News Chinese Exascale Supercomputers: Not Everything Is as It Seems

[Admin]

China's third exascale supercomputer will need all-new hardware.

Chinese Exascale Supercomputers: Not Everything Is as It Seems : Read more

 

[Endymio]

So China cheated on the test? That shouldn't surprise anyone...

 

[setx]

Is US that desperate that they've lost/losing performance crown? China didn't even announce their supercomputer speed but so many accents in the article "it's not real, probably!"

 

[Endymio]

China didn't even announce their supercomputer speed ...

The PRC-funded National Research Center entered their supercomputer to win the Gordon Bell prize. How is that not announcing its speed?

 

[setx]

The PRC-funded National Research Center entered their supercomputer to win the Gordon Bell prize. How is that not announcing its speed?

That's way more about their simulation algorithm than raw speed.

 

[Endymio]

That's way more about their simulation algorithm than raw speed.

The prize is for achievements in supercomputing, not algorithm design. China knowingly entered it in an attempt to garner prestige and standing, and they knowingly cheated on the benchmark.

 

[crostini]

Is US that desperate that they've lost/losing performance crown? China didn't even announce their supercomputer speed but so many accents in the article "it's not real, probably!"

It's important to realize the amount of money China spends for a worldwide PR campaign that amounts to typical propaganda. From LeBron to John Cena to many US state governors, it includes big names all the way to the nameless. Including those posting on forums of tech sites online. In china you can even decrease your prison sentence by each post you make supporting the peoples republic.

 

[setx]

It's important to realize the amount of money China spends for a worldwide PR campaign that amounts to typical propaganda. From LeBron to John Cena to many US state governors, it includes big names all the way to the nameless. Including those posting on forums of tech sites online. In china you can even decrease your prison sentence by each post you make supporting the peoples republic.

Well, maybe they spend money for propaganda in US, but definitely not where I am. Also, never heard of LeBron nor John Cena...

Personally I welcome news about (potentially) competing new Chinese hardware purely from consumer standpoint of the more competition the better. Just compare Intel's and Chinese videocards: Intel's are pure propaganda for now but every technical site just has to regularly write lengthy articles about them, while there is almost nothing about Chinese cards. Pretty obvious who pays for what.

The prize is for achievements in supercomputing, not algorithm design. China knowingly entered it in an attempt to garner prestige and standing, and they knowingly cheated on the benchmark.

I can't comment on cheating as I'm not going to read it into details, but supercomputing without proper algorithms is just stupid. Computers (and supercomputers) are totally useless without corresponding algorithms.

 

[Nobonita Barua]

Instead of these china should fast track EUV, domestic ones are scheduled for 2023,USA SCs are not of much use given they can not really show what they are doing with those. They always come first in test where they make the question papers.
Next, bring on the quantum computers.

 

[galftanus]

The prize is for achievements in supercomputing, not algorithm design. China knowingly entered it in an attempt to garner prestige and standing, and they knowingly cheated on the benchmark.

No, they did not. The article itself seems to be written by somebody who is not too familiar with HPC and supercomputing.

The fact that they achieved 1.2 Exaflop in a real scientific algorithm in no way contradicts them getting about 1 Exaflop in Linpack. Linpack is a very simple benchmark and achieves a very high percentage of peak performance. Real world algorithms typically achieve much less. Of course it does not prove it either, since the floating point accuracy is different.

The Gordon Bell price is given to " recognize outstanding achievement in high-performance computing" " (https://awards.acm.org/bell). Using mixed precision algorithms to achieve best performance is absolutely allowed, and a typical strategy. It is not like they hide it, read the article yourself: https://dl.acm.org/doi/pdf/10.1145/3458817.3487399