427 GHz Graphene CPUs

[magneto]

Hi,

Not sure if you guys have seen this but it looks pretty promising:

http://gizmodo.com/scientists-just-figured-out-how-to-make-lightning-fast-1177727488

We should already have THz CPUs by now, it's embarrassing in this day and age when the software is so much ahead but bottlenecked by the hardware IMO.

I could live with 427 GHz for now, but still. More hardware-savvy people would know more about these and how they might turn out (Mark? )

Using Houdini at that speed would be a dream come true though.

I also heard about optical processors that are extremely fast. Not sure which one will prevail but it's interesting none the less.

Either way my next CPU will still not be faster than 5 GHz I think.

[Guest tar]

Using Houdini at that speed would be a dream come true though.

Looks nice - though you're already using a dream computer, 10-20-30 years ago the computer you are using is unbelievable! The dream hardware is always a dream

[magneto]

Good point. I am not that optimistic though

Look at this thread from 2002:

http://forums.guru3d.com/showthread.php?t=33377

CPU Timeline (hopefully)

July.............2002.....................2,500 Mhz........2.5 Ghz

January......2004.....................5,000 Mhz...........5 Ghz

July............2005...................10,000 Mhz.........10 Ghz

January.....2007...................20,000 Mhz.........20 Ghz

July...........2008...................40,000 Mhz.........40 Ghz

January....2010...................80,000 Mhz.........80 Ghz

July..........2011.................160,000 Mhz.......160 Ghz

January...2013.................320,000 Mhz.......320 Ghz

July.........2014.................640,000 Mhz.......640 Ghz

January..2016..............1,280,000 Mhz......1.28 Thz

How off that prediction was. Compared to 2002, it's not that impressive. Sure more RAM, newer GFX cards, etc but from 2.5 Ghz 11 years later, we are still at 3-4 Ghz. I hope I can use a Thz CPU before I go to the Underworld (not the movie)

[Guest tar]

Luckily computing is way more than just cpu speed; programs get better, new fields and areas open up all the time. Moores law is a pretty boring metric

[magneto]

Yeah software gets better but be limited by the hardware, and the overhead is offloaded onto the developers to make them faster using multi-threading, etc which is still shaky IMO.

I agree on Moore's "Law" though. I never understand why people even care about him

[WesleyE]

Having 427GHz transistors does not mean 427GHz cpu's per se. The current transistors that we have in our cpu's (silicon-germanium) are quite capable of switching at around 350 GHz at room temeprature. The problem with these speeds is that the packages are quite big, thus in need a lot of power. Imagine that almost every watt you send in comes out as heat it could get quite hot fast, which is why the graphene could come in. So theoretically it is possible to run those things at 427GHz, but it is not all that practical -yet.

Edited August 23, 2013 by WesleyE

[halfdan]

Having 427GHz transistors does not mean 427GHz cpu's per se. The current transistors that we have in our cpu's (silicon-germanium) are quite capable of switching at around 350 GHz at room temeprature. The problem with these speeds is that the packages are quite big, thus in need a lot of power. Imagine that almost every watt you send in comes out as heat it could get quite hot fast, which is why the graphene could come in. So theoretically it is possible to run those things at 427GHz, but it is not all that practical -yet.

It's not just that. With a clock cycle of 427 Ghz, the a single light pulse only gets about 0.7mm (5/64") between pulses. Good luck trying to design synchronous circuits with that

I'd be more interested in someone coming up with memory chips with this, since that's the biggest bottleneck right now. All the Ghz in the world won't help you if the CPU is just sitting idle waiting for main memory.

[magneto]

Optical CPUs should solve the heat and energy issues, right? Here another promising article from 2009:

http://gizmodo.com/5...-terahertz-cpus

Edited August 23, 2013 by magneto

[Skybar]

Everything new is probably very far off still. Right now and for many years to come, more speed will come in terms of shrinking and improving the current designs - which Intel is pushing quite hard.

I believe the big "revolution", if you will, will happen once they are physically not able to shrink the current technology anymore - which will happen in the not too distant future. Once we are there, they have to come up with something new to keep pushing - and anything thats "new" today will probably be more mature by then as well.

My point is that, as long as they can improve the current designs they will do just that, anything new wont magically appear and take its place. I guess its easier and cheaper for them to do what they know as well.

Its like petrol cars, we have technology to not use petrol but as long as we still have petrol left on earth that is what will be used, and improving the current cars. Once we are out of it though, and it will happen, things will start to change. Because then it has to change, now it doesnt.

[magneto]

I wonder when a breakthrough happens, if they will release them in small increments on purpose. For example if they manage Thz speeds, I believe they will first release 10Ghz, then 15Ghz next year, 20Ghz the next year, etc. That way they can milk it way longer, than jumping straight to Thz speeds.

[Guest tar]

I wonder when a breakthrough happens, if they will release them in small increments on purpose. For example if they manage Thz speeds, I believe they will first release 10Ghz, then 15Ghz next year, 20Ghz the next year, etc. That way they can milk it way longer, than jumping straight to Thz speeds.

Of course they will, if they can, as that's what business is all about. The question is, can they? That's what competition is all about.

[malexander]

The main issue with high clockspeeds is that power use quadruples with a doubling of the clockspeed. This can be reduced by shrinking the process size, but then leakage current increases and you have a smaller area producing more heat, which becomes difficult to cool. On the other hand, doubling the number of cores merely doubles the power draw. So this is why the megahertz race ended in the high 3GHz range, and the #cores and execution units began to increase instead.

An entirely new process is required for efficient high GHz operation, graphene being one of them. The silicon process advancement is slowing, as the diameter of a silicon atom is 0.25nm. 3D processes and other materials are being researched, and transitioning to a new process will reset the race once again. I think the next big thing is incorporating GPU-style streaming in CPUs using the same memory space. We're going to have to wait a bit on new process tech.

[magneto]

Thanks Mark. The heat output of current CPU model seems way too much like you said. Not sure if graphene solves it but I read optical CPUs are much cooler.