Whats all the hubbub about upcoming 7nm AMD and 10nm Intel

[_dawn_chorus_]

I keep seeing these numbers but I am not sure what the big deal is as far as real world performance.
Does it mean higher clock speeds are acheivable? Or lower temps so more cores at higher clock speeds?

Looking for more of an answer than "its faster". ..lol

 

[gamerk316]

I keep seeing these numbers but I am not sure what the big deal is as far as real world performance.
Does it mean higher clock speeds are acheivable? Or lower temps so more cores at higher clock speeds?

Looking for more of an answer than "its faster". ..lol

In a nutshell: A smaller processing node allows for more transistors to be put in the same die area, which will (typically) lead to increased performance due to higher clocks (due to reduced power draw) or simply more number crunching power (due to having more transistors).

https://www.maketecheasier.com/processors-process-size/

It should be noted that the terminology used to specify node size is arbitrary; Intel's nodes have typically been superior to the same node used by other manufactures. The real issue is that decreasing nodes is becoming VERY expensive, and that Intel has lost its node advantage for the first time anyone can remember.

 

[geofelt]

All of the above.
Most importantly is that smaller tech lets more chips come out of a given size wafer.
That lowers costs and boosts margins.

The increase is related to the square of the size.
A 10nm wafer can hold twice the number of circuits than a 14nm size(10x10 = 100 vs. 14 x 14 = 196)

 

[USAFRet]

I keep seeing these numbers but I am not sure what the big deal is as far as real world performance.
Does it mean higher clock speeds are acheivable? Or lower temps so more cores at higher clock speeds?

Looking for more of an answer than "its faster". ..lol

They can, in theory, pack more 'transistors' into the same space.
More transistors = able to do more work at the same time (per clock cycle).

Figure a schoolbus:
You can pack it with 20x 300 lb NFL players (30nm)
Or, you can pack it with 60x 10 year olds (20nm)
Or, you can pack it with 100x infants (7nm).

Each bus is going the same speed down the road, 55mph.
Which case gets you the most bodies at the end of the same distance trip?

 

[Eximo]

Process nodes are about how small the transistors (or other measurable features) can be made. Everyone uses a different measurement criteria, so the numbers aren't 100% comparable between companies.

For the manufacturers, it means more profit. Smaller chips equals more chips per wafer, and as long as yields are good, more money on sales. (Intel is having yield problems with their new 10nm process, If I recall they aimed for a 28% reduction in size, which is a bigger jump then they have made before)

When you make the transistors effectively smaller, it takes less voltage to flip their state. Lower voltage does mean lower power consumption, and this typically means that on high end parts, high clock speeds can be achieved for a given temperature envelope. You do run into some complications with capacitance, and they are just about at the limits of what traditional lithography can achieve in terms of size.

More cores is just a matter of using more silicon, but yes, since they run at a lower voltage you can get more cores into things like laptops.

TSMC's 7nm isn't quite as small as Intel's 10nm. But basically if AMD can get that to market first it will seem like a huge advantage. Their CPUs on the 14nm and 12nm processes are competitive already. That next jump will make a pretty big difference.

Intel already has 10nm chips in production, just that the yields are low, so they are probably losing money. Very likely why they haven't gone for mass adoption and keep improving their 14nm lineup.

 

[jimmysmitty]

I keep seeing these numbers but I am not sure what the big deal is as far as real world performance.
Does it mean higher clock speeds are acheivable? Or lower temps so more cores at higher clock speeds?

Looking for more of an answer than "its faster". ..lol

In a nutshell: A smaller processing node allows for more transistors to be put in the same die area, which will (typically) lead to increased performance due to higher clocks (due to reduced power draw) or simply more number crunching power (due to having more transistors).

https://www.maketecheasier.com/processors-process-size/

It should be noted that the terminology used to specify node size is arbitrary; Intel's nodes have typically been superior to the same node used by other manufactures. The real issue is that decreasing nodes is becoming VERY expensive, and that Intel has lost its node advantage for the first time anyone can remember.

Unless the node sucks. Intels 90nm node was vastly inferior to its 65nm node and almost to its 130nm node. I would say it was just Netburst but they had Netburst on 65nm which was better power wise.

Similar to Phenom I on 65nm and then Phenom II on 45nm. It might even be that some uArchs just don't like certain node sizes.

And yea its costly. FAB 42 was originally designed and intended for 10nm and had an inistal cost of $1.5 Billion. Intel is now pumping an additional $7 billion to convert it over to 7nm. Thats more than some large companies gross revenue. And it doesn't include all the R&D costs to research that node size or develop equipment for those node sizes.

I don't eve want to think what, if anyone pushes for it, 5nm will cost to do.

All of the above.
Most importantly is that smaller tech lets more chips come out of a given size wafer.
That lowers costs and boosts margins.

The increase is related to the square of the size.
A 10nm wafer can hold twice the number of circuits than a 14nm size(10x10 = 100 vs. 14 x 14 = 196)

Not always as yields start lower for new process tech. In fact thats whats holding Intels 10nm up, they can't get the yields to a profitable margin.

Process nodes are about how small the transistors (or other measurable features) can be made. Everyone uses a different measurement criteria, so the numbers aren't 100% comparable between companies.

For the manufacturers, it means more profit. Smaller chips equals more chips per wafer, and as long as yields are good, more money on sales. (Intel is having yield problems with their new 10nm process, If I recall they aimed for a 28% reduction in size, which is a bigger jump then they have made before)

When you make the transistors effectively smaller, it takes less voltage to flip their state. Lower voltage does mean lower power consumption, and this typically means that on high end parts, high clock speeds can be achieved for a given temperature envelope. You do run into some complications with capacitance, and they are just about at the limits of what traditional lithography can achieve in terms of size.

More cores is just a matter of using more silicon, but yes, since they run at a lower voltage you can get more cores into things like laptops.

TSMC's 7nm isn't quite as small as Intel's 10nm. But basically if AMD can get that to market first it will seem like a huge advantage. Their CPUs on the 14nm and 12nm processes are competitive already. That next jump will make a pretty big difference.

Intel already has 10nm chips in production, just that the yields are low, so they are probably losing money. Very likely why they haven't gone for mass adoption and keep improving their 14nm lineup.

That and it seems like they might be trying to skip 10nm. FAB 42 is being retooled for 7nm but was originally built for 10nm.

 

[Eximo]

That and it seems like they might be trying to skip 10nm. FAB 42 is being retooled for 7nm but was originally built for 10nm.

That seems like it would make sense. They never moved main production to 10nm, so most of the fabs are still running 14nm. To stay competitive they'll have to start considering their next process shrink otherwise they will be stuck on this node for even longer.

I don't recall which process node Intel was targeting for EUV (probably 7nm), but that is a whole other change everyone will have to make as well.