• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Intel 14nm+... compared to AMD's TSMC 7nm chips under microscope

llien

Member
German hardware overclocker and hacker, der8auer, has decided to see how production level silicon compares one to another, and he put it to the test. He decided to use Intel's Core i9-10900K processor and compare it to AMD's Ryzen 9 3950X under a scanning electron microscope (SEM).





To get a fair comparison as possible, he used the L2 cache part of both processors as they are always the best representatives of a node. The logic portion of the chip is different for every chip, so it would be difficult to compare two different microarchitectures. That is the reason level two cache is used to get a fair comparison. The results? Well, the Intel chip features transistors with a gate width of 24 nm, while the AMD chip has a gate width of 22 nm. (lower is better obviously)
TPU

In other words, although Intel is behind, it's nowhere "two times" behind as it might seem based on the node names.
Which makes AMD's lead even more impressive.
 

D.Final

Banned
German hardware overclocker and hacker, der8auer, has decided to see how production level silicon compares one to another, and he put it to the test. He decided to use Intel's Core i9-10900K processor and compare it to AMD's Ryzen 9 3950X under a scanning electron microscope (SEM).





To get a fair comparison as possible, he used the L2 cache part of both processors as they are always the best representatives of a node. The logic portion of the chip is different for every chip, so it would be difficult to compare two different microarchitectures. That is the reason level two cache is used to get a fair comparison. The results? Well, the Intel chip features transistors with a gate width of 24 nm, while the AMD chip has a gate width of 22 nm. (lower is better obviously)
TPU

In other words, although Intel is behind, it's nowhere "two times" behind as it might seem based on the node names.
Which makes AMD's lead even more impressive.


And this is the basic of the nexgen.
Right?
Or maybe I'm confusing :messenger_grinning_sweat:
 

Dr.D00p

Gold Member
All I take from this is just how good Intel's refined & perfected 14nm process is, the fact that it can keep up with a modern 7nm die like Ryzen 2 and still surpass it in some areas, shows that jumping to a smaller node doesn't deliver the performance up tick that the headline figure suggests.

People who like to throw 'LOL..still on 14nm!!!' at Intel, really are missing the point.
 

DonMigs85

Member
Mm, this is Zen 2.
Next gen consoles will have Zen 3, on a somewhat better 7nm.
No, the consoles use Zen 2, which first came out last year. Zen 3 is what's coming out on PC later this year. Zen+ is what came out in 2018 (Ryzen 2600, 2700X, etc)
 

JohnnyFootball

GerAlt-Right. Ciriously.
All I take from this is just how good Intel's refined & perfected 14nm process is, the fact that it can keep up with a modern 7nm die like Ryzen 2 and still surpass it in some areas, shows that jumping to a smaller node doesn't deliver the performance up tick that the headline figure suggests.

People who like to throw 'LOL..still on 14nm!!!' at Intel, really are missing the point.
Intel has gotten mileage out of 14nm due to necessity and it is impressive, but they are at the point where they have pretty much maxed it out as far as what they can get out of it.

10nm has been one of the biggest epic fails in probably the history of CPUs. It's actually sad at this point.

Intel is far from in trouble, but they are extremely unexciting and are paying the price for all those years of stagnation.
 
Last edited:

sn0man

Member
OP: good post. Interesting that they were able to do this.

All I take from this is just how good Intel's refined & perfected 14nm process is, the fact that it can keep up with a modern 7nm die like Ryzen 2 and still surpass it in some areas, shows that jumping to a smaller node doesn't deliver the performance up tick that the headline figure suggests.

People who like to throw 'LOL..still on 14nm!!!' at Intel, really are missing the point.
pro intel thought: It also reveals, only partially, that marketing of the node size is coming into play on some level.

pro amd thought: though this portion of the processor is much closer in size, there may be other portions of the processor that are definitively different in size.

My two cents: This article suggests it would be impractical to measure every section of a cpu at this level to get an “average” node measurement or a node ground truth so to speak. One particular measurement tells us the AMD node may be a bit of marketing but we don’t have an accurate picture
 

Silver Wattle

Gold Member
Gate width is a bad metric to compare node advancement, it barely changes when moving to a newer node these days.

While these are not much different, TSMC's node is still much denser compared to Intel's - TSMC's 7 nm produces chips with a transistor density around 90 MT/mm² (million transistors per square millimeter), which is comparable in density to Intel's 10 nm node used on recent mobile processors
From the article you linked OP.
 
Last edited:
Huh. Honestly did not expect intel to be so close.

still, transistor size is one thing, power draw is another ; so lol 14nm++++++ is still on point.
 

diffusionx

Gold Member
All I take from this is just how good Intel's refined & perfected 14nm process is, the fact that it can keep up with a modern 7nm die like Ryzen 2 and still surpass it in some areas, shows that jumping to a smaller node doesn't deliver the performance up tick that the headline figure suggests.

People who like to throw 'LOL..still on 14nm!!!' at Intel, really are missing the point.

It's true that this stuff has become marketing hype somewhat, and it's not like Intel's processes have stayed still, but also remember that Intel has been working on a 10nm process and it has been a total clusterfuck that just led to a massive shakeup at the company. So it's still ok to laugh at them.
 

LordOfChaos

Member
Intel of all companies should be leading a shift to using MTr/mm², which while not perfect (sometimes you want to give up density for clock limiting logic or hotspotting), is at least something real, and something they can objectively show they're not that behind in.

An 11 billion transistor chip could have a single feature at 3nm length and marketing would call that a 3nm node, that's how this works right now.

That said, TSMC is shipping 5nm in under a month in the iPhone and iPad and is also objectively ahead of Intel 10nm too.
 
Last edited:
I need to watch a documentary about how cpu are made.
anyone knows one?
In english or French.

It's not a huge long video but I feel like short and sweet is sometimes better. This video is nice and informative. Sometimes the human race can feel like a sh*t storm but when you see magical stuff like this it really makes you boggle at the kind of stuff us silly monkey's have figure dout.

 
Last edited:
Top Bottom