Nasa runs competition to help make old Fortran code faster

[Notebook]

Nasa is seeking help from coders to speed up the software it uses to design experimental aircraft.
It is running a competition that will share $55,000 (£42,000) between the top two people who can make its FUN3D software run up to 10,000 times faster.
The FUN3D code is used to model how air flows around simulated aircraft in a supercomputer.

The sensitive nature of the code means the competition is only open to US citizens who are over 18.

One for the programmers!

http://www.bbc.co.uk/news/technology-39803425

N.

ps extra points if you can identify the aircraft, without the internet.

 

[Loru]

One for the programmers!

http://www.bbc.co.uk/news/technology-39803425

N.

ps extra points if you can identify the aircraft, without the internet.

XB-70 Valkyrie?

 

[Urwumpe]

Damn discrimination of 99% of the world.

I would have suggested simply making a LLVM front-end for FORTRAN and include some high-level optimizations in the middleware...

 

[Notebook]

Correct Loru! Came to an unfortunate end, I think one is in a museum?

N.

 

[Urwumpe]

Correct Loru! Came to an unfortunate end, I think one is in a museum?

N.

Not just unfortunate. It was also developed at a time, when it was already obsolete. Flying high and fast was no longer an option to defeat Soviet air defenses. The same already killed the B-58 in the end. Not high and fast enough. Compared to ICBMs especially.

 

[Ravenous]

The nationality requirement is, sadly, the way it goes these days for anything with saucy military applications.

An yep, of course that plane was the XB-70. Marvellous machine. Wrong question to ask on this forum

 

[martins]

It is running a competition that will share $55,000 (£42,000) between the top two people who can make its FUN3D software run up to 10,000 times faster.

Why is there a 10,000x ceiling to the challenge? Is it like blackjack where you want to get as close to 10,000x performance but not over, or you lose the prize money?

 

[Fabri91]

I think it's one of the parts of English which sound a bit weird to those who don't grow up with it, since it's quite a subtlety: your interpretation would also be the most intuitive one to me, but I think it means something more along the lines of "we reasonably expect it to run 10000 or so times faster in the best case, but sure won't be displeased if it runs even faster".

 

[martins]

but I think it means something more along the lines of "we reasonably expect it to run 10000 or so times faster in the best case, but sure won't be displeased if it runs even faster".

I agree that this is the most likely intended meaning, but it raises even more questions:

- will the prize money be distributed to the two fastest solutions regardless of actual speedup, in which case, why mention the 10,000 at all?

- will no money be handed out if the best solution is significantly below 10,000x, because it didn't meet expectations?

 

[Notebook]

More info here:

https://herox.com/HPFCC

 

[boogabooga]

Correct Loru! Came to an unfortunate end, I think one is in a museum?

N.

Yes, there were two and one is near me at the National Museum of the United States Air Force in Dayton, Ohio. I've seen it many times.

The other one crashed in 1966 when it collided with an F-104.

It's an XB-70 because it was used for research flights once it's military limitations became apparent.

Regarding the code, I've never heard of FUN3D, but it seems that each NASA center has its own CFD code, or 2 or 3. There are newer and more modern Navier-Stokes solvers floating around. Wonder what their attachment to this is?

 

[Thunder Chicken]

Regarding the code, I've never heard of FUN3D, but it seems that each NASA center has its own CFD code, or 2 or 3. There are newer and more modern Navier-Stokes solvers floating around. Wonder what their attachment to this is?

Probably existing code validation. Knowing how to properly implement a solver requires the ability to understand exactly what the solver is doing. Each kingdom needs to know their own tools intimately. They also each seem to have their own meshing codes as well. I did a lot of work with CGNS.

As always, I think XKCD has something relevant:

 

[Keatah]

This prize money.. To me it looks like a zero-risk contract. All the contenders have to do the work first, then NASA looks it over. Then NASA says, "I want thatta one there!". Pay the money and get the solution.

A contract in disguise!

 

[Andy44]

This prize money.. To me it looks like a zero-risk contract. All the contenders have to do the work first, then NASA looks it over. Then NASA says, "I want thatta one there!". Pay the money and get the solution.

A contract in disguise!

It's like one of those TV talent shows, except for nerds.

 

[Notebook]

Talent and nerds, mutually exclusive?

N.

 

[Andy44]

Talent and nerds, mutually exclusive?

N.

Not at all.

 

[Enjo]

Talent and nerds, mutually exclusive?

N.

You've got a really old school definition of "talent".

 

[RGClark]

In logic there are techniques for simplifying a sequence of logical statements to the shortest statement(s) possible.

Doesn't the same thing exist in computer science?

Bob Clark

 

[Col_Klonk]

Oh jeezz, Is NASA full of Dumbass nutjobz.

First of all what platform are they running the fortran code - Early 80's.
After that they SHOULD have the whole hardware (including circuit design) 'description', in order to do the optimisation.

Next.. you're not going to get faster than assembler programming and optimisation, maybe via hex editing.

10Kx faster sounds like they have screwed up big time, or BSing big time.

So ... what's out there, that was launched a long time ago.. that is so far away, that now needs quick comms ----> Voyager's maybe ?
:thumbup:

Nota Beta: I'm game for the project

 

[Keatah]

It'd be funny if a 15,000x solution came from out of the country!