Intel intros a new node naming scheme - 10nm Enhanced SuperFin is now Intel 7

[WYP]

If you can't beat them, join them.



Read more about Intel's lithography roadmap.

[NeverBackDown]

I don't think it has anything to do with if you can't beat them join them. They were effectively equal. Intel was just being far more honest about it based off previous measurement standards. If anything they should be commended for not using marketing tricks.


The new method of lithography measurement should be based on Transistors/mm^2. Or as Intel used to use before this change unless they keep this going, MTr/mm2(mega-transistor per mm squared).

Or if you want to follow a different way if measuring the industry should adopt LMC, though it's more of a system measurement for CPU/RAM purposed by nine authors from Stanford, MIT, Berkeley, and TSMC. It does have it's own downsides though but at least it would be standardized
https://ieeexplore.ieee.org/document/9063714

[Warchild]

Is Intel 20A a link to the amount of amps you need to run through the new nodes in order to beat AMD?

just kidding. Its been a long long time since I had any interest in CPU media, mainly because we just dont see innovation. I wonder how this will change things.

[WYP]

Originally Posted by NeverBackDown

I don't think it has anything to do with if you can't beat them join them. They were effectively equal. Intel was just being far more honest about it based off previous measurement standards. If anything they should be commended for not using marketing tricks.


The new method of lithography measurement should be based on Transistors/mm^2. Or as Intel used to use before this change unless they keep this going, MTr/mm2(mega-transistor per mm squared).

Or if you want to follow a different way if measuring the industry should adopt LMC, though it's more of a system measurement for CPU/RAM purposed by nine authors from Stanford, MIT, Berkeley, and TSMC. It does have it's own downsides though but at least it would be standardized
https://ieeexplore.ieee.org/document/9063714

Yes, but as things have moved on it has become clear that there is more to lithography technology than transistor density. While I agree that nm is a bad metric, it conveys the point that manufacturers look at.

For example, GF12nm is an optimised form of 14nm that can offer more performance and silicon density. Hence 12nm. TSMC 6nm is an optimised form of 7nm that can deliver some more density and performance, hence 6nm. The numbers may be marketing, but it conveys the required message, albeit inaccurately.

TBH, there is no single metric that can be used to convey the right message, as both silicon density and performance matter. There is a reason why only 10nm Enhanced SuperFin will be called Intel 7. Intel's other 10nm nodes weren't even good enough for huge scale production. There is a reason why Intel has launched 14nm CPUs this year.

Transistor density alone is not a good enough metric, and while the "standard" nm naming scheme is not accurate it is understood, which is the entire point of the naming scheme.

The expectation is that nodes with smaller numbers are more transistor dense and offer performance/watt improvements over their predecessors. Going beyond that goes above the paygrade of most folks and anyone actually building processors can look at the full specs of each node.

At the end of the day, Intel's name changes are just marketing. It is them getting away from the 14nm+++++ era and into something new.

Should Intel be commended? Not after their prior attempts at 3rd party manufacturing. Node names with less marketing spin is a small deal when compared to that.

Remember when LG planned to use Intel 10nm to make mobile chips in 2016? Now that was a disaster that Intel did a great job hushing up. Now LG has exited the phone industry entirely... Intel had other customers planned for their original 10nm node, it was a disaster.

https://www.vox.com/2016/8/16/125072...re-korea-intel

[NeverBackDown]

Transistor density is a far better metric than anything they use now. It directly correlates to performance, so your point of density and performance matter, is exactly why it's a good metric. This is just an agreement you are making while trying to prove it's not a good metric. The more transistors per square mm, 1) correlates to actual smaller lithography, 2) means more performance in a given area, 3) means you can make smaller CPUs which can lead to higher efficiency, or translate to more complex designs for high performance (adding new instructions for say AI, etc), 4) allows a greater margin of profitability since you can make more per wafer

Using any example of an "optimized" version of a node is not that strong of an argument imo. It could very easily have been called Manufacturer name :XXnm+. As it should have be since it was an optimized version of the previous. It conveys the same message of advancement and better.

None of the examples really have any merit. Ya people understand it but it doesn't make a difference to the actual real life measurement. People who care are the ones who would understand the new technique of measuring. The average Joe already doesn't know or care about what NM size their CPU is.

Should also remember those efficiency gains quoted by manufacturers such as 20% improved perf/watt aren't always accurate as they could be testing some very simple logic unit for say a radio chip, whereas it wouldn't directly translate to a high performance CPU. This was what Lisa Su said a few years back about the Foundries in general (think it was in reference to GF). I can't find the link. Been a long time. This is just another reason that going from 7nm to 6nm really doesn't mean anything. Especially when looking at the actual density increase being very marginal.


I don't see how Intel's initial failure has really anything to do with this topic of measurement. If you care to elaborate then please do.

If you are concerned about people not understanding, then that's why I mentioned LMC. It can be easily understood and abstracted enough for someone to comprehend. It's just measuring the CPU(L) Memory (M) and Interconnect/cache between CPU and memory(C)

[tgrech]

At the end of the day, a public naming scheme for tech is always marketing, there may have been times when this was ostensibly an actual measurement, but even for Intel it's long been a calculation whose result is intended to convey a comparative or marketing purpose, rather than a measurement that is practically useful to anyone on the technical side. The LMC approach is useful for making comparisons, but even then these are very regressive comparisons given the wide range of performance characteristics you would want to asses when comparing modern nodes, and the results are more of a loose guide than something you could use for practical detailed comparisons.

But regardless, this is a lot simpler and easier, and I don't think consumers are losing anything here, with Intel's previous naming scheme they had what was technically a measurement, but nothing else in the industry was being measured with the same ruler, and even if other people in the industry theoretically did use the same ruler, there are so many nuances in that calculation that the resulting comparisons would probably need a bucket of caveats just as they do now.

[NeverBackDown]

But that's the problem. Nobody is still using the same measurements. They just tweak it to make it sound better. Now that Intel is more closely similar they will just change their tune. It essentially makes the name change pointless. They'll just keep spinning it. It's an endless cycle.

There are better ways that can still benefit the consumer. That's really my point. The whole you can't beat them join them makes little sense when considering they were already equal and the confusion lies in the archaic naming scheme every company uses which is why they should adopt a new one as a whole. Heck even TSMC helped purpose the LMC.