Discussion Intel current and future Lakes & Rapids thread

[Zucker2k]

LOL. It is energy efficiency that matters, especially in mobile. In that metric, apple silicon is somewhat more than "slightly faster". Also, nice attempt at moving the goalposts by citing market share as a replacement for technological leading edge.

If technological advancement is not going to automatically translate into market share then what really is your argument here?

 

[dmens]

If technological advancement is not going to automatically translate into market share then what really is your argument here?

Why you mad? I even said a few posts earlier, that even the massive technological difference *might* change peoples’ minds. Because obviously superior performance is only part of the mindshare.

That said, I didn’t bring up market share, someone else did. My argument is simply that Intel has definitively lost the tech race permanently in mobile and likely in every other market as well.

 

[SAAA]

He pointed out that Intel doesn’t have many EUV lithography machine, especially relative to the number of wafers they manufacture every month. It true, it’s another sign of gross mismanagement. The only other scenario that makes sense to me is that 7nm is going to be a short node and Intel will be relying on 10/7nm 3D hybrid packaging. Intel would have to go all in on their 5nm. I don’t see away for Intel to get back on a leading edge process in the next 10 years, if ever. That’s probably why the US semiconductor funding initiative is focused mainly on R&D.

Isn't that exactly their plan? To make CPU tiles for Meteor on 7 nm (yield and performance reasons) while the rest of the core could be fabbed on 10 nm? Maybe IGP on a separate 10-7 nm tile too, so they can finally have either small or big graphics without being tied to 2-3 layouts from were they cut variants.
Ideally that would mean they can start mass producing sooner than when 7nm is fully ready to yield 250 mm^2 dies, though it does bring new issues with packaging.

 

[Ajay]

Ideally that would mean they can start mass producing sooner than when 7nm is fully ready to yield 250 mm^2 dies, though it does bring new issues with packaging.

That's pretty big for a relatively small number of EUV machines at present.

 

[TransientStudent]

That's fine. The boat that sailed is Intel being the leading edge in mobile (or anything else for that matter).

They could always turn to TSMC, and in that way return to the leading edge. Not as a wafer manufacturer, but as a CPU supplier, with packaging and some tiles done in-house. Isn't that the way the wind is blowing, with the talk of Intel doing some stuff on TSMC 3nm?

Also, it is no natural law that Apple is a node ahead of everybody else at TSMC. That is only a question of money. If some in the Windows world (Microsoft, HP, Dell,..) decides they must up the CPU budget, then both AMD and Intel are perfectly capable of creating large dies (as large as Apple) on the bleeding edge. AMD certainly is, if only the customers are willing to pay.

 

[dmens]

They could always turn to TSMC, and in that way return to the leading edge. Not as a wafer manufacturer, but as a CPU supplier, with packaging and some tiles done in-house. Isn't that the way the wind is blowing, with the talk of Intel doing some stuff on TSMC 3nm?

Also, it is no natural law that Apple is a node ahead of everybody else at TSMC. That is only a question of money. If some in the Windows world (Microsoft, HP, Dell,..) decides they must up the CPU budget, then both AMD and Intel are perfectly capable of creating large dies (as large as Apple) on the bleeding edge. AMD certainly is, if only the customers are willing to pay.

Even if TSMC lets Intel play on the leading edge (they won't because Apple has already paid for that privilege), your statement implies Intel architecture and design is competitive without their historically overwhelming process advantage. Pro-tip: they are not.

 

[DrMrLordX]

He pointed out that Intel doesn’t have many EUV lithography machine, especially relative to the number of wafers they manufacture every month.

Exactly.

Allegedly Intel's 14nm production in 2018 was 70 wkpm. Not sure if that's their peak, but it's the last year that they produced a new (er, mostly new?) Xeon that actually saw widespread industry adoption: Cascade Lake-SP. Cascade Lake-AP and Cooper Lake were duds, so we can safely ignore those.

By 2023, Intel's 10nm capacity will supposedly be 65 wkpm while 7nm will be at 20 wkpm (which has been cited multiple times in this thread already). At that point it's a question of: how many working CPUs can Intel harvest per 10nm wafer vs. what they achieved three years ago on 14nm? XCC dice did take up a lot of Intel's 14nm supply.

It true, it’s another sign of gross mismanagement. The only other scenario that makes sense to me is that 7nm is going to be a short node and Intel will be relying on 10/7nm 3D hybrid packaging.

Not sure how they're going to remain competitive doing that. Intel's 7nm will be in risk production until 2023, while TSMC 3nm (which is the node most of their competitors will be using or will wnat to be using) will leave risk production in H2 2022. That'll put Intel at a pretty big disadvantage.

Intel would have to go all in on their 5nm.

. . . which is what people were saying about their 7nm process, last year. We can see how that worked out.

 

[TransientStudent]

Even if TSMC lets Intel play on the leading edge (they won't because Apple has already paid for that privilege), your statement implies Intel architecture and design is competitive without their historically overwhelming process advantage. Pro-tip: they are not.

But can't that be turned around? Surely it must be possible for Intel to build some great CPU architecture/design teams if they do not already have that at the moment. Or is the mismanagement that deep?

 

[dmens]

But can't that be turned around? Surely it must be possible for Intel to build some great CPU architecture/design teams if they do not already have that at the moment. Or is the mismanagement that deep?

LOL. Great teams take many years to build up. They can be eviscerated in a matter of months. Intel management has been deliberately encouraging brain drain for 5 years now. You do the math.

 

[TransientStudent]

Not sure how they're going to remain competitive doing that. Intel's 7nm will be in risk production until 2023, while TSMC 3nm (which is the node most of their competitors will be using or will wnat to be using) will leave risk production in H2 2022. That'll put Intel at a pretty big disadvantage.

I think it is almost certain Intel will use TSMC 3nm also. There is then the question about capacity, and of course also that Intel's own manufacturing may be dead for the bleeding edge when they start to rely on TSMC for the most critical tiles.

 

[TransientStudent]

LOL. Great teams take many years to build up. They can be eviscerated in a matter of months. Intel management has been deliberately encouraging brain drain for 5 years now. You do the math.

But still ADL does not sound so bad. Maybe too power hungry, but maybe not so much if it had been on TSMC 5nm.

 

[eek2121]

LOL. It is energy efficiency that matters, especially in mobile. In that metric, apple silicon is somewhat more than "slightly faster". Also, nice attempt at moving the goalposts by citing market share as a replacement for technological leading edge.

Tiger Lake already beats the M1 in single threaded performance. If tiger lake were on TSMC 5nm, it would be just as efficient.

Why you mad? I even said a few posts earlier, that even the massive technological difference *might* change peoples’ minds. Because obviously superior performance is only part of the mindshare.

That said, I didn’t bring up market share, someone else did. My argument is simply that Intel has definitively lost the tech race permanently in mobile and likely in every other market as well.

Nobody is mad, we just aren’t blinded by fanboyism. Intel is the largest laptop chip maker in the world. That is unlikely to change for the next 5-10 years (if ever). Apple isn’t suddenly going to become a major player because they switched from x86 to ARM. Much of the world is firmly entrenched in Windows, Linux, and x86. There are no competitive desktop or laptop chips that run Windows outside of x86. Apple chips have no official Linux support.

Because of this, Intel’s future offerings will always be of interest to many. Intel is a very important part of the computing market and they will continue to be.

 

[Ajay]

Not sure how they're going to remain competitive doing that. Intel's 7nm will be in risk production until 2023, while TSMC 3nm (which is the node most of their competitors will be using or will wnat to be using) will leave risk production in H2 2022. That'll put Intel at a pretty big disadvantage.
<…>
. . . which is what people were saying about their 7nm process, last year. We can see how that worked out.

Well, Intel hasn’t boasted about 10nm, tells us all we need to no, IMHO.

The 10/7 isn’t going to work well for them because there are too few 7nm wafers - so they will have to choose. 7nm cores will have to go to mobile for sure - what about server ? Desktop is going to be in the lose column against N5 Zen4. The current 'Cove' architectures just suck in power efficiency and they are burning too many watts to keep performance up.

Intel just didn’t buy enough tools for 7nm, if they have upped their buy on tools for 5nm in time, maybe they’ll be in a better position - maybe. Stupid decision to spend money on huge stock buybacks instead of fab equipment. So sad, smh.

 

[eek2121]

Well, Intel hasn’t boasted about 10nm, tells us all we need to no, IMHO.

The 10/7 isn’t going to work well for them because there are too few 7nm wafers - so they will have to choose. 7nm cores will have to go to mobile for sure - what about server ? Desktop is going to be in the lose column against N5 Zen4. The current 'Cove' architectures just suck in power efficiency and they are burning too many watts to keep performance up.

Intel just didn’t buy enough tools for 7nm, if they have upped their buy on tools for 5nm in time, maybe they’ll be in a better position - maybe. Stupid decision to spend money on huge stock buybacks instead of fab equipment. So sad, smh.

I'm going to play devil's advocate here and point out we know nothing about their 7nm process. I am inclined to agree with you in theory, but who knows, maybe Intel is keeping it's EUV use down to a minimum? Have there been ANY details leaked surrounding their 7nm process?

 

[IntelUser2000]

Alder Lake's cache specs revealed? Seems the Big Cores have their L2 assositivity at 10 ways (down from 20 in Willow Cove). The Small Cluster L2 is 16 and it gets 2 MB per cluster.

Some juicy information. Golden Cove keeps the 48KB L1D, 32KB L1I configuration. Gracemont gets the 32KB L1D, 64KB L1I like the leak months ago.

and set the laptop on fire?

The iGPU takes a small portion of power only when it's running. When it's not it uses practically zero power.

 

[Gideon]

I'm going to play devil's advocate here and point out we know nothing about their 7nm process. I am inclined to agree with you in theory, but who knows, maybe Intel is keeping it's EUV use down to a minimum? Have there been ANY details leaked surrounding their 7nm process?

Looking at the density it's just no way in hell they could go "light" on EUV. 10nm already uses quad-patterning and other rather extreme solutions and 7nm has more than double the density. Intel's 7nm is somewhere between TSMC 5nm and 3nm According to Anandtech. Wikichip says TSMC 5nm uses 10 EUV layers. And once you use EUV you need the machines in every fab ...

 

[IntelUser2000]

@Gideon I doubt it's 2.3x now. They stated a more conservative 2x. Still 200 million transistors is a lot.

Whatever Intel says about density doesn't apply squat to majority of their product lines, since the big Core scales at a historical 0.5x no matter their claims. Chips like Atom and graphics do more but that's a much smaller proportion of product and line.

So for Core it would go like this:
22nm - 16
14nm - 32
10nm - 64
7nm - 128

 

[dmens]

Tiger Lake already beats the M1 in single threaded performance. If tiger lake were on TSMC 5nm, it would be just as efficient.

Bwahahahaha and you actually call others fanboys? Wow. You could just compare the perf/power metrics on the N7 vs N5 for Apple and figure out an approximate process scaling improvement and realize it isn't anywhere close to being the 2x or more that Intel would need to even approach Apple M1.

Nobody is mad, we just aren’t blinded by fanboyism. Intel is the largest laptop chip maker in the world. That is unlikely to change for the next 5-10 years (if ever). Apple isn’t suddenly going to become a major player because they switched from x86 to ARM. Much of the world is firmly entrenched in Windows, Linux, and x86. There are no competitive desktop or laptop chips that run Windows outside of x86. Apple chips have no official Linux support.

Because of this, Intel’s future offerings will always be of interest to many. Intel is a very important part of the computing market and they will continue to be.

Even if you use the Windows compatibility argument, AMD is winning against Intel in the technological race. If you ditch Windows then it is already game over. Don't worry, you can continue to sneer at the casuals sitting in their locked garden with processors that are superior in literally every engineering metric. LOL.

 

[eek2121]

Bwahahahaha and you actually call others fanboys? Wow. You could just compare the perf/power metrics on the N7 vs N5 for Apple and figure out an approximate process scaling improvement and realize it isn't anywhere close to being the 2x or more that Intel would need to even approach Apple M1.



Even if you use the Windows compatibility argument, AMD is winning against Intel in the technological race. If you ditch Windows then it is already game over. Don't worry, you can continue to sneer at the casuals sitting in their locked garden with processors that are superior in literally every engineering metric. LOL.

You need to find a different sub to troll. Did I mention Intel was beating Apple from an efficiency standpoint? No, I stated that if they were on 5nm, they would be as efficient, which you will find it very difficult to argue against with any type of realistic mindset.

You are simply here to troll those who are trying to have a productive conversation about future Intel products. There is an Apple thread for apple lovers to go to, and multiple AMD threads for AMD users to go to. P.S. you'll find I regularly contribute to both of those. Buzz off unless you have something productive to contribute.

Oh and PS, I own 2 x Zen 2 chips, and have a Zen 3 chip and a Tiger Lake H laptop coming in the mail. I ALSO own an M1 Mac Mini and an Intel Macbook Pro, so get lost. (fun fact I also have both AMD and NVIDIA GPUs).

 

[dmens]

No, I stated that if they were on 5nm, they would be as efficient, which you will find it very difficult to argue against with any type of realistic mindset.

Actually, it would be extremely easy to argue against.

- Intel has always said their enhanced 10nm is as good as TSMC 7nm (and I will give them that)
- N7 to N5 is a marginal improvement
- At no point on any of the Firestorm operating power envelope is Tigerlake anything better than half as good on power efficiency. In fact, at the 4W range that Firestorm tops out at, Tigerlake is literally crippled.

Feel free to have your "productive conversations", just don't be mad when people point out you are wrong.

 

[eek2121]

Actually, it would be extremely easy to argue against.

- Intel has always said their enhanced 10nm is as good as TSMC 7nm (and I will give them that)
- N7 to N5 is a marginal improvement
- At no point on any of the Firestorm operating power envelope is Tigerlake anything better than half as good on power efficiency. In fact, at the 4W range that Firestorm tops out at, Tigerlake is literally crippled.

Feel free to have your "productive conversations", just don't be mad when people point out you are wrong.

Yes, a nearly 1.81X density increase is "marginal". Blocked.

 

[dmens]

Yes, a nearly 1.81X density increase is "marginal". Blocked.

Again with the moving goalposts, confusing density with transistor power/perf performance.

Go look at A13 vs A14 and prove there is a 81% increase in power/perf.

 

[itsmydamnation]

Yes, a nearly 1.81X density increase is "marginal". Blocked.

you really a sore looser!

What does density have to do with efficiency of intels uarch?

Remember when for both zen2 and zen3 AMD gave uarch power improvements agnostic of process numbers ( about 18% for each).
thats the bit intel hasn't improved in a long time.

 

[scannall]

But can't that be turned around? Surely it must be possible for Intel to build some great CPU architecture/design teams if they do not already have that at the moment. Or is the mismanagement that deep?

Intel got rid of a lot of their senior engineers through forced retirement, or outright firing since they are expensive. Add to being heavily poached. They had a HUGE lead everywhere so the bean counters decided to cut costs. Looked great short term on the balance sheet, but it has proven very expensive in the long run.

While they still have good engineers, their bench isn't all that deep any more.

 

[IntelUser2000]

@eek2121 Intel cores lose against AMD, their closest competitor using the same ISA.

Zen 3 increased core size only by 14% over Zen 2. Sunny Cove is how much larger than Skylake? 50%? Willow Cove adds cache but the reason for the addition seems mysterious as it performs pretty much like Sunny Cove.