Why such high demand for computing power?

[Nick Ger]

Why do people still want to make better supercomputers?

Where do laboratories need so much computing power for?

[Lurick]

You fold and extrapolate data on a protein and let me know how that goes for you

[manikyath]

well.. china uses them for traffic analysis (and you bet the US does too...), the weather forecast on TV is being largely computed by a supercomputer, science uses them for complex stuff like explosion analysis, protein folding (f@h) is a supercomputer workload as well, etc.

 

in short, tasks that are highly parallelized, and require A LOT of computing horsepower, are perfect candidates for a supercomputer, and with ever growing datasets and demand, it makes sense that the supercomputer market keeps growing along.

[NineEyeRon]

Ha my lab ran on computers way less powerful than my gaming PC...

 

Some applications really do need all that power though, just not my labs.

[Bartholomew]

4 minutes ago, Nick Ger said:

Why do people still want to make better supercomputers?

Where do laboratories need so much computing power for?

More and more complex research; in many fields, medical, physics, energy, artificial intelligence.

 

Although take the news about it with a grain of salt; a lot of it is cover up of just being able to cs:go at 15000fps among scientists.

 

[Windows7ge]

There are many applications today that can still benefit from ever higher computational capacity in ever shrinking physical size/power consumption.

 

And don't forget this technology often trickles down. People like us often get our hands on it at affordable prices only a couples years after release.

[Nick Ger]

XD Thx for answer

[freeagent]

Benchmarks!

 

And other stuff

[The1Dickens]

4 hours ago, Bartholomew said:

More and more complex research; in many fields, medical, physics, energy, artificial intelligence.

 

Although take the news about it with a grain of salt; a lot of it is cover up of just being able to cs:go at 15000fps among scientists.

 

But why stop at CS:GO, when you can also get DOOM Eternal up there?

[Pickles von Brine]

Here is an example of science done on a super computer. 

Another example

 

[RasmusDC]

my old job, at Vestas developing Windmills, used them for weather analysis, to ensure that placement of mills where done in the most wind heavy areas.. so analysis of weather pattern and geometri of landscape, i am quite sure it was also used by TV because it was such a strong setup.. but not sure.

 

it was only 481,8 teraflops in 2017 so it was far from the biggest, and by todays standards really small, but it had access to 50.000.000 data points from all standing modern windmills. at the time in 2017 i think it was 35.000 mills, each with a large sensory network. 

[Pickles von Brine]

9 minutes ago, RasmusDC said:

my old job, at Vestas developing Windmills, used them for weather analysis, to ensure that placement of mills where done in the most wind heavy areas.. so analysis of weather pattern and geometri of landscape, i am quite sure it was also used by TV because it was such a strong setup.. but not sure.

 

it was only 481,8 teraflops in 2017 so it was far from the biggest, and by todays standards really small, but it had access to 50.000.000 data points from all standing modern windmills. at the time in 2017 i think it was 35.000 mills, each with a large sensory network. 

Now I can imagine there being many more data points. Regardless that is some serious crunching power to chew through all of that. How long did it take to do the modeling?

[RasmusDC]

2 hours ago, Lord Xeb said:

Now I can imagine there being many more data points. Regardless that is some serious crunching power to chew through all of that. How long did it take to do the modeling?

well from what i know it was live monitoring, so it supported a "live" dashboard + it did long term weather forcasting, the sensor network was just to support the generation of weather forecasting, these models sucks up computing power..

 

to ensure right placement of windfarms, which is more important, than just having a mill that is called 2MW.

 

The 50.000.000 sensor points, just created even more information to model from. i am quite sure that it already took all other datapoints in weather forecasting into consideration..

 

They looked at cloud computing at the time, but it was cheaper to build and own it, than it was to rent, since it was continous work.. that was needed. 

 

but the detail i have from 2017 is that the 50.000.000 measurement points each second = data on petabyte streaming level, so the infiniband network. and a isilon storage system from EMC at the time..

 

I am quite sure that the setup today is just "peanuts" compared to what you can get. and i was at  Vestas from 2008-2013 aprox.. so it is a bit after my time, they really invested, we did have a something then too.. 

[Pickles von Brine]

20 minutes ago, RasmusDC said:

well from what i know it was live monitoring, so it supported a "live" dashboard + it did long term weather forcasting, the sensor network was just to support the generation of weather forecasting, these models sucks up computing power..

 

to ensure right placement of windfarms, which is more important, than just having a mill that is called 2MW.

 

The 50.000.000 sensor points, just created even more information to model from. i am quite sure that it already took all other datapoints in weather forecasting into consideration..

 

They looked at cloud computing at the time, but it was cheaper to build and own it, than it was to rent, since it was continous work.. that was needed. 

 

but the detail i have from 2017 is that the 50.000.000 measurement points each second = data on petabyte streaming level, so the infiniband network. and a isilon storage system from EMC at the time..

 

I am quite sure that the setup today is just "peanuts" compared to what you can get. and i was at  Vestas from 2008-2013 aprox.. so it is a bit after my time, they really invested, we did have a something then too.. 

Still impressive regardless. That is a lot of data to churn through especially in real time. A petabyte streaming level is just... wow!.

[RasmusDC]

4 minutes ago, Lord Xeb said:

Still impressive regardless. That is a lot of data to churn through especially in real time. A petabyte streaming level is just... wow!.

i remember just talking here in LEGO about the streaming dataamount of TB from our moulding setup, i we wanted to monitor all sensors in all machines, and the infrastructure it needed. 

 

the PETA byte is from an article on the VESTAS1 Supercomputer, not from me, so it does seem a bit high. i don´t really know what Inifini band can handle. i am not a Data scientist or a IT specialist, so when talking that kind of equipment, i am out of my normal working area.

 

We just used the data in development.

[GDRRiley]

For almost any type of simulation. 
from nukes, to protein folding, to disease spreading predictions, to weather predictions. 

[minibois]

Gotta run those algorithms, which you need compute power for.

Remember, everything a PC does it simple math, wrapped in a layer of more complicated math, layered in programming code.

 

In that sense, everything you do on a PC requires computing power, but with certain sorts of simulation you have a lot of ways it can go. Simulation will often have to act out all scenarios and then do things like calculate the difference and spit out so etching meaningful for whoever ran the simulation.

[Donut417]

15 hours ago, Nick Ger said:

Why do people still want to make better supercomputers?

How else do you think the NSA can analyze all those phone calls, emails and text messages? 

[JZStudios]

18 hours ago, Lord Xeb said:

 

That's just slow motion footage of lighting farts on fire.

[Delicious Cake]

There are also things like LHC experiments producing so much data, that not only are massive levels of computing power needed to make sense of it within a reasonable about of time, massive levels of computing power is also needed just to first sort through and filter the raw data for which bits of data are "relevant" vs "noise".

[Pickles von Brine]

52 minutes ago, Delicious Cake said:

There are also things like LHC experiments producing so much data, that not only are massive levels of computing power needed to make sense of it within a reasonable about of time, massive levels of computing power is also needed just to first sort through and filter the raw data for which bits of data are "relevant" vs "noise".

I believe last I heard when they built them, each detector was creating about 1PB of data per second and that had to be narrowed down to about 1TB of data a second. So they needed some real horsepower to scrub through the data. 

[alpha754293]

On 3/4/2020 at 4:40 PM, Nick Ger said:

Why do people still want to make better supercomputers?

Where do laboratories need so much computing power for?

The Shanghai Supercomputer was used in China when they built the "temporary Coronavirus 'field' hospital" -- they used the supercomputer in order to optimise the design of the ventilation systems in the room, and that's inside the hospital.

 

They then also use the same supercomputer to look at the effects of the ventilation system external to the hospital (i.e. you have to vent the air somewhere, so you don't want to be spitting out contaiminated air back into the environment) and/or also modelling the nearby area that would be at risk for "pollution".

 

So, that's a recent use case of what supercomputers are still used for.

[Kisai]

On 3/4/2020 at 1:40 PM, Nick Ger said:

Why do people still want to make better supercomputers?

Where do laboratories need so much computing power for?

There's always been an arms race for supercomputers, primarily for scientific reasons and bragging rights. As machines got faster and better, the scale and accuracy at which things can be done at improves. Like in 1975 the Cray-1 was 80Mhz. Back before the first PC's that were 4.77Mhz. Also the third 64-bit computer. The Cray-1 was 160 MFLOPS. Comparatively speaking, the Intel i860 at 40Mhz in 1989 was at half the performance (80MFLOPS.) Back when desktop CPU's were 486's and capable of 1 MFLOPS if they had a FPU. The Nintendo 64 did 200MFlops (split between it's CPU and GPU) in 1996, so you're looking at shrinking some massive hardware to the size of book in 20 years.

 

http://www.roylongbottom.org.uk/mips.htm

 

The current fastest Super Computer is the IBM Summit at 148.60 PFLOPS

https://www.top500.org/system/179397

 

While the individual performance of a CPU has kinda hit a wall, super computers are usually scaled by adding more nodes to the cluster. So you could in theory just keep adding systems infinitely so long as you can power them.