GPUs outperform supercomputers at solving quantum mechanic equations

@mark noble,

Yes. I want the link.

0 Replies

@maxdancona,

I only ask - As I have to type it out.

https://www.youtube.com/watch?v=wZzHnZzm_58

@mark noble,

As I suspected, you are misunderstanding the Physics... In my opinion, this Ted Talk is misleading, so I will cut you some slack.

You make the claim that "It will be an instantaneous delivery of info/data over any distance". These are your words. Leo Kouwenhoven doesn't make this claim (and if he did he would cause a stir in the Quantum Mechanics community).

Modern Physics still understands that faster than light communication is impossible. Quantum Mechanics doesn't change this... it is a well known law of Quantum Mechanics that faster than light transfer of information never happens.

There is a difference between entanglement... which has been shown to happen at distances that would require faster than light speeds... and transfer of information.

I already gave you a link... but you can google "faster than light communication" yourself to see that there is no reputable scientific source that says this is possible given our current understanding.

This is a mathematical discussion, but it it a mathematical fact that you can't communicate at faster than light without breaking modern laws of physics.

3 Replies

@maxdancona,

So - 'QE' operates 'instantaneously' and delivers info/data 'instantaneously'?
Therefore a circuit founded on this - Operates 'instantaneously' 'Much faster' than 'c'.
Yes or no?

1 Reply

@parados,

parados wrote:

What supercomputers use GPUs?

Quote:

* In the Top 10, the No. 2 system, Tianhe-2, uses Intel Xeon Phi processors to speed up their computational rate. The No. 3 system Titan, the No. 8 system Piz Daint is using NVIDIA GPUs to accelerate computation.

* A total of 93 systems on the list are using accelerator/co-processor technology, down from 104 on November 2015. Sixty-seven (67) of these use NVIDIA chips, 26 systems with Intel Xeon Phi technology, three use ATI Radeon, and 2 use PEZY technology. Three systems use a combination of Nvidia and Intel Xeon Phi accelerators/co-processors.

* The average number of accelerator cores for these 94 systems is 76,000 cores/system.

http://www.top500.org/lists/2016/06/highlights/

0 Replies

@InfraBlue,

InfraBlue wrote:

Thanks, I hadn't considered obsolescence.

I was amazed the first time I saw my graphics card do ten hour BOINC workunits in less than a minute, over and over again.

For what it's worth, here is a graphics card vs supercomputer comparison (the second graphics card in each pair contains two GPU chips):

Code:

Cray 1 			0.2 GFLOPS

Cray XMP		0.9 GFLOPS

Cray 2			1.9 GFLOPS

Cray YMP		2.7 GFLOPS

Cray C90		15.2 GFLOPS

Cray T90		57.6 GFLOPS

Cray T3D		153.6 GFLOPS

ASCI Red (1997)		1,830.4 GFLOPS

ASCI Red (1999)		3,207.0 GFLOPS

ASCI White		12,288.0 GFLOPS





Radeon HD 3870		Nov 19, 2007		99.2 GFLOPS

Radeon HD 3870 X2	Jan 28, 2008		211.2 GFLOPS



Radeon HD 4870		Jun 25, 2008		240.0 GFLOPS

Radeon HD 4870 X2	Aug 12, 2008		480.0 GFLOPS



Radeon HD 5870		Sep 23, 2009		544 GFLOPS

Radeon HD 5970		Nov 18, 2009		928 GFLOPS



Radeon HD 6970		Dec 15, 2010		675 GFLOPS

Radeon HD 6990		Mar 8, 2011		1276.88 GFLOPS



Radeon HD 7970		Jan 9, 2012		947.2 GFLOPS

Radeon HD 7990		Apr 24, 2013		1894 GFLOPS

1 Reply

@maxdancona,

LC States 'twice' 'the exchange of info is 'instantaneous'.
Don't cut me 'slack'.
Prove otherwise?
And highlight Einsteins' fallacy too, please?

0 Replies

@maxdancona,

You see, Max - You defend deadend physics, wholeheartedly, because you must.
That's fine.
I have no time for 'going round in circles'. with 'theories' that do nought but stretch the imagination.
I want actual results.
And people like Leo are providing them.

1 Reply

@mark noble,

If you want to understand Quantum Engineering, you need to understand the math (as I consistently say on this forum). But I think I can simplify this one point with a simple example. This example will illustrate the point that you can have something operate instantantneously without gaining the ability to transfer information instantaneously.

Let's say you and I have a pair of magic quarters that have this one property. If you flip your quarter randomly (i.e. you can't cheat) and I flip my quarter (randomly) my quarter and your quarter will magically come up with opposite answers (i.e. if your is heads, mine will always be tails and if yours is tails mine will always be heads).

The rules of these magic quarters are that for the magic to work, the flip has to be random... there is no way for you to determine whether your quarter is heads or tails. Although once you see what your result is you instantaneously know what my quarter did.

Here is the question... using this pair of magic quarters with you in San Francisco and me in Boston... how can you send me a message that contains any information?

As this example illustrates... having instantaneous action at a distance doesn't mean that you can exchange any information at faster than the speed of light.

Again... anyone can google "faster than light communication" to see that reputable scientists agree that this breaks our currently understanding of the laws of Physics.

1 Reply

@mark noble,

Quote:

And people like Leo are providing them.

People like Leo are using what you are calling "deadend physics" to do really cool things. He is not breaking any laws of physics, nor is he claiming to (although I think his Ted talk is misleading hype). If you can show me where Leo is claiming to be able to transfer information at faster than light speeds.... I will change my mind.

0 Replies

@oralloy,

oralloy wrote:

InfraBlue wrote:
Thanks, I hadn't considered obsolescence.

I was amazed the first time I saw my graphics card do ten hour BOINC workunits in less than a minute, over and over again.

For what it's worth, here is a graphics card vs supercomputer comparison (the second graphics card in each pair contains two GPU chips):

Code:Cray 1 0.2 GFLOPS Cray XMP 0.9 GFLOPS Cray 2 1.9 GFLOPS Cray YMP 2.7 GFLOPS Cray C90 15.2 GFLOPS Cray T90 57.6 GFLOPS Cray T3D 153.6 GFLOPS ASCI Red (1997) 1,830.4 GFLOPS ASCI Red (1999) 3,207.0 GFLOPS ASCI White 12,288.0 GFLOPS Radeon HD 3870 Nov 19, 2007 99.2 GFLOPS Radeon HD 3870 X2 Jan 28, 2008 211.2 GFLOPS Radeon HD 4870 Jun 25, 2008 240.0 GFLOPS Radeon HD 4870 X2 Aug 12, 2008 480.0 GFLOPS Radeon HD 5870 Sep 23, 2009 544 GFLOPS Radeon HD 5970 Nov 18, 2009 928 GFLOPS Radeon HD 6970 Dec 15, 2010 675 GFLOPS Radeon HD 6990 Mar 8, 2011 1276.88 GFLOPS Radeon HD 7970 Jan 9, 2012 947.2 GFLOPS Radeon HD 7990 Apr 24, 2013 1894 GFLOPS

I focused on ATI/AMD cards because NVIDIA cripples non-graphics computing power on most of their cards that are marketed to gamers. That way they can charge a huge amount of money selling non-crippled cards to supercomputer builders.

However I kept thinking about that. ATI/AMD graphics cards do not have a lot of interconnects that allow one calculation to easily get data from an adjoining calculation. That works fine for some kinds of computations, and is broadly comparable to most of the Crays that I listed (vector processing also didn't allow for the freeflow of data between each calculation). But it is unlike modern supercomputers where data is easily exchanged between parallel calculations.

NVIDIA cards, crippled though they are, allow data to be exchanged between calculations, making them more like modern supercomputers than the ATI/AMD cards.

So I made a quick chart of the NVIDIA cards too. I decided to include supercomputer cards alongside game cards so the non-crippled versions could be seen. Any card called "GeForce GTX" is a graphics card marketed to gamers. Any card called "Tesla" is a very expensive card marketed to supercomputer companies. The "Titan" cards were marketed to gamers, but were VERY expensive.

Code:

GeForce GTX 480		March 26, 2010		168.12 GFLOPS

Tesla C2050		May 2010		515.2 GFLOPS



GeForce GTX 580		November 9, 2010	197.63 GFLOPS

GeForce GTX 590		March 24, 2011		311.04 GFLOPS	(double GPU)

Tesla M2090		June 2011		665.6 GFLOPS



GeForce GTX 680		March 22, 2012		128.77 GFLOPS

GeForce GTX 690		April 29, 2012		234.24 GFLOPS	(double GPU)

Tesla K20		Nov 2012		1175 GFLOPS



GeForce GTX Titan	February 21, 2013	1300 GFLOPS

Tesla K20X		July 2013		1312 GFLOPS

Tesla K40		Nov 2013		1680 GFLOPS

GeForce GTX Titan Black	February 18, 2014	1706.9 GFLOPS

GeForce GTX Titan Z	March 25, 2014		2707.2 GFLOPS	(double GPU)

Tesla K80		Jan 2015		2912 GFLOPS	(double GPU)



GeForce GTX 1080	May 27, 2016		257 GFLOPS

Tesla P100		coming soon		4700 GFLOPS

0 Replies

@maxdancona,

Let's introduce 'magic-quarters' into the debate, eh.

No.

1 Reply

@mark noble,

There is no debate here Mark. You simply misunderstand the science and I am trying to explain it to you.

1 Reply

@maxdancona,

I don't misunderstand, Max.
And I appreciate your kind gesture of tuition.

Let me ask you one question.
Does the moon rotate upon its own axis?

0 Replies

GPUs outperform supercomputers at solving quantum mechanic equations

Related Topics

Quick Links

My Account

able2know