Let's suppose it is true, then the gain by replacing the current infrastructure with this new one needs to justify the downtime, training costs, ... The better we get, the harder it will become to gain from such a switch as the possible gains get smaller and the cost gets higher.
In the scenario that I “dreamt of” with LLVM Flang being used on DragonflyBSD, training costs will apply mostly to administrating OS (which is not hard to grasp), researchers are already familiar with NVIDIA/PGI
In NetBSD I made the mistake to switch to Clang and building world takes a lot of time. Switched back to GCC and it's faster. Compiling DragonflyBSD is a breeze. I wish FreeBSD could be compiled faster like this.
Given that running world binaries does not take up much time anyway, esp. for HPC, you may switch to -O1 and try again. If compiling fast is your goal, that is. Once upon a time kernel builds were using -O0, if I remember correctly.
That is only a fraction of the training involved. You also need to educate the higher ups. That will cost you.
Well, tell to higher-ups that better number crunching will save the $ and at the end of the day get them bigger bonusses – that’s all of the edu they’ll need
You need to take this into account:Well, tell to higher-ups that better number crunching will save the $ and at the end of the day get them bigger bonusses – that’s all of the edu they’ll need
Sorta how Unix was born when Patriarchs fooled Bell that they need new PDP for the legal dept to write forms faster
No wonder I’m not a successful businessman
I get your point, and it’s valid one from business perspective, because business is making their own $’s which they can burn as shareholders want, but from pure research perspective, which depends on other people money is not so simple if they look ahead.
)IBM and Oracle are entangled with so many governments around, that they will probably last until end of this civilization cycle.
Boffins are usually not admins; I didn't check FortranBSD, but if everything is set up for them to install it and start coding in Fortran, I see merit in that.
The author specializes in kernel modifications for improved performance. Based on that, I guess that's what he did with FortranBSD. It would make more sense for Fortran to have a special bootstrap, and make it a wider purpose math processing operating system. Let Python and other mathematical languages have a bootstrap there too.
I myself never tried, but if DragonflyBSD has full gcc in its kinda ports, it shoud have gfortran.
Yeah, gfortran is compiler in FortranBSD. When I said "dreamt of" I was referring to what I said here, in #17:
I think that will interesting to watch what will happen with FortranBSD by Mohamad Badiezadegan (mbzadegan).
Quote: “FortranBSD, a new operating system designed specifically for Fortran developers who demand high-performance and seamless multithreading support. FortranBSD is based on DragonflyBSD which is an outstanding OS for HPC application” /q
It will be very interesting to watch it when/if DragonflyBSD gets better LLVM support, and LLVM Flang gets production ready status (Flang is an open-source Fortran compiler based on the NVIDIA/PGI commercial compiler).
Boffins gonna make those supercomputers full of Nvidia cards go brrr on BSD
av.tib.eu
NetBSD was started 1993/04 and FreeBSD later on 1993/11.
You're right. I don't remember if I was looking at when 1.0 was released in 1994, or if I just messed up. NetBSD's first version 0.8, but the first release is what matters, as it's subjective to anyone, including those who released it, what makes it to be 1.0: https://www.netbsd.org/about/history.html. NetBSD was always the first to have formed a foundation. Edited it, to adjust to that, as it's information I intended to have there then.
Yw. I like putting things together for organizing concepts.
