Intel Says It Won’t Regain the Lead until 5 Nm

M

Marees

[H]ard|Gawd
Joined
Sep 28, 2018
Messages
1,382
Intel expects to get back to parity in the 7 nm generation(2022 & beyond) and regain leadership in the 5 nm generation

https://www.thefpsreview.com/2020/0...m-10-nm-isnt-going-to-be-quite-the-performer/

Screenshot_2020-03-04-11-04-31-802.jpeg
 
Last edited:
Isn't this timeline a little optimistic? It shows 10nm in 2019. But Intel hasn't been able to produce 10nm in any volume. Certainly not on relevant desktop parts whether for consumers or for the server space. In fact, I know of precisely zero SKU's that are 10nm in any of Intel's current or upcoming product lines.
I say that to say I find 10nm+ very unlikely in this year at least in any quantity or for any SKU that makes sense in the most relevant markets. Which (as you can see where I'm going) would have a domino effect on the years after.
It took TSMC 2 years to move from 10nm to 7nm and Samsung 3 years to do the same, at least according to that graph. But Intel is going to do it in less than half that? Granted a lot of the tech now exists and it might be possible for them to not have to learn the same things their competition did, but I still find it rather unlikely.
 
UnknownSouljer said:
Isn't this timeline a little optimistic? It shows 10nm in 2019. But Intel hasn't been able to produce 10nm in any volume. Certainly not on relevant desktop parts whether for consumers or for the server space. In fact, I know of precisely zero SKU's that are 10nm in any of Intel's current or upcoming product lines.
I say that to say I find 10nm+ very unlikely in this year at least in any quantity or for any SKU that makes sense in the most relevant markets. Which (as you can see where I'm going) would have a domino effect on the years after.
It took TSMC 2 years to move from 10nm to 7nm and Samsung 3 years to do the same, at least according to that graph. But Intel is going to do it in less than half that? Granted a lot of the tech now exists and it might be possible for them to not have to learn the same things their competition did, but I still find it rather unlikely.
Click to expand...
Well, they've been working on 10nm for a long time, they've got to go to volume production eventually. If it doesn't happen this year or next, it almost wouldn't be surprising if they just gave up and converted the fabs to a different node or sold them.
 
Let's also keep in mind that traditionally Intel doesn't use as strong marketing numbers for their process nodes.

Intel's 10nm, while not very capable of producing anything - has minimum feature sizing similar to TSMC's 7nm process.

So, when they say 5nm, it is unclear to me if they mean a TSMC 5nm or a proper Intel 5nm.
 
Zarathustra[H] said:
when they say 5nm, it is unclear to me if they mean a TSMC 5nm or a proper Intel 5nm
Click to expand...

5nm refers to Intel's 5nm which is scheduled to come up against TSMC's 3nm

Intel is expecting their 5nm to be better than TSMC's 3nm
They also expect their 7nm to match TSMC's 5nm but NOT beat it
 
if all tech companies release their latest and greatest without earning back the cost to make the previous products they would be poor.

I can see why tech companies keep doing this.

business is business still in the end.
 
Good. I hope they do. Get back into the game and come back strong. They've been knocked down before by AMD and came back strong. Real strong.

I'm all for it. A little skeptical, but if they can pull it off, great. It'll be nice to see them back to innovating and adding a lot more performance to new releases instead of a 1-2% increase... I'm not too upset because if the way they are going, I don't really need to upgrade for a couple years anyway. They don't have anything that'd really be called an "upgrade" from what I have now.
 
I'll believe it when I see it. Intel has been lying to investors for years about the state of it's 10nm process, and I just don't see them releasing two new process nodes in two years.

It was hard enough for Global Foundries to bump it's 14nm rollout to two years - that was all it was doing at the time.
 
Last edited:
Snowdog said:
How could they possibly predict this with any accuracy?
Click to expand...

Best guess. Accuracy doesn't matter. It's a goal, it's not a defined plan with a set completion date. But, they aren't trying to predict it with accuracy. It's just what they think they can accomplish with what they think they know.

I wouldn't bet on it. If it happens, great. If not, oh well. They missed a goal. Wouldn't be the first time.

Like "You're going to get some tonight!", then she ends up having a bad day at work, period starts, diarrhea, etc. and you don't get some peutange (she's French). She had the intentions and put it out there, but ... shit happens.
 
Sounds like they are assuming that AMD will not continue to innovate and refine. Intel is going to need more than a die shrink. They need a whole new architecture. And, realistically, two whole new architectures.
 
Riccochet said:
Sounds like they are assuming that AMD will not continue to innovate and refine. Intel is going to need more than a die shrink. They need a whole new architecture. And, realistically, two whole new architectures.
Click to expand...

That's the part of the plot they don't want you to think about, and probably (just my guess) why 10nm didn't come out full force.

The shrink helps improve efficiency of the entire part, but they NEED new architecture because they know exactly how riddled with security holes their existing products relying on speculative execution are.

They are probably spending big money in R&D on two things, and need time to figure out which is the right path.
  1. Can they adapt/improve usage of speculative execution, or create new tech with similar results, that will allow them to maintain their single-core speed advantage.
  2. Can they design a new architecture that doesn't rely on speculative execution, and still remain competitive?
Only time will tell.

Anyone know when Intel adopted SE?
 
Riccochet said:
Sounds like they are assuming that AMD will not continue to innovate and refine. Intel is going to need more than a die shrink. They need a whole new architecture. And, realistically, two whole new architectures.
Click to expand...

I don't think it has anything to do with AMD. It's purely about process. Fab vs Fab.
 
Snowdog said:
I don't think it has anything to do with AMD. It's purely about process. Fab vs Fab.
Click to expand...

Getting back on top has nothing to do with AMD? Then who are they trying to get back on top of? TSMC? It's going to be beyond 2021 before Intel gets ahead of TSMC, at least at this point. TSMC will be on a 5nm node with 3nm in testing before Intel gets to 7nm with acceptable yields.
 
lostinseganet said:
Isnt TSMC going to release 3nm in 2022?

https://www.eteknix.com/tsmc-confirm-3nm-is-on-track-for-2022-release/
Click to expand...

Again, TSMC uses marketing nm numbers, whereas Intel's numbers have traditionally been more accurate with respect to actual feature sizes.

The feature sizes involved with TSMC's current "7nm" process are actually closer to what Intel is calling 10nm. If Intel's 10nm had actually worked it would probably have had slightly smaller minimum feature sizes than TSMC's 7nm.

Intel's 5nm (if they can actually make it work) will probably be equivalent to or smaller than TSMC's 3nm process.

At this point it is best to view the Process Node numbers coming out of the likes of TSMC and Samsung as being more like a "generation" number, and less like anything that resembles the actual measured minimum feature sizes.

It wi be interesting to see if Intel will adopt this same type of process node size lie going forward in order to not lose the perception of how far along their processes are.
 
Last edited:
Inacurate said:
Anyone know when Intel adopted SE?
Click to expand...

Officially adopted? Pentium 4. AMD responded with the similar Speculative Loads on the Athlon XP.

The first aggressive-enough architecture (with actually exploitable bugs) was Core 2.
 
Riccochet said:
Getting back on top has nothing to do with AMD? Then who are they trying to get back on top of? TSMC? It's going to be beyond 2021 before Intel gets ahead of TSMC, at least at this point. TSMC will be on a 5nm node with 3nm in testing before Intel gets to 7nm with acceptable yields.
Click to expand...

They are talking about process fabrication, not the overall CPU market.

AMD no longer has a process foundry. So ZERO to do with AMD.

Everything to do with TSMC and Samsung.
 
Inacurate said:
That's the part of the plot they don't want you to think about, and probably (just my guess) why 10nm didn't come out full force.

The shrink helps improve efficiency of the entire part, but they NEED new architecture because they know exactly how riddled with security holes their existing products relying on speculative execution are.

They are probably spending big money in R&D on two things, and need time to figure out which is the right path.
  1. Can they adapt/improve usage of speculative execution, or create new tech with similar results, that will allow them to maintain their single-core speed advantage.
  2. Can they design a new architecture that doesn't rely on speculative execution, and still remain competitive?
Only time will tell.

Anyone know when Intel adopted SE?
Click to expand...

Partially. There are other aspects here too.

Intel relies on their greater reputation, which crazily enough does not appear to be too tarnished by the speculative execution security bugs to drive sales.

I'd imagine they will continue with their current approach assuming that if a few years after launch performance is hurt by the need to patch, users will just upgrade.

I have no confidence that they will actually change.

Even if they do - however - they have the size advantage. Intel has more engineers dedicated to CPU design than AMD has employees in total. They have the ability to out design AMD by a factor of 10-1. Because to this it really does boil down to process more than anything else.

Unless AMD has grown their market share sufficiently to afford expanding CPU engineering significantly and taken continuous improvement of their architecture very seriously they will find it very difficult to keep up once Intel's process dominance is back in place.

Intel's process node failure has given AMD a window of opportunity. Short term it is relatively straight forward to take advantage of. Long term it will be much more difficult for a much smaller company like AMD to make that permanent.

The speculative execution errors are probably just a minor detail to Intel.
 
Just going to point out that that slide is not from Intel... and the screenshot above conveniently cuts off the fine print at the bottom:

1583425822365.png
 
clockdogg said:
Intel admitting they have relinquished their lead. Seems like bigger news than the possibility of regaining it.
Click to expand...

True, but that never stops them from claiming they are better than everyone else. Again like a boxer whose getting his ass beat down, yet still has the gall to claim he's gonna beat everybody.
 
thesmokingman said:
True, but that never stops them from claiming they are better than everyone else. Again like a boxer whose getting his ass beat down, yet still has the gall to claim he's gonna beat everybody.
Click to expand...
I am pretty sure Intel will get back on track, 1/2 their 10nm woes are design based. They are using this process to also implement a new transistor design which supposedly reduces the size of the transistor circuits which when combined with the node decrease triples the density. It was too ambitious a project they flew too high and got burned. They should have done it in baby steps and scaled back their ambitions a smidge.
 
It didnt help they wrote off AMD as never being a threat again and coasted along for a decade. Then trying to push too far in one go, and then doing nothing to scale back and change what wasnt working for too long. Plus the CEO not doing anything to resolve all the problems under his reign when thing started going south. It has been a perfect storm of mistakes and failures for intel.
 
Last edited:
You must log in or register to reply here.
Back
Top