Getting Exaflop computers by 2020:

Originally posted by DeepThought
Humans have built computers. How many computers have built a human?

Its not 2020 yet, and the article suggests that even by 2020 we won't have a super computer with the raw power of one brain. How many humans have built a computer? Zero. It took a civilization and thousands of years.

A skilled go player will still beat the crap out of a computer, provided the board is large enough.
Not for long. Its hard to predict, but I guess that within 10 years or so the best Go player will be a computer.

Originally posted by twhitehead
Its not 2020 yet, and the article suggests that even by 2020 we won't have a super computer with the raw power of one brain. How many humans have built a computer? Zero. It took a civilization and thousands of years.

[b]A skilled go player will still beat the crap out of a computer, provided the board is large enough.
Not for long. Its hard to predict, but I guess that within 10 years or so the best Go player will be a computer.[/b]

My guess is we are just not getting go programs to the best computers. If you want to program a Cray supercomputer to play go they would probably just laugh at you. Unless you wanted to invest 100 mil and buy your own! THAT would be awesome, our own computer to (A, beat the crap out of those putzes using engines on our own site and (B, to see if a Cray super comp can beat the best go players.

Anyone want to donate to the cause🙂

Originally posted by sonhouse
Anyone want to donate to the cause🙂

It's the sort of project that is not urgent and therefore the cost effective solution is to throw time at it. Wait 5 years and you will have a PC on your desk top that is equivalent to the Cray supper computer of today.
Wait 10 years and your cell phone will have the equivalent processing power.

You are correct that we are currently not throwing computing power at the problem. The top Go software runs on a PC.
We are also not yet throwing much programming power at the problem either, AI development is still very much in its infancy.

Originally posted by twhitehead
It's the sort of project that is not urgent and therefore the cost effective solution is to throw time at it. Wait 5 years and you will have a PC on your desk top that is equivalent to the Cray supper computer of today.
Wait 10 years and your cell phone will have the equivalent processing power.

You are correct that we are currently not throwing compu ...[text shortened]... much programming power at the problem either, AI development is still very much in its infancy.

well don't forget about National Security issues...It's my humble opinion that our governments will decide that the general public doesn't need supercomputers at their fingertips. The computers makers will stop offering more powerful machines to the public....better get yours while you still can....

Originally posted by woadman
well don't forget about National Security issues...It's my humble opinion that our governments will decide that the general public doesn't need supercomputers at their fingertips. The computers makers will stop offering more powerful machines to the public....better get yours while you still can....

While there may be a problem with quantum machines being able to crack asymmetric key encryption. In terms of off the shelf machines, an Intel I7 with 6 cores running at 4GHz and AVX2 registers can do 8 single precision multiply-adds per clock cycles, so the upper bound on it's speed is 8*4*6 = 212 Gigaflops. Although it'll never hit that due to memory latency. If you really want to spend money you can get an Intel Xeon Phi with 52 cores (as I remember) for ~$2,000 and that will hit a teraflop double precision.

Assuming Moores law, doubling the transistor count every year, continues to apply in a decade you can expect expect retail machines to be around 200 teraflops single precision. With things like a Xeon Phi giving of the order of a petaflop double precision.

This is still not enough to break a decent cipher with a brute force attack. By the time quantum machines are available off the shelf someone will have found an alternative to the factorization problem to make banking and so forth secure. Governments have no interest in stopping Intel, Arm, AMD, and co. from making their profits.

That last statement partially depends on whether NP != P. If it turns out NP = P (which seems unlikely) then BQP is likely to the same as P (the relationship between NP and BQP is unknown). If P = NP then there might be a problem as the key encryption decryption algorithm would become computationally intense.

P = The set of problems which can be solved in polynomial time on a Turing Machine. Meaning that the time to solve the algorithm grows as some polynomial function of the size of the input.

NP = The set of problems which can be solved in polynomial time on a Non-deterministic Turing machine. Basically a machine with an infinite number of processing cores.

BQP = The set of problems which can be solved in polynomial time on a quantum computer with bounded error (the probability it gets it wrong is no more than 1/3). The requirement is that checking the result can be done in polynomial time too.

Originally posted by DeepThought
While there may be a problem with quantum machines being able to crack asymmetric key encryption. In terms of off the shelf machines, an Intel I7 with 6 cores running at 4GHz and AVX2 registers can do 8 single precision multiply-adds per clock cycles, so the upper bound on it's speed is 8*4*6 = 212 Gigaflops. Although it'll never hit that due to memory ...[text shortened]... more than 1/3). The requirement is that checking the result can be done in polynomial time too.

With a quantum computer, the part I don't get is how does the answer pop up? I mean don't you have effectively a sea of answers crowding each other for attention? Like getting the factors of a large number, you input the numbers and the answer pops up but how is that answer separated from the jillians already present inherent in the superpositions of say, hundreds of qubits?

Originally posted by sonhouse
With a quantum computer, the part I don't get is how does the answer pop up? I mean don't you have effectively a sea of answers crowding each other for attention? Like getting the factors of a large number, you input the numbers and the answer pops up but how is that answer separated from the jillians already present inherent in the superpositions of say, hundreds of qubits?

I don't know enough about quantum computers. It only started being included as standard in undergraduate degrees after I finished. It's something I've been meaning to read up on anyway so I'll do some reading and if I find the answer to that question I'll post it if no one else does in the meantime.

Originally posted by woadman
well don't forget about National Security issues...It's my humble opinion that our governments will decide that the general public doesn't need supercomputers at their fingertips. The computers makers will stop offering more powerful machines to the public....better get yours while you still can....

The computer you have now is already equivalent to the super computers of 10 years ago. The government didn't stop you having it.

Originally posted by DeepThought
Assuming Moores law, doubling the transistor count every year,..

Performance actually doubles at closer to 18 months, and transistor counts approximately every two years.
http://en.wikipedia.org/wiki/Moore%27s_law

Originally posted by twhitehead
Performance actually doubles at closer to 18 months, and transistor counts approximately every two years.
http://en.wikipedia.org/wiki/Moore%27s_law

Yeah, I didn't check, I was just going by Moore's original statement.

Originally posted by DeepThought
Yeah, I didn't check, I was just going by Moore's original statement.

Although I believe the 'law' still holds, there has been a noticeable shift away from megahertz to reducing power consumption.
Also, although hard disc sizes have been expanding at a Moore's law type rate their performance has lagged behind making them the main bottle neck in PCs today. I recently upgraded to SSDs and it makes a massive difference.

There has also been a significant reduction in processing need. It is less beneficial to invest in newer hardware every two years, and many people wait up to 10 years or more before upgrading. Instead we are putting a significant amount of our investment into faster internet.

I don't have actual figures, but Google does a lot of parallel computing tasks and they probably have the largest supper computer in the world, its just not labelled as such.

Another contender is Folding@home which averaged 20.7 native petaFLOPS as of June 2014 according to Wikipedia

http://en.wikipedia.org/wiki/FLOPS

Originally posted by twhitehead
Although I believe the 'law' still holds, there has been a noticeable shift away from megahertz to reducing power consumption.
Also, although hard disc sizes have been expanding at a Moore's law type rate their performance has lagged behind making them the main bottle neck in PCs today. I recently upgraded to SSDs and it makes a massive difference.

T ...[text shortened]... .7 native petaFLOPS as of June 2014 according to Wikipedia

http://en.wikipedia.org/wiki/FLOPS

Right now the record is in the hands and bragging rights of the Chinese, reaching 43 petaflops. The next goal is 1 EXAflop, 1000 petaflops. The more computing power you engineer, at this level the higher the energy bill, now over 20 megawatts and climbing, they think the exaflop machine will such up as much as 100 megawatts.

Originally posted by DeepThought
Humans have built computers. How many computers have built a human?

A skilled go player will still beat the crap out of a computer, provided the board is large enough.

Can you build a computer, DeepThought? No you can't. Nor can a computer. Samo samo.

But even if you could build a computer from scratch, do you need a computer or a computer aided machine to do it? Yes, you do.

Originally posted by FabianFnas
Can you build a computer, DeepThought? No you can't. Nor can a computer. Samo samo.

But even if you could build a computer from scratch, do you need a computer or a computer aided machine to do it? Yes, you do.

Yes, I can build a computer. I can even do it from raw materials if you insist, although it'll require someone to turn the handle to make it work. I'm thinking of something on the lines of Charles Babbage's Analytical machine. All it needs to qualify is to be Turing Complete - it doesn't need to be electronic or particularly fast.

Originally posted by DeepThought
Oh yes it is. Human's can take on NP hard problems and get it right. No matter what the processing power of a Turing machine it simply cannot compete with a human on some sorts of problems.

Bear in mind that we have significant philosophical differences. You seem to regard the world as deterministic. I do not. So from your point of view a ...[text shortened]... ied to determinism, a quantum Turing machine is qualitatively different from a conventional one.

You are making a positive claim, that it's impossible for a Turing Machine to emulate
a biological brain.

Positive claims require positive evidence, proof.

Show me the PROOF that it's IMPOSSIBLE for a Turing Machine to emulate a biological
brain.

Human's can take on NP hard problems and get it right.

So?

We also very frequently get it wrong.

Most NP hard problems nowadays are being handled by computers, handling logistics
and optimisation problems, and they do far better at it than we do.

Deterministic or not, there is nothing humans do that I have ever heard of that a
sufficiently powerful and suitably programmed Turing Machine cannot do.