LLVM to get Fortran compiler that targets parallel GPUs in clusters

Oko

Daemon

Reaction score: 771
Messages: 1,620

The fact that you are posting this link in off-topic subsection clearly demonstrate how irrelevant BSDs are for scientific computing. Unfortunately since the demise of Solaris scientific computing ecosystem is the one of Linux monoculture.
 

Oko

Daemon

Reaction score: 771
Messages: 1,620

LLVM is made for BSD.
People in academia don't care about any particular OS. As somebody who is familiar with U.S. academic computing I would have very hard time believing that anyone from the original LLVM project team (with exception of the oldest members) from the University of Illinois at Urbana-Champaign who started LLVM in 2000 under the direction of Vikram Adve and Chris Lattnerresearch to investigate techniques for static and dynamic programming languages have ever seen a computer running any BSDs and FreeBSD in particular. I work at Carnegie Mellon school of computer science and if I have to put a percentage 80% of our computers run Windows, 15% MAC and 5% Linux (mostly Ubuntu and some CentOS/ROCKS on clusters). I am one of few people in the buildings (we have three) having both OpenBSD and FreeBSD machines.
 
OP
OP
sidetone

sidetone

Daemon

Reaction score: 396
Messages: 1,118

There are a lot of condescending remarks about BSD or people's thoughts. Then move to Linux, or Windows.
 

Oko

Daemon

Reaction score: 771
Messages: 1,620

There are a lot of condescending remarks about BSD or people's thoughts. Then move to Linux, or Windows.
There is nothing condescending in stating the fact that nobody uses BSDs for scientific computing these days. That is well known even to new comers in the field let alone yo somebody who is actually using computers since late 80s. The fact that my laptop runs OpenBSD and that I use nvi/TeX to write the papers doesn't alter the fact that Hadoop is develop on Linux and that one can't use even such a trivial thing as MATLAB on BSDs due to the lack of vendor support. I run RHEL/ROCKS when I have to and I am day dreaming of a mystery investor who would put couple hundred of million dollars in Free/DF or Net and make it worth while scientific computing platform just like I remember from late 80s and early 90s. Land of Linux is full of misery but unfortunately has no alternative for scientific computing. You see I am not even talking about my beloved OpenBSD which is security network appliance OS which can be abused as more general OS.
 

Beastie7

Well-Known Member

Reaction score: 164
Messages: 409

There is nothing condescending in stating the fact that nobody uses BSDs for scientific computing these days. That is well known even to new comers in the field let alone yo somebody who is actually using computers since late 80s. The fact that my laptop runs OpenBSD and that I use nvi/TeX to write the papers doesn't alter the fact that Hadoop is develop on Linux and that one can't use even such a trivial thing as MATLAB on BSDs due to the lack of vendor support. I run RHEL/ROCKS when I have to and I am day dreaming of a mystery investor who would put couple hundred of million dollars in Free/DF or Net and make it worth while scientific computing platform just like I remember from late 80s and early 90s. Land of Linux is full of misery but unfortunately has no alternative for scientific computing. You see I am not even talking about my beloved OpenBSD which is security network appliance OS which can be abused as more general OS.
One of the reasons why Linux is prevalent in more fields is because the Linux community is actually loud about themselves. The BSD community is way too conservative and/or lazy when it comes to promoting BSD outside of the inner circle. And when attempts are actually made, the effort (or on some occasions, half-assed) put in doesn't make significant effect on mind-share. There's a shit ton of conferences the Linux community gets involved with that has absolutely nothing to do with Linux in an effort to show what they have to offer, where's the BSDs? There's a big world out there outside of the usual old *BSDCons. It's actually frustrating.
 
OP
OP
sidetone

sidetone

Daemon

Reaction score: 396
Messages: 1,118

Ok. It's fine to say FreeBSD or any other BSD lacks gpu scientific computing, and it's an interesting subject. The way it was said, was, why is it in an off-topic section, as if FreeBSD is not good because it lacks gpu scientific computing, or why did I post it at all in an off topic forum.

It shows potential of what can be added on top of LLVM. That's in FreeBSD's base system, and made for BSDs.
 

kpa

Beastie's Twin

Reaction score: 1,809
Messages: 6,318

It shows potential of what can be added on top of LLVM. That's in FreeBSD's base system, and made for BSDs.
Emphasis on word potential. Now you just need a commercially operating vendor that can create well tested and robust software (under strict deadlines too!) that they can support for years on end without compromise and in a way that the customers don't feel cheated after paying big bucks for the software. This is what separates heavy lifting from hobbyist doodlings.
 
  • Thanks
Reactions: Oko

obsigna

Aspiring Daemon

Reaction score: 573
Messages: 966

sidetone many thanks, for posting the link of the Ars Tecnica article. I found it very interesting. I read the article twice, and it does mention the purpose of HPC using GPU's, however, I can imagine that the proposed FORTRAN compiler will not be limited to this. So, chances are, that at some point in time we will see another choice of a FORTRAN compiler for FreeBSD besides that of the GNU Compiler Collection.

To those who don't find an article interesting, "don't read it." And even better, don't comment on it, only to let people know about your boring "better" choices.
 

Oko

Daemon

Reaction score: 771
Messages: 1,620

One of the reasons why Linux is prevalent in more fields is because the Linux community is actually loud about themselves.
Linux is prevalent in more fields due to huge monetary investment of companies like IBM, HP, RedHat, Canonical, (Silicon Graphics in the past) and many others. We are talking about billions of dollars. Actually considering the level of external investment in BSDs it is surprising to see so many new technologies coming out of that kitchen and adopted somewhere else. Loud Linux teens are irrelevant just like loud BSD, Windows, or Solaris users.
 

Oko

Daemon

Reaction score: 771
Messages: 1,620

"Nvidia and US National Labs will open source the results" in late 2016
NVidia has a proven track record of producing closed hardware and pushing binary blobs down the throats of anyone. Pretty sad for a subset of old Silicon Graphics employees which were known for innovation and forward thinking. I don't believe that they will open source anything until I see it. Which US National Lab did you think in particular? U.S. government agencies definitely have monetary power to promote new platforms for scientific computing. DARPA funded TCP/IP is an example of U.S. government promoting adoption of open (at that time technologically inferior) protocol which inclusion in BSD made it the most widely used OS in scientific computing in late 80s and early 90s.
 

Oko

Daemon

Reaction score: 771
Messages: 1,620

So, chances are, that at some point in time we will see another choice of a FORTRAN compiler for FreeBSD besides that of the GNU Compiler Collection.
You nailed. Until we see Intel or Portland compilers working on FreeBSD or any BSD for that mater we can just daydream about doing serious scientific computing.

Between ROCKS cluster distribution now based of CentOS (earlier on Red Hat) was paid largely from NSF grants. BSDs need that kind of investment.
 

Beastie7

Well-Known Member

Reaction score: 164
Messages: 409

Linux is prevalent in more fields due to huge monetary investment of companies like IBM, HP, RedHat, Canonical, (Silicon Graphics in the past) and many others.
That's exactly my point. It was Redhat (and subsequently the greater Linux community) who sought to collaborate with Dell for their Powerdge Line in the early 2000's, it was RedHat who sought to team up with IBM for PowerLinux and usage of RHEL. Including many other large OEMs. All the web framework conferences, ApacheCons, etc all had Linux enthusiasts (and the Linux Foundation) showing off their software. You think the Linux community just sat there until companies flocked to them? No, they sought out to promote Linux in an aggressive fashion; then the money came after that.

Despite my bias for headless servers, the fact that FreeBSD still doesn't ship a pre-built graphical image hinders potential adoption also.

Loud Linux teens are irrelevant just like loud BSD, Windows, or Solaris users.
And those "Loud Linux teens" are the reason why Linux is at the center of open source, and open source IT at large.
 
  • Thanks
Reactions: Oko

Crivens

Moderator
Staff member
Moderator

Reaction score: 928
Messages: 1,929

Could we please get back to the topic? Threads spiraling out usually end in a DE-ORBIT state.
sidetone, thaks for the link. I would like to have an alternative to gfortran :)
Oko, please do not tell our top customer that one can not do scientific computing on FreeBSD. He would find himself suddenly not wanting the product we are making for him.
 
OP
OP
sidetone

sidetone

Daemon

Reaction score: 396
Messages: 1,118

I don't believe that they will open source anything until I see it.
I am doubtful of Nvidia open-sourcing their implementation, but that would mean they would be ahead in selling more of their graphics cards, or related products. It would be a win for them to open-source it.

Which US National Lab did you think in particular?
It's quoted from the subtitle, they're not my own words. I edited my post, to add that.
 
OP
OP
sidetone

sidetone

Daemon

Reaction score: 396
Messages: 1,118

I bought a digital copy of FreeBSD's magazine, the one about science, earlier this year. Hate to say it, but I was mildly disappointed, because it wasn't what I expected. It was about FreeBSD's filesystems for storing scientific data for network use. (If I remember correctly, it was about FFS [Fast Filesystem]) This is great to some people, but I was expecting content like creation of scientific diagrams, data analysis, programs about periodic tables, or mathematical programs using BSD.
 

Oko

Daemon

Reaction score: 771
Messages: 1,620

Oko, please do not tell our top customer that one can not do scientific computing on FreeBSD. He would find himself suddenly not wanting the product we are making for him.
Is the type of the product secret? Could you please say something about the product. I am guilty as a charged of using term scientific computing in the fairly narrow sense of the word to try to make a point. We keep our scientific data on FreeBSD ZFS pools. We use Jails to deliver our analytical tools to our sponsors. ATLAS, BLAS, LAPACK, C/GCC, GSL, Python, R, lately TeX are all very reasonably maintained on FreeBSD. What is really hurting is lack of MATLAB, Mathematica, Maple (yes FreeMat, Octave, and SciLab last time I looked are all very well maintained), proprietary compilers and such. However the biggest source of frustration is lack of clustering suite (SUN had grid engine which worked reasonably well in the past on FreeBSD) possibly a cluster distribution, then lack of any CUDA (I know it is NVidia who is screwing things up but possibly FreeBSD foundation could make some kind deal with them). Hadoop, spark and similar would be nice. I hate to say this but Open Stack and containers are probably going to stay so something of that crap too.

Right now I really like FreeBSD as a storage OS, and light weight virtualization platform. Network is traditionally strong. Scientific computing is maybe another areas where FreeBSD could excel besides let say security appliances. Linux is too big and is trying to be everything for everyone which is a recipe for not being good for anyone. I will shut up now.
 
OP
OP
sidetone

sidetone

Daemon

Reaction score: 396
Messages: 1,118

What is really hurting is lack of MATLAB (yes FreeMat, Octave, and SciLab last time I looked are all very well maintained), proprietary compilers and such
FreeBSD has math/sage available in ports. It's GPL, but that's not important. I've mentioned this before, when it was pointed out how FreeBSD was missing a Matlab alternative. There are even books about it, and free online videos.
 

Oko

Daemon

Reaction score: 771
Messages: 1,620

FreeBSD has math/sage available in ports. It's GNU, but that's not important. I've mentioned this before, when it was pointed out how FreeBSD was missing Matlab. There's even videos on the internet about it, and books.
math/sage is just a kitten sink front and for things like Maxima, GNU Octave, and similar. I am not a big fan of that project. I would prefer to see more of very specialized ports lake math/gap and similar (you would be shocked how many of those specialized tools for various areas of pure, applied Mathematics and science (bio-statistics, genetics, neuroscience) are now available open source. However typically they target Linux only.
 

ronaldlees

Aspiring Daemon

Reaction score: 321
Messages: 756

"Nvidia and US National Labs will open source the results" in late 2016
One can hold out for that to happen, and maybe it will. I think the traditional PC graphics vendors are quite entrenched in their proprietary thinking, it's being driven by competition (originally in the gaming market, - a market to which we really owe all the GPU horsepower improvements).

Some of the ARM based suppliers are perhaps less entrenched, but so far even Mali graphics offers only kernel source and not the userland part (except as blob). But, there are some reverse engineering efforts ongoing (limadriver.org) to open source the Maii driver, albeit for the time being only up to the 400 series. It probably takes something like Mali 628 or higher to do serious GPGPU stuff.

As I posted on another forum, the bitcoin miners proved the viability of these smallish boards, if you put enough of them together. Although the ASIC boards are not exactly like any of the commercial offerings of SoC ARM boards these days, I think serious power could be harnessed with the latter for GPGPU work. The Odroid XU4 comes to mind. A couple dozen of those, and you're probably crunching some numbers.

For non-Linux though, we're still looking at all the reverse engineering, and potential legal issues? Don't know. Interesting though ...
 
Top