Supercomputer and the Milky Way - Computerphile

This Supercomputer is doing some of the most difficult computations in the world, about things that are out of this world.

Thanks to Dr Mark Wilkinson & Gary Gilchrist for showing us the DIRAC Complexity System at the University of Leicester.

This video was filmed and edited by Sean Riley.

[SapphireCrook]

Fun fact:
This machine was originally built to run Crysis 1. 
When that didn't work out, they just donated it to science! 

[CronicX]

Take those 4 cores and the price tag of the supercomputer and you have an Apple product. 

[NickiRusin]

The university I study at also has a similar machine. It's very nice to get an explanation on how they work.
And that Allan key inclusion is goddamn clever.

[LTS]

Its great to see Linux running on a supercomputer of this magnitude. Its scalability is simply astounding.

[elliottmcollins]

Having subtitles in the room was a nice touch, but it's a bit hard to read with the small light green font. Maybe something a little more accessible next time.
Awesome video.

[illdie]

I loved the end, the whole "They spend too much time talking to eachother and not enough time working!"

[EddieHD_]

I wonder how many mods on Skyrim I could play with on this Computer

[Milanevra]

If i may suggest using outline feature on subtitles so they are always easy to read.
This way, it depends on the background colour, and that is too dynamic to be easy on the eyes for the reader.
Cheers

[animalxxv]

For those who are interested, the high performance computer (HPC) probably uses a lsf system to manage the jobs (the programs to run) submitted by the users. Each user is usually set a maximum number of cores that they can use at any one time. To program one of these, you need to know MPI or OpenMPI as well as a programming language like C or Python, if you wish to communicate between the cores/nodes. Though you can run lsf jobs on the HPC without MPI and OpenMPI and still utilise all the cores.

MPI seems to be the standard

I'm cautious about calling it a super computer, from what I can tell its no bigger than the clusters I have access to. We call them high performance computers. Also the number of cores they quoted probably includes hyperthreaded cores so in reality it takes twice as long as they quoted, if that was the case (hyper threading just means that the CPU cores are less idle)

This HPC seems to be just made up of CPUs, you can get GPU ones which are just made up of a lot of GPUs, also there can be a mixture of CPU and GPU clusters (compilers can optimise plain MPI - C code into CPU and GPU operations).

[samramdebest]

I went to the computer farm at my local university earlier this year. They are pretty powerful as well, I will see if I can find any stats.


edit: with earlier this year, I mean earlier this school year

[taids]

Can I have the key to that server room? My body is ready...

[steel]

as soon as you guys are done with this comupter give it it me! i need to play thousands of video games at once!

[MrSlowestD]

I'm curious what is meant by "all to all" - in true parallel fashion, the processors have no need to talk to one another, they shouldn't need to be connected in any way.

But yeah, very cool setup!

[zenzylok]

Glad to see humans are utilizing computers to simulate the universe.

[firstnamelastname]

Truly interesting video, the behind-the-scenes type stuff is great on this channel.

Any more new channels planned Brady? Maybe retire the words and bible ones, and add in something fresh?

[WilliamAndrea]

more videos about cutting-edge computer hardware plz

[RMoribayashi]

I assume the supercomputer was named after Nobel Prize winning British physicist Paul Dirac.  *_He_* would be a good subject for a numberphile video.

[JacobShepley]

Hey Brady, I liked the video :)
but maybe because I'm (red/green) colourblind I sometimes found it hard to read the subtitles except when the background was very contrasted. Could you maybe use a subtitle text that has an outline, like black outline with green text or something?

Thanks 

[amxx]

Oh look, a small supercomputer :P
(working at ANL. we have Mira, the top500's n°5)


By the way running 16 node simulations on a supercomputer is just a wast of ressources. Just use a Xeon-phi and leave the supercomputer's nodes for large applications that really needs them !

[douro]

That is excellent uptime for a supercomputer, even for one made up of blade servers.  I would imagine this system is much cheaper to run than a shared memory system such as a Blue Gene/Q which doesn't have nearly as much reliability.