How long will 10nm last?

Astrowind

Weaksauce
Joined
Sep 28, 2018
Messages
116
I know this seems like a useless question. But, since socket 775 lasted a long while and we've been on 14nm for several years now, when I read Tiger Lake (7nm) (EDIT: I just read it will be on 10nm) will come out soon after Ice Lake (10nm), would it really be worth it to skip 10nm and just wait for 7nm?

Also, it's nanometer right? I heard some guy on Youtube say n m...
 
Last edited:
It is indeed nanometer. While Youtube has some very well educated people, there are a ton of posers and pretenders.

As for process node, process node really has no impact on what socket the processor uses. Socket design (backwards compatibility or not) is all done by the engineering team, with some input from marketing I'm sure.

As for skipping this generation for the next, depends on what you want out of your computer. If it doesn't do what you want/need it to do now, upgrade now. Otherwise, it's always best to wait until you want/need to upgrade.
 
It will last as long as any Intel platform independent of any (pulled out of the intel hat) technical limitations...
 
In terms of socket longevity, SINCE Intel has started Integrated memory controllers Intel has let sockets lasts two generations. It only stopped being "new processor architecture, then die shrink next year " after 10nm took way too damn long.

And that 10nm delay is why you're still seeing two sockets on 14nm:
Skylake + Kaby Lake used same socket
Coffee Lake (6 cores) + Coffee Lake Refresh (8 cores) used same socket

We're still not sure on whether 10nm chips will launch on the same socket as Coffee Lake, but that really depends on how quickly they can be put large chips into production. If it's not until next year, I'd expect a new socket. Coffee Lake Refresh was just Intel anticipating AMD's Zen 2 7nm, and knowing they have no 10nm chips for another year.
 
Last edited:
Well, looks like the competition went down even to the Pentium level because the Pentium G5400 is probably about the same performance (deducting from benchmarks I've seen) as my i5-4570 but less expensive.

I'm still happy with my i5-4570 and GTX 1050 Ti. I'm just trying to plan ahead because once Win7 support ends in Jan, 2020, I expect I'll likely build a new machine. Next one though might be with 16GB RAM, SSD and 6 core CPU. (I still have 8GB RAM and regular HD.)
 
Much like that i5 served you well for 6 years, the 6-core processors you can buy now will have a similarly long-life.
 
My system hasn't changed in the last 6 years (i7-4930k with GTX 690) and likely won't change for another 2-3 years. With the games I play, I don't see a need to until I upgrade to 1440p monitors.
 
Buy what you need when you need it. Only exception is when it's obvious that there's a release imminent, and I do mean really imminent- not just 'something's always around the corner' imminent.
 
I upgrade my GPU the most frequently but have been rocking my i7 980X for 10 years...I was tempted with the latest Coffee Lake refresh but held out...I plan on upgrading once the 10nm Ice Lake CPU's are released at the end of 2019
 
I upgrade my GPU the most frequently but have been rocking my i7 980X for 10 years...I was tempted with the latest Coffee Lake refresh but held out...I plan on upgrading once the 10nm Ice Lake CPU's are released at the end of 2019
get VR headset and immediately your CPU will become obsolete

myself struggling not to upgrade to Coffe Lake waiting for Zen2 or 10nm Intels... which will most probably stomp AMD in games... and my wallet also = (
 
7 nm will be coming out in 2017.

TT6_575px.png
 
90nm took everyone about a year longer than expected, but the industry recovered and kept to 2 years/node change.

Then 14nm came along,and the industry lost.
 
Here's the funny thing about process nodes: the "shrink every two years" didn't happen until the 1990s. Before that, process tech shrunk about every four years, because the industry was still learning. Different process types ( e.g going from nmos to nmos depletion mode), or a larger silicon wafer size could have as much positive effect as an actual node shrink.

Moore's Law only existed in the senses that tons of easy-to-implement process optimizations , along with increasing levels of automation meant actual production would roughly double every 18 months. And though the costs did not stand still, they tended to rise a lot slower than the profits.

Just to give you some idea of how quickly these things changed:

img_report04_02.jpg


The Motorola 6800 was fabbed on a cutting-edge process featuring 3 inch wafers. Jumping from 3" to 4" gives you 75% more area, so if you maintain yields, that is how much you raise your profits (minus cost of larger wafers and upgrade). We're till trying to work the way through 18 inches, but if they ever can it will increase throughput per-wafer by over 100%.

Most off the amazing performance improvements came from doubling the die size from one new architecture to the next (as you got to know the new larger wafer sizes, you tended to put them to use making something vastly more powerful). See here:

8086 = 33 mm^2
80286 = 47 mm^2
80386 = 104 mm^2
80486 = 160 mm^2
P5 = 293 mm^2
P6 = 307 mm^2 ( plus L2 cache

It was just as much about learning to mass-produce larger processors with less defects as it was the march to smaller node sizes. Otherwise, the PC performance revolution would never have happened.

It wasn't until Intel hit the power and 12-inch wafer size wall with the the P6 that they cared about their process node feature size at all.

This is the same system that drove the 3D graphics industry: die sizes ballooned from 90 mm^2 (TNT) to 500 mm^2 (8800 GTX) in just ten years, as the companies got better at custom silicon. The the lack of competitors meant they could ram-up production.
 
Last edited:
OK, I want to add here it looks like I made a mistake. It looks like Tiger Lake will come out at 10nm. And I just found out before Ice Lake comes out, there will be yet another version of 14nm coming out (lol) the Comet Lake coming out mid 2019 or later.

So, it looks like it will be: Coffee Lake refresh (14nm) > Comet Lake (14nm) > Ice Lake (10nm) > Tiger Lake (10nm)

In that case, I may not wait for Tiger Lake. However, when support ends for Windows 7 in January 2020, I'll still have a choice between choosing Comet Lake or waiting for Ice Lake. Since I have a Core2Duo (45nm) for Windows XP and my main machine with Win7 is a Haswell (22nm) and my backup machine a Skylake (14mn), I may very well wait for 10nm so I can feel like I actually had an improvement lol.
 
OK, I want to add here it looks like I made a mistake. It looks like Tiger Lake will come out at 10nm. And I just found out before Ice Lake comes out, there will be yet another version of 14nm coming out (lol) the Comet Lake coming out mid 2019 or later.

So, it looks like it will be: Coffee Lake refresh (14nm) > Comet Lake (14nm) > Ice Lake (10nm) > Tiger Lake (10nm)

In that case, I may not wait for Tiger Lake. However, when support ends for Windows 7 in January 2020, I'll still have a choice between choosing Comet Lake or waiting for Ice Lake. Since I have a Core2Duo (45nm) for Windows XP and my main machine with Win7 is a Haswell (22nm) and my backup machine a Skylake (14mn), I may very well wait for 10nm so I can feel like I actually had an improvement lol.

Comet lake is Intel's 9th gen version of the 8086k, a way to steal some of the thunder away from Ryzen 3000.

Many expect Icelake will be widely available early 2020. However, it will be interesting if they choose DDR5 or DRR4. If they choose to use Icelake/Tigerlake to use DDR5, they may have to push even further as I doubt DDR5 will be ready for prime time.

For all your future Intel speculating:
https://hardforum.com/threads/intel...celake-tigerlake-sapphire-rapids-etc.1945886/
 
@juangra - TSMC was predicting 10 nm in 2016. I would say they were much closer considering they have 7 mm in early 2019.

Besides, it was just a joke. No need to go all nuclear on me.
 
I read Intel bought a company to accelerate their progress with GPUs. That doesn't tell me whether Ice Lake or Tiger Lake will get the improved GPU. However, it does tell me now I'm not sure which one I'll get, Icel Lake or Tiger Lake. This is because when I complete a machine, I don't want to immediately be required to buy a video card.
 
@juangra - TSMC was predicting 10 nm in 2016. I would say they were much closer considering they have 7 mm in early 2019.

Besides, it was just a joke. No need to go all nuclear on me.

TSMC launched 10nm on late 2017, so they missed their own roadmap, and this 10nm is more close to Intel 14nm than to Intel 10nm.

My goal was to put things in context to explain why Intel is having so many trouble and delays with the 10nm node. Since Intel announced 10nm in the roadmap that you linked, many things happened, including that Glofo canceled two 10nm nodes (first 10XM then 10LP), IBM foundries closed because couldn't keep developement, and resulting Glofo-IBM join finally healt 7nm (7LP). Only TSMC has been able to get a low performance node (7FF) with a similar density to Intel 10nm.
 
The last CPU I bought was for my backup, the i3-6100 was $140 Canadian. The i5-4570 I bought for my main machine was $208 Canadian. The GTX 1050 Ti I bought was $200 Canadian but only because I wanted the model with heatpipes. I actually downclocked the vidcard to regular speed so it would still use only 75 watts and wouldn't turn on the fans. And the only expensive thing I usually do is add a Coolmeraster Gemini S4 cooler to my CPU because with heatpipes, if a fan fails for whatever reason, I have protection with the heatpipes.

So, with a quality power supply and 120mm fans everywhere, I have near-silent machines that are reliable.

By the way, are they still using silicon for 10nm and 7nm?
 
TSMC launched 10nm on late 2017, so they missed their own roadmap, and this 10nm is more close to Intel 14nm than to Intel 10nm.

My goal was to put things in context to explain why Intel is having so many trouble and delays with the 10nm node. Since Intel announced 10nm in the roadmap that you linked, many things happened, including that Glofo canceled two 10nm nodes (first 10XM then 10LP), IBM foundries closed because couldn't keep developement, and resulting Glofo-IBM join finally healt 7nm (7LP). Only TSMC has been able to get a low performance node (7FF) with a similar density to Intel 10nm.

How do you even know what Intel's density at 10nm even is? The one i3 they released on the 10nm node was a huge turd. Other than having impressive AVX performance, it was worse than other i3's on their 14nm node. So, for the time being I would imagine that no one knows anything meaningful about Intel's long speculated and massively delayed 10nm node.
 
How do you even know what Intel's density at 10nm even is? The one i3 they released on the 10nm node was a huge turd. Other than having impressive AVX performance, it was worse than other i3's on their 14nm node. So, for the time being I would imagine that no one knows anything meaningful about Intel's long speculated and massively delayed 10nm node.

We don't know- but we do have Intel's history of product execution to go by. So if you were to make a bet, it'd be prudent to bet that Intel will have it figured.
 
How do you even know what Intel's density at 10nm even is? The one i3 they released on the 10nm node was a huge turd. Other than having impressive AVX performance, it was worse than other i3's on their 14nm node. So, for the time being I would imagine that no one knows anything meaningful about Intel's long speculated and massively delayed 10nm node.

We know the densities of the libraries of the nodes developed by both TSMC and Intel. And for Intel, data isn't only coming from Intel foundry engineers at the 64th IEEE International Electron Devices Meeting (IEDM) that was held in San Francisco in December of past year (aka two months ago).

I already explained Intel 10nm node is "massively delayed", because it is a very complex node. Rest of foundries either canceled a similar node (Glofo) or are out business (IBM) or released an inferior node (TSMC).

The 7FF node released by TSMC has density slightly lower than the 10nm (HD) libraries by Intel (96 MTr/mm² vs 101 MTr/mm²), but the TSMC node only targets clocks in the 2--3GHz range. Moreover, this 7FF node is being used on small dies due to yields/costs.

The TSMC node that will by used by AMD for Zen2 is 7HPC, which target clocks in the 4--5GHz, but is delayed (compared to 7FF) and has only a density of about 67 MTr/mm², which is closer to Intel 10nm UHP libraries (except that UHP targets higher clocks than 7HPC).
 
By the way, are they still using silicon for 10nm and 7nm?
Yup, amazing ain't it? Not by choice mind you... they've just not been able to come up with a viable, low-cost solution. :(
I mean, there HAVE been new material advancements made over the years, but clearly they've not been adopted for one reason or another, which is a bummer. However, on the flip-side of it... thank you, Silicon, for getting us even this far!! heh
 
Back
Top