically limit it to 95 Watts in the BIOS on most boards and take a sizeable clockspeed and performance hit.

Can't wait for the 10th gen Core series processors that requires liquid nitrogen cooling to run at stock speeds.

 

[volkgren]

Right, they should have said it was 150W TDP or, like I was saying, just clocked it at 4.2Ghz or something and left the rest up to any enthusiasts looking to push it that far.

Will 10nm (or any other future improvements) lead to considerably better IPC that could result in lower frequencies (such as 3.6GHz) being that much better so we can stop trying to go higher and higher with frequencies?

 

[Supernova1138]

Right, they should have said it was 150W TDP or, like I was saying, just clocked it at 4.2Ghz or something and left the rest up to any enthusiasts looking to push it that far.

Will 10nm (or any other future improvements) lead to considerably better IPC that could result in lower frequencies (such as 3.6GHz) being that much better so we can stop trying to go higher and higher with frequencies?

Well Intel didn't want to leave it at 4.2GHz on all cores, because then the 9900k wouldn't be much better than the R7 2700x out of the box and they couldn't charge the huge premium for the 9900k.

As for shrinking down to 10nm, there may be some architectural improvements but mostly it's improving power efficiency so they can cram more cores onto the chip without having to resort to a 150+ Watt TDP. AMD has basically kicked off a core count war, since we have hit the limit in terms of clockspeed and even IPC improvements are likely to be limited going forward, so all that can be done for more performance is throw more cores onto the CPU and hope software will be written to utilize them.

 

[Isaiah4110]

Putting this out there again for everyone: I know it has been explained many times before (I think in this very thread), but TDP does NOT equal power draw.

Power draw is how many watts of power (volts times amps) the CPU itself is 'using' (and needs supplied from the wall just for the CPU through the PSU), counting all the power actually used to do work AND the power escaping the processor in the form of heat generated by inefficiencies.

TDP (Thermal Design Power) is the maximum amount of HEAT ENERGY the CPU will generate through normal operation and is used to identify what the COOLING REQUIREMENTS are for that CPU.


TDP and power draw are typically directly related (meaning as power draw rises TDP usually does as well at a similar rate), but they are NOT equal. It is perfectly acceptable (and actually NORMAL IIRC) for a CPU to have a TDP lower than its power draw. Since TDP comes from heat generated by the CPU, which is created by inefficiencies (electrical resistance, etc. turning power drawn into heat)... yeah, they shouldn't be equal.


I'm pretty sure this is one of the most commonly misunderstood CPU specifications. If you want more detail (I'm not going to bother typing more) surrounding the science (primarily physics) of TDP and power draw then feel free to read the Wikipedia entry on Thermal Design Power: https://en.wikipedia.org/wiki/Thermal_design_power

 

[volkgren]

Putting this out there again for everyone: I know it has been explained many times before (I think in this very thread), but TDP does NOT equal power draw.

Power draw is how many watts of power (volts times amps) the CPU itself is 'using' (and needs supplied from the wall just for the CPU through the PSU), counting all the power actually used to do work AND the power escaping the processor in the form of heat generated by inefficiencies.

TDP (Thermal Design Power) is the maximum amount of HEAT ENERGY the CPU will generate through normal operation and is used to identify what the COOLING REQUIREMENTS are for that CPU.


TDP and power draw are typically directly related (meaning as power draw rises TDP usually does as well at a similar rate), but they are NOT equal. It is perfectly acceptable (and actually NORMAL IIRC) for a CPU to have a TDP lower than its power draw. Since TDP comes from heat generated by the CPU, which is created by inefficiencies (electrical resistance, etc. turning power drawn into heat)... yeah, they shouldn't be equal.


I'm pretty sure this is one of the most commonly misunderstood CPU specifications. If you want more detail (I'm not going to bother typing more) surrounding the science (primarily physics) of TDP and power draw then feel free to read the Wikipedia entry on Thermal Design Power: https://en.wikipedia.org/wiki/Thermal_design_power

Yes, I am aware of that. I'm sure others are as well. I admit it usually get's forgotten.

Nonetheless, ~150W power draw does not equal 95W TDP.

 

[goldstone77]

What is TDP (Thermal Design Power)?
With every processor, Intel guarantees a specific frequency at a specific power, often with a particular grade of cooler in mind. Most people equate a chip's TDP rating directly to its maximum power draw, given that the heat energy that needs to be dissipated from the processor is equal to the power consumed in doing calculations. Normally, the TDP rating is that specific power.

But TDP, in its strictest sense, relates to the ability of the cooler to dissipate heat. TDP is the minimum capacity of the CPU cooler required to get that guaranteed level of performance. Some energy dissipation also occurs through the socket and motherboard, which means that technically the cooler rating can be lower than the TDP, but in most circles TDP and power consumption are used to mean the same thing: how much power a CPU draws under load.

For any given processor, Intel will guarantee both a rated frequency to run at (known as the base frequency) for a given power, which is the rated TDP. This means that a processor like the 65W Core i7-8700, which has a base frequency of 3.2 GHz and a turbo of 4.7 GHz, is only guaranteed to be at or below 65W when the processor is running at 3.2 GHz. Intel does not guarantee any level of performance above this 3.2 GHz / 65W value.

https://www.anandtech.com/show/13544/why-intel-processors-draw-more-power-than-expected-tdp-turbo?_ga=2.254249096.1462579559.1541509107-1754874025.1532872917

Guess how much power is being pulled by that 65WTDP Intel processor at base clocks and no turbo. 65W. Reviews have been using TDP as a reference for many years now. Guess how many watts are being pulled by it during turbo? 65x1.25=81W. This is exceeding it's 65W TDP!

PL1 = TDP = 95 W
PL2 = TDP * 1.25 = 118.75 W
Tau = 8 seconds

While TDP helps create cooling solutions, it remains very much attached to power consumption. I don't know the exact amount of heat that is dissipated through inefficiencies, and into the socket/motherboard, but we do know that coolers are made with TDP ratings and most seem up to the task of their Rating.

Ian was discussing how motherboard manufactures have modified Intel spec to achieve longer or unlimited Tau so those processors are always running at boost speeds, which is exceeding TDP.

Edit:
Do We Need to Re-Review the Core i9-9900K? Please Help!
Hardware Unboxed
Published on Nov 8, 2018
https://youtu.be/XfGz22ZjeGk?t=309

2nd Edit:

https://www.techspot.com/news/77313-do-need-re-review-core-i9-9900k.html

 

[-Fran-]

In all fairness, that is the desired behaviour by consumers anyway... I know it is for mine.

There's little situations where you'll be thermally constrained, outside of a silent HTPC, but then again you wouldn't use something with a big power budget in that scenario anyway.

As usual, it's important to know and let other people understand what those numbers mean. That is where Intel dropped the ball: telling average Joe the CPUs WILL suck more juice than advertised and they need to get a good PSU AND cooler for it if they want the reviews numbers.

Now, the interesting thing is comparing it to AMD's implementation of Turbo and the advertised numbers. Intel is pushing the thermal limits now and that is evident for everyone (I hope), so their power consumption will go up the roof; AMD has been doing that for a good while now, but not going too overboard.

I wonder if that has to be taken into reviews as well? It would be darn interesting to see how the "top" and "average" cooling solutions fare for the i9 9900K! I know for the 2700X is a grim picture, since this puppy is really power hungry, so the heat it needs to move is a lot.

Cheers!

 

[volkgren]

Yeah, if I had to have some issue with a CPU (for me) it would be that is uses more power than necessary to run faster for longer.

 

[jimmysmitty]

Why Intel Processors Draw More Power Than Expected: TDP and Turbo Explained
by Ian Cutress on November 9, 2018 12:00 PM EST
https://www.anandtech.com/show/13544/why-intel-processors-draw-more-power-than-expected-tdp-turbo

I think Intel should go back to having the K-series with lower frequencies and leaving the overclocking to the enthusiasts. They've gone down this road of having higher and higher frequencies, and then having to out-perform themselves with every new CPU release.

Intel's problem is that because they can't get 10nm working, all they can do is ramp up the clockspeed, throw on more cores and hope the chip doesn't melt in order to get more performance. With the 9900k, we are effectively at the limit of 14nm, and that's a chip that despite the official 95 Watt TDP, tends to draw 150+ Watts under full load unless you specifically limit it to 95 Watts in the BIOS on most boards and take a sizeable clockspeed and performance hit.

Can't wait for the 10th gen Core series processors that requires liquid nitrogen cooling to run at stock speeds.

I am not sure it is the limit of 14nm. Intel has had an 8 core, be it on a lower clock speed, for a while on 14nm but even so the 7820X runs cooler and uses less power than the 9900K even with more memory lanes, more cache, more PCIe lanes and a much faster interface even around the same clock speed.

However 14nm is very much a wall. Intel needs to either get 10nm up and running or push 7nm. However that wont be an easy task either way since their 10nm is extremely dense and 7nm requires a lot of new tools that are not quite ready.

 

[goldstone77]

Intel Allegedly Readying 10 Core, Comet Lake-S Processors – Based on 14nm Process Node, Dual-Ring Bus Rumored
By Hassan Mujtaba
Nov 25

The rumor comes straight from Taiwanese forums where it was mentioned that a 10 Core CPU under the Comet Lake-S family was mentioned during a partner meeting. The family will still be based on the 14nm process node and was recently listed in an updated DT/IOTG roadmap. The roadmap updates every quarter but we haven’t seen such roadmap in public yet.

https://wccftech.com/intel-comet-lake-10-core-processors-rumor/

 

[aldaia]

Intel Allegedly Readying 10 Core, Comet Lake-S Processors – Based on 14nm Process Node, Dual-Ring Bus Rumored
By Hassan Mujtaba
Nov 25

The rumor comes straight from Taiwanese forums where it was mentioned that a 10 Core CPU under the Comet Lake-S family was mentioned during a partner meeting. The family will still be based on the 14nm process node and was recently listed in an updated DT/IOTG roadmap. The roadmap updates every quarter but we haven’t seen such roadmap in public yet.

https://wccftech.com/intel-comet-lake-10-core-processors-rumor/

With coffee lake refresh just being released, and Ice Lake coming in late 2019 (according to Intel), I see no need for Comet Lake.

Unless .... 10nm is further delayed, or as some rumors suggest, even dead.

The most plausible rumor i heard is that late 2019 we will see a few low volume Ice Lake -Y. A "Déjà vu" of the Canon Lake paper launch of late 2017.

 

[jimmysmitty]

Intel Allegedly Readying 10 Core, Comet Lake-S Processors – Based on 14nm Process Node, Dual-Ring Bus Rumored
By Hassan Mujtaba
Nov 25

The rumor comes straight from Taiwanese forums where it was mentioned that a 10 Core CPU under the Comet Lake-S family was mentioned during a partner meeting. The family will still be based on the 14nm process node and was recently listed in an updated DT/IOTG roadmap. The roadmap updates every quarter but we haven’t seen such roadmap in public yet.

https://wccftech.com/intel-comet-lake-10-core-processors-rumor/

With coffee lake refresh just being released, and Ice Lake coming in late 2019 (according to Intel), I see no need for Comet Lake.

Unless .... 10nm is further delayed, or as some rumors suggest, even dead.

The most plausible rumor i heard is that late 2019 we will see a few low volume Ice Lake -Y. A "Déjà vu" of the Canon Lake paper launch of late 2017.

The dual ring bus is interesting as that has never been on mainstream before only HEDT. Although unless they plan to give more PCIe lanes to the CPU itself I am not sure it will be of any benefit to mainstream and I doubt they will give more than 16 PCIe lanes to the CPU since AMD isn't doing more than 16 for mainstream either and leaving more for HEDT.

 

[audiospecaccts]

They really need to work on the chip sets. These last few generations of them have really poor performance compared to what they should be IMHO. 10nm is not a hard thing if they removed all that extra "features" that is probably creating RF standing waves on the wafer and slowing the bus clock internally. There is diminishing returns when it comes to integrating extra circuits on a processor die.