News Intel to Launch 7nm in 2021, 10nm Servers in 1H'20

[PaulAlcorn]

Intel CEO Bob Swan announced that Intel will launch its 7nm process in 2021 to challenge TSMC's 5nm products.

Intel to Launch 7nm in 2021, 10nm Servers in 1H'20 : Read more

 

[Giroro]

I am very confused what "USB Type-C on-die " means.

Isn't USB-Type C is a connector style, not really a version of USB.

 

[alextheblue]

I am very confused what "USB Type-C on-die " means.

Isn't USB-Type C is a connector style, not really a version of USB.

Marketectural enhancements!

That aside, these ARE interesting times.

 

[valeman2012]

Marketectural enhancements!

That aside, these ARE interesting times.

Well they are smart stepping back on Moore Law and let AMD CPU Business hit dead end on Moore Law first.

 

[alextheblue]

Well they are smart stepping back on Moore Law and let AMD CPU Business hit dead end on Moore Law first.

AMD hasn't been following Moore's Flaw for a while now, and neither has Intel. Just because TSMC is slightly ahead of Intel on process rollouts doesn't mean Intel is intentionally "stepping back". They had ISSUES with their 10nm (equivalent to TSMC's 7nm), it's LATE.

I know you're a fan of Intel, and that's totally cool. They've put out some really nice processors over the years. But that doesn't change the truth, so I'll just repeat this one more time: Intel didn't PLAN on waiting until 2020 to mass rollout 10nm. They ran into technical problems. End of story.

 

[Karadjgne]

He's also a fan of large print.
As if the standard 15point isn't loud enough to get the point across. Just to be fair.

Moore's Law is dead. It was dumb to begin with. Stating that a cpu will double in speed and capability every 2 years simply due to the amount of transistors a die can field is rediculous.

By that reasoning, a current cpu would be operating at close to 13.6GHz as compared to a 6 year old FX at 3.4GHz and a fx8 has 1.2Billion transistors, an R7 2700 has 4.8Billion, so that math is wrong too.

 

[valeman2012]

AMD hasn't been following Moore's Flaw for a while now, and neither has Intel. Just because TSMC is slightly ahead of Intel on process rollouts doesn't mean Intel is intentionally "stepping back". They had ISSUES with their 10nm (equivalent to TSMC's 7nm), it's LATE.

I know you're a fan of Intel, and that's totally cool. They've put out some really nice processors over the years. But that doesn't change the truth, so I'll just repeat this one more time: Intel didn't PLAN on waiting until 2020 to mass rollout 10nm. They ran into technical problems. End of story.

AMD hasn't been following Moore's Flaw for a while now, and neither has Intel. Just because TSMC is slightly ahead of Intel on process rollouts doesn't mean Intel is intentionally "stepping back". They had ISSUES with their 10nm (equivalent to TSMC's 7nm), it's LATE.

I know you're a fan of Intel, and that's totally cool. They've put out some really nice processors over the years. But that doesn't change the truth, so I'll just repeat this one more time: Intel didn't PLAN on waiting until 2020 to mass rollout 10nm. They ran into technical problems. End of story.

Not fan of any AMD or Intel...the only fans i see is AMD Fanboys getting upset over the truth.. (You look at other sites articles relating to AMD or Intel comments (Joking around or not joking around) ...see how AMD Fanboys get upset.
Just like many other games

I am fan of none but i am only going for product that would be "better" for Gaming right which is obviously intel...(If AMD was "better" i would went there ...its not really choice...mostly go for the "better" product.

 

[TJ Hooker]

Moore's Law is dead. It was dumb to begin with. Stating that a cpu will double in speed and capability every 2 years simply due to the amount of transistors a die can field is rediculous.

By that reasoning, a current cpu would be operating at close to 13.6GHz as compared to a 6 year old FX at 3.4GHz and a fx8 has 1.2Billion transistors, an R7 2700 has 4.8Billion, so that math is wrong too.

Moore's law is about is transistor density. It says nothing about performance.

With regard to FX 83xx vs R7 2700, you're forgetting about die sizes. The 2700 has about 6 times the density, meaning density doubled about 2.5 times in 6 years rather than 3 times as predicted my Moore's law. Not too far off really, despite the slowing of node progression.

Edit: And Ryzen 3000 on 7 nm will be coming out soon (~7 years after FX 83xx), which may even bring things further in line with Moore's law.

 

[bit_user]

Moore's Law is dead. It was dumb to begin with. Stating that a cpu will double in speed and capability every 2 years simply due to the amount of transistors a die can field is rediculous.

By that reasoning, a current cpu would be operating at close to 13.6GHz as compared to a 6 year old FX at 3.4GHz and a fx8 has 1.2Billion transistors, an R7 2700 has 4.8Billion, so that math is wrong too.

It was never forecast to go on, forever. In 1965, it was forecast for at least a decade. Then, in 1975, another (and at a slower rate; source: https://en.wikipedia.org/wiki/Moore's_law ). That doesn't make it dumb.

Also, you misunderstand the speed increases as relating directly to clock speed. Being able to pack more transistors on-die enabled various sorts of optimizations that could reduce the number of clock cycles per operation, and then increase the number of operations that can be performed concurrently. So, it was the combination of clock speed and efficiency which delivered the observed performance gains.

 

[bit_user]

I am very confused what "USB Type-C on-die " means.

Isn't USB-Type C is a connector style, not really a version of USB.

Nomenclature aside, I assume the point was to support USB-connected displays.

 

[JaSoN_cRuZe]

If the rumors were true about AMD's Zen 2 chipsets, it would be a hard sell of Intel processors.
I think by the time they release 10nm desktop parts, AMD would be sitting @ 5nm.
Intel better hustle.

 

[Deleted member 14196]

Moore’s law was never a law all. it was an observation! and was never prove to be a law and anyone who thinks it’s a law is wrong and has been wrong forever now. It is really dumb to call something a law that is merely an observation and anyone in science could tell you that it wasn’t the law based on physics

 

[Armani316]

I was hoping to learn more about the H-series mobile CPUs within the 10nm Ice Lake or 10nm Tiger lake families. Any ETA on those? Looks like next year's 10th Gen CPUs will introduce yet another 14nm refresh for the H-series in the form of Comet Lake. Now word about the 7nm H-series either.

 

[jimmysmitty]

If the rumors were true about AMD's Zen 2 chipsets, it would be a hard sell of Intel processors.
I think by the time they release 10nm desktop parts, AMD would be sitting @ 5nm.
Intel better hustle.

TSMC expects tape out for their 5nm to be in 2020. AMD stated its 7nm+ Zen 3 will be out in 2020. Intels 10nm will probably be equivalent to TSMCs 7nm and Intels 7nm will probbaly be equivalent to TSMCs 5nm. Remember the nm naming isn't quite as it once was. Intels 10nm was originally, not as sure now without the information on it, going to be quite a bit more dense than TSMCs 7nm.

And I think they are hustling. Planned launch of 10nm next year and 7nm the year after? Thats a pretty fast rollout. It is to be seen though as with anything predicted release dates and actual release dates are not always 100%.

 

[TJ Hooker]

Moore’s law was never a law all. it was an observation! and was never prove to be a law and anyone who thinks it’s a law is wrong and has been wrong forever now. It is really dumb to call something a law that is merely an observation and anyone in science could tell you that it wasn’t the law based on physics

I don't think anyone really though it was a law in the scientific sense of the term (nor was it supposed to be interpreted that way). Anymore than people think Poe's or Godwin's laws are physical laws of science.

 

[Karadjgne]

For example, the 2010 update to the International Technology Roadmap for Semiconductors predicted that growth would slow around 2013,[18] and in 2015 Gordon Moore foresaw that the rate of progresswould reach saturation: "I see Moore's law dying here in the next decade or so."[19]

Even Gordon said it was about dead, in 2015. With the advent of the 7nm process its seriously going to slow down further, the only viable alternative is a backwards trend, making cpu dies either bigger (Intel) or more numerous (amd). They've about hit the 5nm brick wall where silicon is to the point it just won't conform to a smaller process, physical characteristics and electromigration being that wall. The only way you'll see a viable 5nm or smaller solution is a change from silicon as a base, and neither company has figured out what that could possibly be, yet. It's going to be considerably longer than 2 years before you'll see a doubling of transistors, speeds and capabilities. You'll get maybe 2 or 3 more generations, possibly with a die shrink on Intels part, without either some sort of miricle breakthrough or some really large dies.

 

[Geef]

Intel is SO full of sh**. They have continued giving time frames for their 10nm and now 7nm and we still have not seen either from them. I will believe it when I see it which won't be in 2021.

 

[nicholas70]

So by 2030 will we see the 1nm chips finally start rolling out?

 

[Karadjgne]

Nope. Not without a miricle. Silicon won't go that small. Apart from thermal insulation between the transistors, the silicon also has to provide electrical insulation. Which it doesn't. You get far too much arcing at even 1v. You'd have to figure out how to get a tx to run at half of that, to not have the power sufficient to jump, and that reduction in power takes away the ability to get any speed. They could probably figure out a 5nm process that works, but the cpu would be back to running @ 1GHz.

 

[jimmysmitty]

Nope. Not without a miricle. Silicon won't go that small. Apart from thermal insulation between the transistors, the silicon also has to provide electrical insulation. Which it doesn't. You get far too much arcing at even 1v. You'd have to figure out how to get a tx to run at half of that, to not have the power sufficient to jump, and that reduction in power takes away the ability to get any speed. They could probably figure out a 5nm process that works, but the cpu would be back to running @ 1GHz.

Which is why Intel and plenty of others like IBM have been working for probably a decade on other materials to move to from silicon.

 

[TJ Hooker]

They could probably figure out a 5nm process that works, but the cpu would be back to running @ 1GHz.

TSMC already has a working 5nm node (although it's hard to say how it will compare to Intel's eventual 5nm node). Apparently HPC is one of the areas they're targeting with the node, so it seems unlikely it'd be limited to low frequencies.

https://www.anandtech.com/show/1417...chnology-pdk-drm-eda-tools-3rd-party-ip-ready

 

[nicholas70]

I think I read somewhere that a 1nm transistor had been built already, but I think it was made from carbon nanotubes or graphine. I just can't imagine in a decade that we would be stuck at 5nm. Well actually I guess in light of whats happened with 14nm maybe, but it will be sad if we get stuck at the next node or two and cpus can only start growing out vs becoming more dense with transistors.

 

[jimmysmitty]

I think I read somewhere that a 1nm transistor had been built already, but I think it was made from carbon nanotubes or graphine. I just can't imagine in a decade that we would be stuck at 5nm. Well actually I guess in light of whats happened with 14nm maybe, but it will be sad if we get stuck at the next node or two and cpus can only start growing out vs becoming more dense with transistors.

I think I read somewhere that a 1nm transistor had been built already, but I think it was made from carbon nanotubes or graphine. I just can't imagine in a decade that we would be stuck at 5nm. Well actually I guess in light of whats happened with 14nm maybe, but it will be sad if we get stuck at the next node or two and cpus can only start growing out vs becoming more dense with transistors.

I think thats the theory is that anything below 5nm will have to be with carbon nanotubes. The problem is that thus far nothing they have successfully tested with has beaten silicon or if it has its probably just too expensive.

At some point we will be stuck. We will have to think of other ways to increase density such as chip stacking or something else. We are hitting the limits of materials physically and beyond 7nm silicon has shown it will have a lot of problems to overcome. It might be too expensive to even go after.

I know Intel has been working on alternate materials for a long time. Thats why they went to Hi-K metal gates and started also using Hafnium in their process technology.