The World's First Petaflop SuperComputer

[neutrino]

The honours go to Japan in creating it. Although it is not officially the world's fastest supercomputer, it may well be when the list of the fastest 500 comes out next time.

http://www.newsfactor.com/news/Japa...uter-Stakes/story.xhtml?story_id=1220059R0ADY

EDIT: This should be probably be in the Technology section, so feel free to shift it there if any of the mentors feel so.

 

[Chi Meson]

So, my donations to peta weren't such a waste then?

 

[Moonbear]

I don't know, sounds like quite a flop to me.

 

[Jamesnns]

1) And I THOUGHT I was saving the whales.
2) We can now find the largest Prime number.
3) 3.14159? (come on, you KNEW it was coming.)

4)1st post...what fun.

 

[LowlyPion]

2) We can now find the largest Prime number.

Which one is it?

 

[turbo]

Which one is it?

Is there one Largest Prime? Wouldn't we have to construct a computer more complex than the Universe to even stand a chance of locating that?

 

[Phrak]

The honours go to Japan in creating it. Although it is not officially the world's fastest supercomputer, it may well be when the list of the fastest 500 comes out next time.

http://www.newsfactor.com/news/Japa...uter-Stakes/story.xhtml?story_id=1220059R0ADY

EDIT: This should be probably be in the Technology section, so feel free to shift it there if any of the mentors feel so.

The link you supplied didn't work well. Wiki supplies a value of 4.1 PFLOPS as some sort of current record.

 

[Ivan Seeking]

2) We can now find the largest Prime number.

Mathematicians found that last week under a '56 Buick parked between 2nd and Main.

 

[Borek]

The link you supplied didn't work well. Wiki supplies a value of 4.1 PFLOPS as some sort of current record.

You have noticed that you comment on over two years old post, didn't you?

 

[Jimmy Snyder]

Does this mean they can send one quadrillion credit card applications to my goldfish every second now?

 

[Jamesnns]

I don't know, twas a joke.(prime nimber)
And yea, I knew it was a two year old, so what. IBM just announed new Petaflop comp, caught my interest.

 

[Jamesnns]

Mathematicians found that last week under a '56 Buick parked between 2nd and Main.

Great humour!

 

[Jamesnns]

Does this mean they can send one quadrillion credit card applications to my goldfish every second now?

Doesn't sparky, your trusted pet dog,sand perhaps yout cat too get "pre-approved" apps. in the mail?

 

[Dr Lots-o'watts]

Is it a desktop or a laptop?

 

[LowlyPion]

Is there one Largest Prime? Wouldn't we have to construct a computer more complex than the Universe to even stand a chance of locating that?

No they can just estimate it and round to the nearest hundred thousand.

 

[Jamesnns]

I think they're using a standard PDA.
I know I am.

 

[Phrak]

I'm not sure I want a faster computer. Much faster and it would crash faster than it boots.

The size scales are interesting. Light travels .008 inches in a pico second. (1 pico = 1/tera). What are the feature sizes of transistors in todays supercomputers?

 

[Evo]

I remember when a teraflop was the big thing. One Teraflop is a trillion floating-point operations per second, so a petaflop would be a thousand trillion.

Here is the latest, or was. Only costs $133 million.

An American military supercomputer, assembled from components originally designed for video game machines, has reached a long-sought-after computing milestone by processing more than 1.026 quadrillion calculations per second.

http://www.nytimes.com/2008/06/09/technology/09petaflops.html

 

[LowlyPion]

I'm not sure I want a faster computer. Much faster and it would crash faster than it boots.

Consider a non-Microsoft Operating system then.

 

[LowlyPion]

I remember when a teraflop was the big thing. One Teraflop is a trillion floating-point operations per second, so a petaflop would be a thousand trillion.

My first IBM PC was 16kb memory and ran at about 5 mhz. I had to shell out $2000 for it. I added extra chips myself to get it up to 64kb to save a few hundred bucks. That was k in kilo-bytes - not a typo.

Moore's Law has been running riot ever sense.

 

[Phrak]

I remember when a teraflop was the big thing. One Teraflop is a trillion floating-point operations per second, so a petaflop would be a thousand trillion.

I wonder if the indentured servitude of twentyfive 24/7 technicians and engineers are included in the price, or are they extra?

This computer appears to be running 13,000 floating point operations in parallel, or perhaps some multiple of it. This is a computer involving several racks of floating point processors spread out over a computer room.

In (1/tera) seconds, 10^-15 sec, light in vacuum travels 8 microns.

The current smallest process feature size of a semiconductor about 0.09 microns, spacing adjacent FET transistors about 0.9 microns from center to center (my best guess).

The Pentium 4's feature size is currently at 0.13 microns.

http://arstechnica.com/articles/paedia/cpu/moore.ars/2"

 

[Redbelly98]

Is there one Largest Prime? Wouldn't we have to construct a computer more complex than the Universe to even stand a chance of locating that?

It would be better to find the largest integer first. Then we'd have an upper bound on the largest prime, greatly simplifying the task.

 

[Borek]

It would be better to find the largest integer first.

When I was younger there were no integers larger than 65535. It has changed since then.

 

[Jimmy Snyder]

It would be better to find the largest integer first. Then we'd have an upper bound on the largest prime, greatly simplifying the task.

First look for the largest even integer. Then you'll have an upper bound for the largest even prime.

 

[Phrak]

First look for the largest even integer. Then you'll have an upper bound for the largest even prime.

So it isn't 42...

 

[Redbelly98]

So it isn't 42...

No, but that may be an upper bound.

 

[Borek]

https://www.youtube.com/watch?v=drE5cHe6c3s

45000000000

 

[Jimmy Snyder]

The largest prime is a function of time. Setting the current time to $$t_0$$, we get $$\epsilon(t_0) = 2^{43112609} - 1$$, where $$\epsilon > 0$$ is the largest prime. Now apply A. Einstein's speed limit of 300 km/s to get a upper limit for the current time (which is no longer what the current time was when I began this post), in terms of the current space. We find that for each $$\epsilon$$, there is a $$\delta > 0$$ such that $$|f(x) - f(y)| < \epsilon$$. (See Introduction to Elementary Quantum Field Theory for Poets, by Gumm, page 486257). Now we use the standard mathematician's subterfuge of allowing zero to approach $$\delta$$ from behind. This sneak attack will insure that $$\delta$$ is circumspect while we let $$\epsilon \rightarrow 0$$. Thus the largest prime is zero.

 

[Jamesnns]

I remember when a teraflop was the big thing. One Teraflop is a trillion floating-point operations per second, so a petaflop would be a thousand trillion./QUOTE]

Or roughly the eqiv. of 250000 calcs per second per living person on this planet.

 

[mgb_phys]

The size scales are interesting. Light travels .008 inches in a pico second. (1 pico = 1/tera). What are the feature sizes of transistors in todays supercomputers?

The switching time is more a question of the capacitance of the junction which fortunately drops with feature size. Even desktop PCs now have <45nm features.

Your point is a big problem in network switches.
A 100Gb network switch means the light only goes a couple of mm as each new bit arrives.
You have to read the packet header and decide where to route the signal in the time it takes light to cross the PCB

 

[mgb_phys]

The largest prime is a function of time. Setting the current time to $$t_0$$, we get $$\epsilon(t_0) = 2^{43112609} - 1$$, where $$\epsilon > 0$$ is the largest prime. Now apply A. Einstein's speed limit of 300 km/s to get a upper limit for the current time (which is no longer what the current time was when I began this post), in terms of the current space. We find that for each $$\epsilon$$, there is a $$\delta > 0$$ such that $$|f(x) - f(y)| < \epsilon$$. (See Introduction to Elementary Quantum Field Theory for Poets, by Gumm, page 486257). Now we use the standard mathematician's subterfuge of allowing zero to approach $$\delta$$ from behind. This sneak attack will insure that $$\delta$$ is circumspect while we let $$\epsilon \rightarrow 0$$. Thus the largest prime is zero.

Fortunately the adoption of the "one,two,many,lots" counting system allows to simply state that the largest prime has the value "lots" - in fact all primes except "many" have the value "lots"

 

[Phrak]

The switching time is more a question of the capacitance of the junction which fortunately drops with feature size. Even desktop PCs now have <45nm features.

MosFets are more widely used, aren't they? There must be a rough way to characterize delay time with feature size. Something like t = f(RC+t_drift), using lumped values. What to do with power disipation--hold it constant with die area? And is there a lower bound on bias voltage for mosfets? It hadn't occurred to me to ask if there were a lower limit on flipping channel. Deposition thinkness would effect both R and C. Has it remained fairly constant as of late?