News on China's scientific and technological development.

Quickie said:

Moore's Law ...

Click to expand...

well at first my point could be summarized by

Jura said:

...
now this business sure looks like a boondoggle to me

Click to expand...

and I thought ever-ever-ever-increasing supercomputer power was as useful as increasing calibers of Battleships' main gun to 20" and beyond would've been ... in an attempt to show it, I made a prediction of a zetta-computer (LOL)

Jura said:

Click to expand...

which might look absurd, since an exa-computer is expected only around 2020, but ...
later I wasn't so sure if people at the point

Jura said:

1993 59.7 GFlops

Click to expand...

had not thought, common, Moore's law wouldn't allow an EXA-computer (relatively accurately predicted: 2018 https://www.sinodefenceforum.com/ne...ogical-development.t4270/page-296#post-426366
by just simple exponential curve) ... now I sure hope China indeed builds it LOL!

but I'm guessing (actually bluffing) for real-world applications like predicting weather in large areas with complex landscape for example, exaflop would need to be achieved using that HPCG metrics
(note "no system reached a single petaflop running HPCG" so far: https://www.sinodefenceforum.com/ne...ogical-development.t4270/page-297#post-426392), not "just" in Linpack

and now, please, I'm not naysaying Chinese achievements in this field

Jura said:

well at first my point could be summarized by

and I thought ever-ever-ever-increasing supercomputer power was as useful as increasing calibers of Battleships' main gun to 20" and beyond would've been ... in an attempt to show it, I made a prediction of a zetta-computer (LOL)
which might look absurd, since an exa-computer is expected only around 2020, but ...
later I wasn't so sure if people at the point

had not thought, common, Moore's law wouldn't allow an EXA-computer (relatively accurately predicted: 2018 https://www.sinodefenceforum.com/ne...ogical-development.t4270/page-296#post-426366
by just simple exponential curve) ... now I sure hope China indeed builds it LOL!

but I'm guessing (actually bluffing) for real-world applications like predicting weather in large areas with complex landscape for example, exaflop would need to be achieved using that HPCG metrics
(note "no system reached a single petaflop running HPCG" so far: https://www.sinodefenceforum.com/ne...ogical-development.t4270/page-297#post-426392), not "just" in Linpack

and now, please, I'm not naysaying Chinese achievements in this field

Click to expand...

Ok, we all know that Usain Bolt can not run a Marathon at his 100m speed (37.58km/h). Ferrari is not as fuel efficient at a Toyota Prius. Different game, different objectives. Orange and Apple. Chicken and Duck etc. etc.

Well, people only joined the hot argument before they realize that a naysayer was naysaying. Given time, people may learn to ignore.

Janus-faced is his hallmark.

Jura said:

but I'm guessing (actually bluffing) for real-world applications like predicting weather in large areas with complex landscape for example, exaflop would need to be achieved using that HPCG metrics
(note "no system reached a single petaflop running HPCG" so far: https://www.sinodefenceforum.com/ne...ogical-development.t4270/page-297#post-426392), not "just" in Linpack

and now, please, I'm not naysaying Chinese achievements in this field

Click to expand...

Common sense will tell you that it's not the case mainly for the following 2 reasons.

1. The Tauhulight, like any big projects, must've have been built in accordance with the type of applications it's intended to be used for and this would include weather prediction and simulation application which is what the Tauhulight (which excel more in the HPL benchmark than HPCG) was tasked to do in winning the 2016 ACM Gordon Bell prize.

2. The Tauhulight come in at 4th place behind the second placed Tianhe-2 in the HPCG benchmark. (Btw, the Tianhe-2 is only about < 4% slower than the K-computer according to the listing). If the HPCG performance is that important for that kind of application, the Tianhe-2 would be a better choice to enter the competition since according to the benchmark the Tianhe-2 would be over 50% faster than the Tauhulight in running the application. But we know this is just not what has actually happened: the Tianhe-2 did not enter the competition, and instead the Tauhulight, which excel in the HPL type of benchmark, performed superbly in the kind of application.

Jura said:

but I'm guessing (actually bluffing) for real-world applications like predicting weather in large areas with complex landscape for example, exaflop would need to be achieved using that HPCG metrics
(note "no system reached a single petaflop running HPCG" so far: https://www.sinodefenceforum.com/ne...ogical-development.t4270/page-297#post-426392), not "just" in Linpack

and now, please, I'm not naysaying Chinese achievements in this field

Click to expand...

No matter how fast the computer is you'll won't be able to predict the weather without weather observation stations every few kilometers to feed the computer with data required to predict it.
Japan has around 1700 stations around the nation. The system is called and collects various data ever 10 minutes and is fed into a computer to predict the weather.

Who started this contest and why now?

As in the June 2016 HPCG results list, Tianhe-2 takes the fist place, nobody out there claims the HPCG- benchmark is the better and more accurate benchmark for supercomputer.
So why now?

Quickie said:

Common sense will tell you that it's not the case mainly for the following 2 reasons.

1. The Tauhulight, like any big projects, must've have been built in accordance with the type of applications it's intended to be used for and this would include weather prediction and simulation application which is what the Tauhulight (which excel more in the HPL benchmark than HPCG) was tasked to do in winning the 2016 ACM Gordon Bell prize.

2. The Tauhulight come in at 4th place behind the second placed Tianhe-2 in the HPCG benchmark. (Btw, the Tianhe-2 is only about < 4% slower than the K-computer according to the listing). If the HPCG performance is that important for that kind of application, the Tianhe-2 would be a better choice to enter the competition since according to the benchmark the Tianhe-2 would be over 50% faster than the Tauhulight in running the application. But we know this is just not what has actually happened: the Tianhe-2 did not enter the competition, and instead the Tauhulight, which excel in the HPL type of benchmark, performed superbly in the kind of application.

Click to expand...

well assuming you've reacted to the part you highlighted in

Jura said:

but I'm guessing (actually bluffing) for real-world applications like predicting weather in large areas with complex landscape for example, exaflop would need to be achieved using that HPCG metrics
(note "no system reached a single petaflop running HPCG" so far: https://www.sinodefenceforum.com/ne...ogical-development.t4270/page-297#post-426392), not "just" in Linpack

and now, please, I'm not naysaying Chinese achievements in this field

Click to expand...

I'm sorry to tell you I'm not convinced by your argumentation, as 'common sense' (that's what you brought here, to my surprise) would tell me it's time to check how much bigger the investments into ever-ever-ever increasing supercomputers are still needed before intended real-world applications can be solved! is it ... (bluffing again) ... ten bil to build an exa-computer and predict whatever they want to predict (by the way I'm curious what would this be ... and wouldn't it be better to build instead several smaller centers so that an engineer can quickly connect and (just an example) wiggle the prototype in 3D using 4096 cores and won't be charged for this?? ... this type of questions if you know what I'm talking about, I mean the benchmarks, the competitions, the pride it's all fine, but I wonder if people are not overdoing it, if it's not some kind of bubble ...

on a purely technical side, I see what you meant by "Tianhe-2 is only about < 4% slower than the K-computer according to the listing": HPCG (Pflop/s)
0.5800/0.6027

ans =

0.9623

but what's kinda scary is the parameter (for all computers on the list!) called Fraction of Peak

SamuraiBlue said:

No matter how fast the computer is you'll won't be able to predict the weather without weather observation stations every few kilometers to feed the computer with data required to predict it.
Japan has around 1700 stations around the nation. The system is called

Please, Log in or Register to view URLs content!

and collects various data ever 10 minutes and is fed into a computer to predict the weather.

Click to expand...

I'm guessing here it's even worse because if weather will or won't change typically depends on where sometimes local fluctuations may or may not cause for example a cold front to pass (or stay back in Germany!) ... if it passes, it's possible more or less accurately say when it'll arrive somewhere in the valley, but if it doesn't pass while "the model thought" it would LOLOL

(just an illustration of the general problem, I'm not going to talk Czech weather here

duncanidaho said:

...

As in the June 2016 HPCG results list, Tianhe-2 takes the fist place, ...

Please, Log in or Register to view URLs content!

Click to expand...

indeed:


now ():


hey SamuraiBlue what's happened to the K in the meantime?

SamuraiBlue said:

No matter how ...

Click to expand...

Jura said:

I'm guessing here it's even worse because if weather will or won't change typically depends on

Please, Log in or Register to view URLs content!

where sometimes local fluctuations may or may not cause for example a cold front to pass (or stay back in Germany!) ... if it passes, it's possible more or less accurately say when it'll arrive somewhere in the valley, but if it doesn't pass while "the model thought" it would LOLOL

(just an illustration of the general problem, I'm not going to talk Czech weather here

Click to expand...

Predicting local weather if it will rain or not depends on various variables and those localized variables actually changes the out come in a larger scale event. It's called the where a very small local effect creates a ripple causing a cascade effect resulting into a major event in a large scale.
Like if a small cloud disrupts the amount of solar radiation in a small region causing a cold spot that eventually grows into a cold front resulting into a blizzard.
You can't predict this kind of chain of events by only looking at the big picture.

Jura said:

indeed:

now (

Please, Log in or Register to view URLs content!

):

hey SamuraiBlue what's happened to the K in the meantime?

Click to expand...

Aughhh, it's ranked as #1 in November 2016 according to the chart you had posted.