Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

5nm is only a marketing term and has no relation to any actual phyiscal feature

twitter.com

Post reply on HN
Print view View original

31–39 of 39 posts

Re: 5nm is only a marketing term and has no relation to any actual phyiscal feature

#31

Post by staticassertion »

duskwuff wrote: post #29
staticassertion wrote: post #25

Earlier quoted context omitted.

Because you're advertising a product in a way that's deceitful, obviously. If I sell you a CPU and I say "yeah it's 4.5ghz", and it turns out it runs at 1.6ghz, it should be illegal to say "ghz is just a brand, it's marketing".

Semiconductor processes aren't a feature that's generally advertised to end users. I've never heard a mobile phone or computer manufacturer describing their devices as having a "5 nm processor", for example. The clients who are seeing these descriptions (like, purchasing managers in electronics manufacturers) are generally aware of these issues.

So it's just marketing and therefor it's ok to lie, because it's not marketing and people who see it will know it's a lie?

Re: 5nm is only a marketing term and has no relation to any actual phyiscal feature

#32

Post by buu700 »

anyfoo wrote: post #26

Slight (and somewhat related, even) throwback to when (desktop) CPUs' primarily marketed performance indicator was their core frequency (i.e. "how many MHz a CPU has"), which also got divorced from reality over time. First bus and core frequency got separated, with bus frequency growing significantly slower and getting advertised much less, then the designation got at times decoupled entirely from frequency. For inst…

Goodhart's law strikes again.

Re: 5nm is only a marketing term and has no relation to any actual phyiscal feature

#34

Post by sylware »

wmf wrote: post #27
phkahler wrote: post #19

But EUV is a real thing and cleanly divides the haves and the have nots. While Intel 7 may have similar density to TSMC 7, there is a huge difference in the process and cost of making them. Intel will not go beyond that without EUV, and neither will anyone else.

To be clear, Intel 7 and TSMC N7 are both non-EUV processes. Later this year Intel 4 and TSMC N4, both using EUV, may be available.

I wonder which EUV steppers they bought from ASML, the 100 millions ones or the new half of billion ones (don't think those EUV steppers are available yet).

Re: 5nm is only a marketing term and has no relation to any actual phyiscal feature

#35

Post by netizen-936824 »

recursive wrote: post #15
clintonb wrote: post #12

Earlier quoted context omitted.

> For one thing, it makes it harder for me to understand when it will be necessary for new processors to use exotic materials. If you have a need to understand this, shouldn't you be close enough to the field to discern marketing speak from the underlying technical details?

If you're going to devise some marketing-speak, shouldn't you be able to do it in a way that doesn't just happen to be an actual lie?

Glad to know others feel the same way as me about marketing

Re: 5nm is only a marketing term and has no relation to any actual phyiscal feature

#36

Post by wmf »

sylware wrote: post #34
wmf wrote: post #27

Earlier quoted context omitted.

To be clear, Intel 7 and TSMC N7 are both non-EUV processes. Later this year Intel 4 and TSMC N4, both using EUV, may be available.

I wonder which EUV steppers they bought from ASML, the 100 millions ones or the new half of billion ones (don't think those EUV steppers are available yet).

I think they ordered regular ones that are already installed and they also have some high-NA machines coming later.

Re: 5nm is only a marketing term and has no relation to any actual phyiscal feature

#37

Post by rstuart4133 »

wolf550e wrote: post #18

Some technical people plain refuse to read technical sites about hardware and prefer to read marketing aimed at completely non technical people. Read https://en.wikichip.org/wiki/5_nm_lithography_process

I love this from that page:

> The term "5 nm" is simply a commercial name for a generation of a certain size and its technology, and does not represent any geometry of the transistor.

And I learnt something new: it seems like MTr/mm² is the metric in use now. I wonder if marketing will find a way to co-opt that?

Re: 5nm is only a marketing term and has no relation to any actual phyiscal feature

#38

Post by wolf550e »

rstuart4133 wrote: post #37
wolf550e wrote: post #18

Some technical people plain refuse to read technical sites about hardware and prefer to read marketing aimed at completely non technical people. Read https://en.wikichip.org/wiki/5_nm_lithography_process

I love this from that page: > The term "5 nm" is simply a commercial name for a generation of a certain size and its technology, and does not represent any geometry of the transistor. And I learnt something new: it seems like MTr/mm² is the metric in use now. I wonder if marketing will find a way to co-opt that?

The problem with "density" is that designers use power optimized transistors for some things and speed optimized transistors for other things, on the same manufacturing process. Also, the quality of standard cell libraries matters, kinda like the quality of the compiler affects the SPEC score of a chip.

Re: 5nm is only a marketing term and has no relation to any actual phyiscal feature

#39

Post by sylware »

wmf wrote: post #36
sylware wrote: post #34

Earlier quoted context omitted.

I wonder which EUV steppers they bought from ASML, the 100 millions ones or the new half of billion ones (don't think those EUV steppers are available yet).

I think they ordered regular ones that are already installed and they also have some high-NA machines coming later.

It seems, from random stuff I read on internet, that even with current ASML EUV steppers, there is a lot of transistor size reduction possible. And with the high-NA improvement, we can expect even more. This is very good news for the 2/3 next generations of CPU/GPUs. We can "reasonably" expect "really" more, namely significantly faster/less power hungry of what we have now. But it seems we are missing foundries to scale for the worldwide demand, and many video game AAA studios (which will have the decency to provide builds for linux based OSes, meh). On the mobile front maybe some new radio modulation beyond 5G in order to properly use chips based on those new processes.
Post reply on HN