AMD's Future Chips & SoC's: News, Info & Rumours.

Page 41 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.


https://www.anandtech.com/show/11850/intel-displays-10nm-wafer-commits-to-10nm-falcon-mesa-fpgas

http://fudzilla.com/news/processors/45091-intel-10nm-is-better-than-everyone-thinks
 


This is self fulfilling prophecy for everything known, and unknown is physics! As with anything, it's only a matter of time before what was once thought impossible is made reality.
 


The laws of physics don't stop working when someone desires to ignore them.
 


What is intelligence without creativity to drive it? Useless. Science is constantly evolving like our understanding of physics. Don't create barriers where none exist.That chart is just a timeline showing the relatively recent lack of innovation by all those blue squares! Don't worry though, the times are a changin!
 


Science evolves, the laws of physics are immutable. The electromagnetic and quantum laws that drive the behavior and properties of silicon have not changed in last 100 years and will not do in next 100.

That frequency wall is a well-known physical limit.
 


My last comment on the subject. As I expressed previously nothing about physics, known or unknown, has changed. Only our understanding of it.
 


If only there were CPUs that were over 6Ghz... Oh, wait a minute, there are!

Caveated as you will, but it's already been done before. You can indeed push silicon beyond 6Ghz, so I have no idea what "wall" you're talking about here. You sure it's not just you own imagination? As far as I'm aware, meeting the conditions for quick gate switching and low leakage is only a matter of resistance, temperature and material composition (in a very simplified way).

But well, let's see. Just because you can do it theoretically, doesn't mean you will do it commercially.
 
Law's of physics stop working when you shrink something small enough. Who know's whats ahead. Quantum Computing is here and graphene chips.. Not sure which way it's gonna go from here.

There's still plenty of room for improvement though... an by the looks of things Intel will most definitely have their hands full in the very near future. Their about to forfeit their process lead.

An just look how close Ryzen got in it's first generation on an inferior process no less...
With some tweaks an a shrink it looks like it will be Intel playing catch up as their Core uArch is already being pushed beyond it's limits an their process lead well that's old news it doesn't exist anymore.

Actually their being attacked on every front an it's about time. An I think the market agrees with me as the big players all start to fall in behind AMD... It's seem's everyone is pro choice.
The cards are dealt an the die is cast roll on 7nm.

Happy New Year Everyone :)
 


Personally i don't see 6ghz in a lab with exotic cooling as being impossible if the design is made for high frequency's over IPC or "Performance" per cycle.
This is from 2010 and they got 5.2ghz. Heck Amd was able to pull of a 4.7ghz FX chip that's AMD and on Global foundries even haha, so yeah i'm betting my money that its possible but not likely and for sure not a product consumers will use at least not in our personal machines.
https://www.engadget.com/2010/09/06/ibm-claims-worlds-fastest-processor-with-5-2ghz-z196/


I'm hoping Amd finds some money for K12 or Intel gives up on X86 only and builds their own ARM high performance CPU.
 
i still wasnt expecting a 5.0 GHz chip...and yet here we are.
all this has to be done at room temperature or else there is condensation...so no cheating with ln2 😀

nvidia seem to be doing pretty well with their new Volta ..it's just a gargantuan sized chip which works.

sure there are limits. but the smaller nodes seem to consistently push the frequency limits higher. anyway. not my problem :) just nerdism curiosity.

i read that glofo's 7nm is really an 'equivalent' of intel's 10nm...whatever that means. and i have a feeling that all manufacturer's / companies will be maximising the 10nm/'7nm' nodes lifespan. 2020 at the earliest for 5nm as far as i have read on it...which probably means a bit later than that.

its only 2014 when intel introduced 14nm cpu's just 3 ish years ago. not long really.

should be more interested in the graphics chips for raw compute. they are monsters....massively parallel processing.

why can't cpu's hurry up with the parallelism. certain operations and instructions just arent suited for parallelism ...but that's all there is to increase compute.

some scientists are even experimenting with sub-1nm 'transistors' which is getting down to the atomic scale of larger elements :lol: 360 picometers i read.

------
moar coars!!!! more ARM clusters...more more more
 


LOL. This stuff is well-known. The frequency wall is one of the several walls that were identified in microarchitecture many time ago.

https://books.google.es/books?id=48eOVJYAe9oC&pg=PA348&lpg=PA348&dq=%22Frequency+wall%22+cpu&source=bl&ots=CHteanLD3Q&sig=Bxot1Ffb1NmdPmORuwQLtgOVoqE&hl=en&sa=X&ved=0ahUKEwjQid3_tZHYAhUCNxQKHerYDoEQ6AEIRjAI#v=onepage&q=%22Frequency%20wall%22&f=false

http://slideplayer.com/slide/5189974/16/images/4/The+Mystifying+Walls+Power+Wall+Memory+Wall+Frequency+Wall.jpg

But let us ignore the laws of physics and believe in those magic 6--7GHz CPUs coming on 7nm node. :sarcastic:
 


False. The laws continue to work. Precisely any modern CPU is fabricated thanks to engineers and scientists applying those laws.
 


I predicted that Intel would hit 5GHz with 14nm. My explanation and graphs can be found in ancient threads here in the forums.



You can push the frequency higher and reach the limit asymptotitcally. The nature of the wall is easily seen in the figure that I provided above. There is no reason to negate reality.
 


The frequency wall affects to all the companies, because all the companies are bound by the same laws of physics. And the graph clearly shows that the wall affects to all companies including those green circles and those red triangles.

Negating evidence doesn't make it go away. Those imagined 6GHz RyZen chips aren't coming.
 

TCRoLlB.png


The frequency wall affects to all the companies
Click to expand...
Maybe you couldn't see, so let me blow up the image for you! That's IBM shattering the Intel Wall created on that graph you linked. If you notice IBM isn't behind the wall it has passed it significantly. Intel design limits don't apply to IBM.

Edit: As of mid-2013, the highest clock rate on a production processor is the IBM zEC12, clocked at 5.5 GHz, which was released in August 2012.

2nd Edit: I've gave my estimations for 7nm SOC and 7nm HPC.
14nm 1.3 Normalized Frequency 7nm Soc 1.7 Normalized Frequency
1.3 to 1.7 is 30.7%@less power consumption
30.7% X 3.6GHz = 4.7052GHz
30.7% X 4.1GHz = 5.3587GHz
7nmHPC
1.7 to 2.1 is 23.5% @~2.5x power consumption
23.5% X 4.7052 = 5.810922GHz
23.5% X 5.3597GHz = 6.6179945GHz

AMD and IBM will fall within these lines based on the normalized frequency and power chart if designs permit. 7nmSOC ~5GHz for AMD and 7nmHPC ~6GHz for IBM.
 


I don't think so Juan my statement is completely accurate !

To the best of my understanding it goes something like this,
If you shrink them small enough the gates in the chip stop working an energy starts to just to pass through closed gates (It's like it teleport's to the other side)...Really strange behaviour for sure, hence a breakdown in the laws of physics as we know it. Quantum Tunnelling it's called !

Hello this is the reason were gonna hit a wall with shrinking nodes (even if the tech makes it possible)... eventually physics just will not behave as it should as we enter the quantum realm...

By the way didn't you also claim that Epyc was dreadful, An predict it would not make it as a server chip in the data centre's, an never receive any design wins due to it's really high latency ?
 


Nobody said anything about 6GHz Ryzens's, but IBM is something else.

On current GloFo 14 nm technology.
AMD 1800x maximum frequency 4.1 4.2 GHz (note AMD uses 14LPP)
IBM Z14stock frequency 5.2 GHz (note IBM uses 14 HP)

 


Documents from 2006 and a document talking from the Intel POV. Right...

Ok, I'll just leave it here.
 


You are wrong. There is no frequency wall here. This figure actually shows we have hit a POWER WALL (actually it should be named cooling wall). Frequency wall is well beyond current processor frequencies and nobody in micro architecture considers it a wall TODAY. The real wall is POWER.
Why is IBM getting higher frequencies than anybody else (even with inferior process)? Because they are willing to spend more on a cooling solution. IBM usually designs for TDPs in the 200-400W range. IBM MCM's are famous for disipating beyond 2KW.

In fact, with the proper cooling, it is possible to reach frequencies well beyond 8 GHz, almost 9 GHz.
https://valid.x86.fr/lpza4n
This is proof that, so far frequency, is only limited by power, and therefore cooling. We have yet to reach any frequency wall.

 


200-400W range? this is achieved with air coolers? or air cooled rads?
interesting to hear. the obsession with low tdp parts presumeably is for the laptop, tablet and phones??

...and yes ultimately cheaper transistors is what moore's law is about isnt it?
----
https://www.cnet.com/news/ibm-ships-5-2ghz-chip-its-fastest-yet/

so all we need is a gigantic heat spreader; no worries!! :lol:
 


Pretty much this right here. Provided sufficient voltages, you can keep increasing clocks. The problem is that power draw makes it impossible to keep most parts cool, especially at the consumer level. High end parts can break 5GHz, but those are typically limited production and have more advanced cooling solutions to go with them.

So basically, there's a "reasonableness" wall at ~4.5GHz or so, where power draw (Heat) increases almost exponentially, making it not worth it in most cases to go beyond. It *can* be done at a technical level, but the performance typically isn't worth the cost.

So, lets stop going over the finer points of the English dictionary and move on, kay?
 


Anyone can check that IBM also hitt the frequency wall. Until 2005 IBM was increasing frequencies at a faster pace. Then the GHz race stopped almost abruptly when they did hit the wall.



Quantum tunneling is a physical phenomena that follows from the laws of physics. The physical laws that drive quantum tunneling have been known during 100 years.

I didn't say that EPYC wouldn't get some desing win, even Bulldozer and Itanium got some desing wins, and both were terrible designs. What I said is that EPYC would be massively rejected, with customer choosing other CPUs. I predicted EPYC would be crushed by both sides: ARM and Intel. And data just demonstrates it. Datacenters are massively purchasing Xeon, Cavium and Qualcomm, and rejecting EPYC. Did you read the links that I gave? What server CPU chose Microsoft for powering most of its datacenters? Also latency is only one of the problems that EPYC has.





The IBM Z series also shows the frequency wall. Until 2008 the increase in clocks was very fast. Then the GHz race stopped almost abruptly and even clocks reduced a bit when IBM hit the wall

05-ghz.jpg




The laws of physics and the atomic structure of silicon have not changed since 2006. Also the same laws apply to Intel than to AMD, IBM, or anything else.



The power wall and the frequency wall are related, but the concepts wouldn't be confused, because the two walls aren't the same. That is the reason why the power wall and the frequency wall are discussed by separate.

https://books.google.es/books?id=48eOVJYAe9oC&pg=PA348&lpg=PA348&dq=%22Frequency+wall%22+cpu&source=bl&ots=CHteanLD3Q&sig=Bxot1Ffb1NmdPmORuwQLtgOVoqE&hl=en&sa=X&ved=0ahUKEwjQid3_tZHYAhUCNxQKHerYDoEQ6AEIRjAI#v=onepage&q=%22Frequency%20wall%22&f=false

The+Mystifying+Walls+Power+Wall+Memory+Wall+Frequency+Wall.jpg


In fact, it is the power wall which has an incorrect name, because the problem is not on the amount of power, but in the power density. One can double the power if the area is doubled. The problem is when one double the power maintaining the area. Or alternatively when one maintains the power whereas reducing the area to one half with a die shrink.

We know that with exotic cooling as LN2 we can go above 5GHz. That doesn't disprove the existence of a frequency wall, because the limit takes different values depending of the cooling used: air, liquid, oil, LN2,...

The wall is around 6GHz on air and around 9GHz on LN2.
 
As a Physicist I can point this out:

CONSERVATION OF MASS AND ENERGY
Albert Einstein introduced his famous equation E = mc2 in 1905 journal submission titled, "On the Electrodynamics of Moving Bodies." The paper presented his theory of special relativity, based on two postulates:

Principle of relativity: The laws of physics are the same for all inertial reference frames.
Principle of constancy of the speed of light: Light always propagates through a vacuum at a definite velocity, which is independent of the state of motion of the emitting body.
The first principle simply says that the laws of physics apply equally to everyone in all situations. The second principle is the more important one. It stipulates that the speed of light in a vacuum is constant. Unlike all other forms of motion, it is not measured differently for observers in different inertial frames of reference.

We also know that Like in other fields of science, new laws of physics build on or modify existing laws and theoretical research. Albert Einstein's theory of relativity, which he developed in the early 1900s, builds on the theories first developed more than 200 years earlier by Sir Isaac Newton. I do think that the laws of physics as we know it will stay the way they are for a while until we discover new worlds or get to understand better our expanding universe.
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom