Supercomputers ...

Joined
16 Sep 2006
Messages
4,171
Reaction score
775
Location
Fife
Country
United Kingdom
Not sure if it's still the world's fastest (too lazy to conduct further research) however as of June 2020 the most powerful supercomputer in the world is Japan's Fugaku with a processing speed of 415.5 petaflops. This basically means it can perform the following quantity of floating point calculations per second ... hopefully I've got my number of zeros right!

415,500,000,000,000,000

Mainly used for computational scientific tasks that require extremely intensive processing e.g. molecular modeling, quantum mechanics etc.

If you're lost in the zeros:

1,000,000: Million
1,000,000,000: Billion
1,000,000,000,000: Trillion
1,000,000,000,000,000: Thousand Trillion (1 petaflop)

Again I'm too lazy to work out how this compares to the average home computer (although granted there's no such thing really) hopefully someone reading this has the skills to answer :)

I suppose I just think it's quite impressive when you consider how quickly a second comes and goes!
 
Sponsored Links
If your stock pc is clocking at a few Ghz and has a few cores; of the order of say 10Gfps, a bit more maybe.
That's 10^10 where "yours" is 10^15.
100,000 x as fast then.
They'll probably follow a Moore's type exponential growth, so we may see them something like 100x faster every 10 years.

Loads of good youtubes on quantum computing, many are aimed low. Fascinating things; I've spent a few hours going down rabbit holes of understanding, but you need to be very versed in quantum physics to fully "get" how they work. Ordinary mortals apply the maths and "it just works", though that's well beyond A level, say.

Then there's just the software issue. When experts can't get a schoolkids' exam prediction algorithm right, processor speed may not be the limiting factor......
 
Last edited:
If your stock pc is clocking at a few Ghz and has a few cores; of the order of say 10Gfps, a bit more maybe.
That's 10^10 where "yours" is 10^15.
100,000 x as fast then.
They'll probably follow a Moore's type exponential growth, so we may see them something like 100x faster every 10 years.

Loads of good youtubes on quantum computing, many are aimed low. Fascinating things; I've spent a few hours going down rabbit holes of understanding, but you need to be very versed in quantum physics to fully "get" how they work. Ordinary mortals apply the maths and "it just works", though that's well beyond A level, say.

Then there's just the software issue. When experts can't get a schoolkids' exam prediction algorithm right, processor speed may not be the limiting factor......
Thanks. If I'm using Moore's law correctly, the average home computer will therefore equal Fugaku's computing power in around 17 years then? Yeah I've read one or two articles about software often being the limiting factor over the hardware.

Mini-interesting side note. Many years ago I was offered a (low level) job as part of a university team. They were involved in moving a supercomputer from the UK to USA. Would have meant a minimum of 6 months in New York if I recall correctly. I turned the job offer down, too much of a home bird!
 
Sponsored Links
Thanks. If I'm using Moore's law correctly, the average home computer will therefore equal Fugaku's computing power in around 17 years then? Yeah I've read one or two articles about software often being the limiting factor over the hardware.

Mini-interesting side note. Many years ago I was offered a (low level) job as part of a university team. They were involved in moving a supercomputer from the UK to USA. Would have meant a minimum of 6 months in New York if I recall correctly. I turned the job offer down, too much of a home bird!

Moore's law is pretty much dead now, certainly dying as the limits to silicon technology are being reached.
 
Probably end up with a monkey brain processor eventually...

Screenshot_20210108-131643_Samsung Internet.jpg
 
Moore's law is pretty much dead now, certainly dying as the limits to silicon technology are being reached.
Its demise was announced in 2010, but in 2018 it was still going, and looking at the graph, isn't far off in 2020
https://en.wikipedia.org/wiki/Moore's_law#/media/File:Moore's_Law_Transistor_Count_1970-2020.png

One of the "cool" things about quantum computers is ( oversimplifying) that they don't dissipate heat on either the "0" state or the "1" state. Heat limits concentration of computing power on a chip, which is why we don't see current types getting much thicker. Qubits use a maybe state :).

For laptops, I can only think of games which "need" them to be faster. Video editing used to be the heavy load but that's been covered adequately for a while now. I have a 10 ish year old i7 which a few times slower than my whizzy desktop, but it's very rare to notice any difference in normal use between them.
 
We passed the point where CPU processing power was the major bottleneck in home or business use five or ten years ago. Especially with more of the heavy lifting being shifted into the cloud.

Moore's law has mutated over the years anyway. It started as processing doubling every year, then went up to every two. Is it speed in gigaflops or transistor count or even gigaflops per dollar.

Also, bigger in one place isn't always the point. After all the decrease in tansistor size, cost and power that lets an i7 outperform a Cray is also the same one that lets me turn my lightbulb off by yelling at a hockey puck in the corner.
 
Its demise was announced in 2010, but in 2018 it was still going, and looking at the graph, isn't far off in 2020
https://en.wikipedia.org/wiki/Moore's_law#/media/File:Moore's_Law_Transistor_Count_1970-2020.png

One of the "cool" things about quantum computers is ( oversimplifying) that they don't dissipate heat on either the "0" state or the "1" state. Heat limits concentration of computing power on a chip, which is why we don't see current types getting much thicker. Qubits use a maybe state :).

For laptops, I can only think of games which "need" them to be faster. Video editing used to be the heavy load but that's been covered adequately for a while now. I have a 10 ish year old i7 which a few times slower than my whizzy desktop, but it's very rare to notice any difference in normal use between them.

Well I'm not sure that's not a bit of a skewing of reality, does that graph show the number of transistors per core? The image is a bit blurred for me.
 
Well I'm not sure that's not a bit of a skewing of reality, does that graph show the number of transistors per core? The image is a bit blurred for me.
It's per CPU rather than core, otherwise numbers would have dropped or leveled off from 2006.
 
I suspected that might be the case. So it's been skewed by multiple cores.
Why does that matter? How does it matter? Moore's law isn't about the structure of chips or the way they're used, it's a high level observation that computing gets cheaper in a fairly predictable manner.

In practical terms without going to multiple cores you start edging even further back into CISC, excessive pipelines like the Prescott core or just bulking out the caches.
 
So, if we assume what we currently refer to as PCs/desktops will still exist even roughly in their current form for the near/medium future, how many years do you reckon it'll be until we're sitting at home browsing Google and using Excel with something as powerful as Fugaku? 10 years? 15 years? More?
 
Why does that matter? How does it matter? Moore's law isn't about the structure of chips or the way they're used, it's a high level observation that computing gets cheaper in a fairly predictable manner.

In practical terms without going to multiple cores you start edging even further back into CISC, excessive pipelines like the Prescott core or just bulking out the caches.

Well according to Moore's law

"Moore's law is the observation that the number of transistors in a dense integrated circuit (IC) doubles about every two years."

I guess you could argue that whilst multiple cores are packaged together in a CPU, each core is its own IC
 
Well according to Moore's law

"Moore's law is the observation that the number of transistors in a dense integrated circuit (IC) doubles about every two years."

I guess you could argue that whilst multiple cores are packaged together in a CPU, each core is its own IC
Well they are, if the cores are on the same silicon die, which they normally are these days.

If not then you'd need to exclude caches from some historic and current CPUs. They haven't always been kn the same die.
 
Sponsored Links
Back
Top