Slashdot Mirror

Sure, if I pay twice or trice as much, you get more.

By that logic, I should get a Summit. Because what's not to like?

Even though #1 & #2 on the TOP500 are NVIDIA based. And NVIDIA DGX-1 lets anyone join TOP500 by buying a few dozens boxes and networking them together (see #50 and #51)

Even though #1 & #2 on the TOP500 are NVIDIA based. And NVIDIA DGX-1 lets anyone join TOP500 by buying a few dozens boxes and networking them together (see #50 and #51)

Even though #1 & #2 on the TOP500 are NVIDIA based. And NVIDIA DGX-1 lets anyone join TOP500 by buying a few dozens boxes and networking them together (see #50 and #51)

Even though #1 & #2 on the TOP500 are NVIDIA based. And NVIDIA DGX-1 lets anyone join TOP500 by buying a few dozens boxes and networking them together (see #50 and #51)

Agreed -- ignoring those 2+ billion Android devices (in 2017, no less) is strange.

If we ignore Android and look at the rest of the consumer space, then yes, we see Linux is a drop-in-the-bucket -- on the desktop it is estimated to roughly have only 2 %

However, Linux is dominating in other areas -- it is a wildly successful in the server space. 100% of the Top 500 Supercomputers in the world run Linux. Not bad for a "hobby" OS. /sarcasm I wish I could "fail" like that. =P

Why does Linux need to upstage Windows before it is considered "successful" ??

Yep! CPUs are definitely easier to program and sometimes GPUs are exactly wrong for certain workloads.

BTW: The current #1 (which will surely be supplanted in the soon to be refreshed Top500) is an all "CPU" machine (somewhat close to what Intel Phi was): https://www.top500.org/system/...

10M actual "CPU" cores. But - they are clocked lower and of quite a bit different architecture from your normal Xeon...

ARM's rise is definitely interesting because it should give us another option for good flops/watt while remaining simple to program.

For a 5,184 socket system a "Peak" performance of 2.3 Petaflops isn't that revolutionary.

I'm assuming that when they say "peak" they mean a LinPack "Rpeak" value which is usually (with a few exceptions) *higher* than the "Rmax" value that's actually used to order the systems by performance. There is no contra-indication in the story that these values are Rmax and in fact the story literally says "theoretical peak petaflops" definitely makes me think Rpeak?

You can see the soon to be outdated list from last November here: https://www.top500.org/list/20...

For perspective, if you go way down that list to #82 you'll see the rather pedestrian Riken Energy Hokusai BigWaterfall system from last year that hits a noticeably higher 2.58 Petaflops peak and only needs 1680 sockets with rather pedestrian 20 core CPUs to do it.

Scale that system down by 10% as a rough estimate to 1512 sockets and you have a 3.4 to 1 socket advantage and 4.8 to 1 socket advantage for a rather generic commercial system from last year that isn't even using a single supercompute accelerator to get its performance.

You are just showing your ignorance here. The toasters probably peaked in ...bread toasting abilities...about 100 years ago.
Current iphone has roughly the same processing power as super computers occupying entire datacenter from 30 years ago. Heck, single nvidia 1070 has roughly the same power as IBM ASCI White from year 2000. 1070 which was released in 2016 is about 60% faster than 970 which was released in 2014
For your reference, here's a picture of the gpu and the supercomputer

Quite a few systems lasted more than six months at number one. among them

ASCI Red
Earth Simulator
BlueGene/L
Numerical Wind Tunnel
Roadrunner
Tianhe-2
Sunway

And some computers were only on top for an instant.

/sarcasm Because in order for Linux to succeed on the desktop you need to run some slow, bloated, shitty ribbon bar IDE right?

It is obvious you don't deal with text day-in and day-out. Vim works because it becomes an extension of your mind once you learn how to use it. It is FAST. It can edit files of almost ANY size.

* Linux won in the server space. 100% of the Top 500 supercomputers in the world run Linux.
* Linux won in the mobile space. Linux runs on over 2 Billion monthly active devices.

That leaves the desktop space.

Guess what, no one gives a fuck that Windows dominates the desktop. People _already_ use Linux on the desktop. The only ones complaining about the "quantity" is you.

Oh, M$ can't just not do small, they can't do big either.
Android is a Google front end using a Linux kernel. It works well in phones because most phones have 2 or more core processors (usually Atom or similar). And Linux is very good at doing multicore. In fact it does it so well, that its used on very large computers too.
Here is a link to the top500 fastest supercomputers in the world. Go for the link that says
statistics - list statistics - operating system family.
Here ya go

I'd agree with your and the parent's assessment as well.

Back in 2007:

* Windows stability was joke. /que Your mouse has moved, please reboot!
* Linux stability was legendary. Custom kernels FTW!

Ironically, in 2017:

* Windows is (finally) stable
* Linux seems to be getting more unstable with each passing year

On the other hand, who knew that:

* Windows would turn into a complete clusterfuck of spyware
* Linux would dominate the Top 500 supercomputers, and be on 2 billion devices
* Windows Phones. MWUAHAHA! LOL. Even with a ~20 year head start MS _still_ couldn't figure out how to make Windows on a Phone sell.

> Linux support of notebook computers

Yup, that has always been an Achilles heel of Linux. All the *nix geeks (myself included) switched over to MBPs (MacBook Pros). We get a good GUI and BSD out-of-the-box. Dual booting, a dedicated Windows box, VMWare are not going away anytime soon but at least we have choices.

--
Want to play an old-school 8-bit RPG for the Apple 2, PC, and Mac?
Check out Nox Archaist

Analogy time:

McDonalds serves _billions._ No one is arguing that their quantity is even remotely comparably to quality. McDonalds excels at selling A LOT of cheap, shit food.

Likewise, the analogy to Operating Systems on the desktop is applicable:

* Windows = Quantity
* Linux = Quality

Although I would argue that Linux on the Desktop was NEVER about quantity, but about Freedom. Quality was always an afterthought.

Linux has failed to gain any serious traction on the desktop because:

1. "Windows is Good Enough" and "Momentum"
a) Disrupting the Windows desktop is almost impossible because the _casual_ Windows user doesn't give a fuck about freedom.
b) Likewise, game developers don't give a fuck about Linux because there is (almost) zero money in compared to Windows. Everyone and their dog is chasing after the "Fee-to-Play" bullshit model because whales and dolphins are where the _real_ money is.

2. Linux has never had anyone understand _good_ UI design.
Linux's core philosophy has aways been to copy what others are doing. What innovation it has done was never in the GUI space. The complete clusterfucks of KDE and GNOME, year and year, prove that they people are out of touch with Function and instead focus on Form at the expense of Function.

3. There is no one to enforce "standards"
Windows succeeded as game machines because everyone bought into the DirectX bullshit. And while we can debate the politics of Microsoft all day, the fact remains that there is nothing equivalent on the Linux side. The LSB, Linux Base Standard, is a step in the right direction, but there needs to be _standard_ APIs _across_ kernel version. The clusterfuck of PulseAudio is another example. Linux has "too many chefs in the kitchen with the majority re-inventing the oven". Instead we get half-assed implementations of everyone doing things their own way because "the other guys suck."

OpenGL ES has been a success because it provided a _standard_ across almost every device. Apple has always had Not-Invented-Here syndrome so they, like MS, have been pushing their proprietary APIs. If developers AND _users_ weren't morons they would _insist_ on ONE API across EVERY platform. But again, most people don't give a fuck about doing things "right" so we end with a shitty heterogeneous environment instead of a homogenous environment where implementation is held accountable to a the design specifications.

Using a bullshit metric of "popularity" as success nullifies the facts where Linux HAS been successful:

* Gee, 99.6% of the Top 500 supercomputers in the world run Linux. /sarcasm I wish I could "fail" like that!

* Android has over 2 BILLION devices, again running Linux. /sarcasm Again, I wish I could "fail" like that.

I've been using Linux off and on since the Slackware days. These days I have a dedicated Linux box, (along with Windows and OSX machines.) The fact of the matter is that _every_ Operating System sucks -- no one cares about switching from one crappy OS to another crappy OS. There are always strengths and weaknesses of every platform.

Apple, Google, and Microsoft are NOT interested in freedom. All they care about is profit and their hawking their proprietary crap. NOTHING will ever change until _everyone_ else decides there needs to be a better system and I don't ever see that happening. The problem is no one has the time, money, expertise, or status to pull this off, so we are stuck with crappy OS's that "sort of" do the job but suck in some way or another.

The rest of us just go back to using these crappy platforms complaining about it. :-/

Which is why I find it odd that instead of trying to introduce backdoors, governments don't just try to keep encryption to a level where only nation states have the resources to crack it

Because that's impossible, for very many reasons.

First, the nation state doesn't need to break just one use of encryption, it needs to break several. If you need to be able to break 1,000 people's encrypted communications, then the criminal needs 1,000th of the computational power that the state can throw at it. If the state needs to be able to break them in a day, then a criminal happy to wait a month needs 1/30,000th the computational power, which is about the difference between a data centre and a mobile phone.

Second, because breaking encryption is an embarrassingly parallel workload and so is ideal for botnets of GPUs. Guess who controls those? It's not the governments.

Third, because encryption is time sensitive. For some things, you care that no one can decrypt or spoof it this week (e.g. 'fire missiles at North Korea now' or 'buy stocks in Whole Foods now, I just learned that Amazon is going to try to buy them'). For some things, you care about whether someone can decrypt them in the next 20 years or longer. These things factor in the growth of computational power in the work factor for the attackers (GPUs and cheap FPGAs caused problems for some users of these) and so expect not to be broken by your non-government attacker with 20 years of growth in available processing power. Given the polynomial growth rate of cheap computational power, there's absolutely no way that you could expect a government to have now, more computational power than a gang of criminals could have in 20 years. To put this in perspective, 20 years ago, a 300MHz Pentium II was a really expensive CPU and the world's fastest supercomputer had a peak of 1,453.0GFLOPS, using 7,264 cores, i.e. about 6 times as fast as the iPhone 7's GPU, or about a quarter the speed of an nVidia GPGPU card. For encryption, integer ops matter more than floating point, but the trajectories are in the same direction.

I suggest bigger cuts to the Office of Science budget. Why do we need to spend money developing better, faster supercomputers? We can let the Chinese do all the expensive R&D, then we can buy the finished product from them. No problem. It worked for drywall, why not quantum puters?

The Chinese are expending significant resources building conventional supercomputers. Which suggests far more promise for quantum computers than current reality.

https://www.top500.org/lists/2016/11/

And that is true for all the computing leaders at present. We know how to build very effective supercomputers. We think that quantum computers might be great, the promise is there in theory, but you'd be a fool to ditch your conventional HPC systems.

And even if quantum computing becomes "a thing", suspicions abound that they might only be good as a specialized co-processor. Need to factor a crazy large prime number? You send that to the quantum computing module. Need to run a web server, database, render some graphics, parse some text, meat & potatoes stuff? You still want a conventional processor.

The US and close allies control more or less the entire semiconductor business. There's no worry of say, China or Russia, competing anywhere in the foreseeable future.

Russia? No. China...

Here's the world's #1 supercomputer:

https://en.wikipedia.org/wiki/...

The US decided to give China a leg up by denying them access to the top chips, thereby inviting them to dump a metric assload of cash (the US actually specified an imperial fuckton, but the metric assload is about a factor of 1.05 larger) on developing their own processes.

And just in case you think they've simply brutalized it:

https://www.top500.org/green50...

It's also #3 on the Green 500 list. Below the best xeons, but not by far, 10% perhaps. It's also not clear what process they were fabbed on and more importantly where. They're also not general purpose processors, they're definitely dedicated number crunchers.

Nonetheless, the system is performing well, very well. They've been very circumspect about the chip specs though.

It's currently Intel, GloFo, Samsung and TSMC who can do 14-16nm chips. SMIC in China does have 28nm chip capability, however. For reference, Haswell chips were on 22nm (Intel had a 32 and 22 node, but no 28 node). They are behind, but they are not that far behind.

So there used to be a lot of supercomputers running microsofts systems. And they had problems. And there were stock exchanges running microsofts systems. And they had problems. I used to use their systems. I had problems. And a long time ago I sought out and found other systems, eventually landing on Linux. And my problems went away. And the stock exchanges have tried a few other systems, and eventually landed on Linux. And their problems have gone away. And the worlds supercomputers have a list of what they run. Its here. And if you select operating system family from the category box and then click on the submit button, you will see what the worlds supercomputers are running. Go ahead, I double-dog-dare ya. How much customer satisfaction comes out of an information systems manager (Infotainment division), when he accepts a large retirement donation in exchange for using a system that gives customers no satisfaction?

Windows 10 has ~21% but I guess MS had to force upgrades to even get those numbers.

Nice to see Windows 7 holding steady at 42%

The funny thing is MS had a 20 years head start on mobile with WinCE and consumers STILL didn't want it. Apple and Google come along and they accomplish in less then 5 years (Android was released on Dec 6, 2010) what MS couldn't do in 20 years!! LOL

Microsoft still has a ways to go when 2 years ago Linux run on over 1 Billion Devices and 99.4% of the Top 500 supercomputers run Linux; hell even iOS had 800 million back then.

But keeping MS, because you're (slowly) becoming irrelevant.

It's popular on Slashdot to loudly pronounce that strong AI is impossible. This is different from years past. I take the change to mean that it's coming very soon. As it seems more inevitable people who don't want change (whether out of fear, distrust, or sour grapes) will decry AI more.

Now weak AI isn't just coming. It has arrived. And Moore's law was supposed to have stopped years ago but supercomputers and video cards are still on a logarithmic slope for performance and price. The human brain is estimated to calculate between 100 petaflops and 1 exaflop. I know that's not a good metric but for this purpose it suffices. But as performance keeps doubling and doubling it becomes more evident that even the highest estimates are a question of a few more doubling periods. And the highest estimates assume direct one-to-one simulation of each neuron. Consider how many neurons are used for breathing, processing vision, and other things that either aren't needed in a machine or have already been done at a much lower computational cost on silicon.

It's true we don't know everything about how the human brain works. But recent progress is undeniable in terms of success stories. Jeopardy. Go. Commodity trading. Corporate resource balancing. Piloting. To keep shouting that strong AI is impossible is to only betray one's own insecurity. You are not special. Your brain doesn't run on quantum magic. You have no soul. Fucking deal with it.

http://www.ft.com/cms/s/0/fe7e...

http://www.bbc.com/news/techno...

http://top500.org/news/china-t...

http://www.theverge.com/2016/6...

http://www.netlib.org/utk/peop...

Or my (rejected) submission at https://slashdot.org/submissio...

Well, from http://www.top500.org/statisti... it seems machines 1-10 are all running Linux.

Oh, and all the machines from 11-209 are running Linux as well.

Only numbers 210, 211, 302, 420, 487 and 488 are not running Linux of the top 500 supercomputers in the world. Those all run other variants of Unix.

It is market leader in the server world, having a market share of 100% in supercomputers,

100%??? I call bullshit. it has 100% of the top 10 and the vast majority of the top 500 but Linux most definitely does NOT have 100% marketshare in supercomputers.

OP pretty much got it right, it's actually 99%. Check for yourself.

if that were true, then why does the NOAA need some of the biggest iron in existence JUST TO RUN CLIMATE MODELS??

#2 at Oak Ridge! If you can do it on a fucking laptop WHY DO THEY NEED THAT??

YOU get YOUR head out of YOUR arse!

You're forgetting one tiny little thing:

Us geeks influence friends and family.

I got my whole family switched over to Macs ~ 10 years back. It was an dead easy "sell" as my Dad was so fed up with Microsoft's constantly nickeling and diming. IF Microsoft would be reasonable and sell Windows (licenses) for $20 instead of $200 Windows 10 Pro USB flash drive then MAYBE people would stick with them but that ship has sailed LONG AGO in our family. Macs are just easier to use & support for non-technical people.

I am sticking with Windows 7 Pro only because of a few Direct3D 11 games and I need a version of Windows for game dev that supports my 32 GB RAM. But future Microsoft upgrades? Fuck 'em. I already have used two MacBook Pro's from work that I've been using for the past 4 years and absolute love them; I just bought my own MacBook Pro once I found out I can I can use my GTX 980Ti as en eGPU.

I tell my friends, "Sorry, I no longer support Windows past ver 7. I recommend a Mac or Linux box unless you have a specific reason you _need_ Windows. Microsoft has no respect for your privacy nor for empowering the UI. Why would you continue to give them money to allow them to constantly abuse you??"

I work for a Fortune 50 company -- you would be surprised at how many Mac's we have here. WAY, WAY, more then I would ever expect. The IT guys love Mac OSX significantly more then Windows. Hell, we're STILL upgrading Windows XP systems to Windows 7 systems.

The harder MS pushes Windows 10 the more people they are going to piss off. Push too hard and they could find themselves irrelevant much like IBM is today. Don't think this is a possibility? Consider the facts:

Android is already used by over 1.4 Billion people.
* http://www.theverge.com/2015/9...

Linux powers 98.8% of the top 500 supercomputers in the world; this is something Microsoft can only *dream* about.
* http://www.top500.org/statisti...

Free is eventually going to win over corporate greed. It doesn't really matter how long it takes; Microsoft's days are numbered. Maybe not today, maybe not tomorrow, but 20 years down the road MS will be struggling to maintain any sense of relevancy.

So yeah, MS _does_ undermine themselves. They are just to big to recognized it and they don't care. That's fine. I'll just take my business elsewhere. Apparently a lot of us geeks do as well.

--
Microsoft Windows 8 and 10, noun: A 64-bit compilation of 32 bit extensions and a graphical shell for a 16 bit patch to an 8 bit operating system originally coded for a 4 bit microprocessor written by a 2 bit company that can't stand 1 bit of competition with 0 bit of understanding good UI.

Exponential growth has stopped? Perhaps, if you're talking about Moore's law. CPU speed hasn't increased much for about 10 yrs. But there are multi-core processors now. Take a look at the list of the fastest computers. (see below) What would you call a chart plotting it's performance? I see something approximating exponential growth. (7 doubles in 10 years)

The key concept to grasp is not the Kurzweilian AI and human/robot mind melds. The key concept is that exponential growth is a hard thing to grasp. Our mental models are linear .

2015 - FASTEST (RMAX) 33,862.70

2005 - FASTEST 280.6

http://www.top500.org/lists/

It is amazing to recall that the world's top supercomputer ASCI Red from 1997 to 2000 was only capable of just over 1 TFLOP.

http://www.top500.org/statisti...

That's some hobby.

98.8% of the top 500 supercomputers run linux. The other six run unix. Not on the list: Windows.

Also worthy of note:

For the sixth consecutive time, Tianhe-2, a supercomputer developed by China’s National University of Defense Technology, has retained its position as the world’s No. 1 system, according to the 46th edition of the twice-yearly TOP500 list of the world’s most powerful supercomputers.

(Source: Top 500 lists November 2015)

Supercomputers are fundamental to leading edge scientific research.

With the exceptions of the IBM machines, which held most of the top spots in 2012. The LLNL one is still #3... http://www.top500.org/list/201... and the ANL #5. The general idea was cheaper processors that doesn't do pipelining, but OTOH the cost of them is massively decreased and they all had an instruction unit capable of two threads per core so the onus is on the programmer to make sure that the alu was kept constant fed.

Uh a few... http://www.top500.org/

That is definitely one area Linux is eating Microsoft for lunch!

Servers.

Gee, why are 97.6% (to be exact) of the top 500 super computers all running Linux? :-)

* http://www.top500.org/statisti...

Again in mobile, Android makes Windows Mobile look like a joke.

Microsoft is going to end up like IBM in 20 years if they aren't careful. Still around, but most people go, meh.

The Tianhe-2 Cluster computer has 3,120,000 cores and handily beats all three of them on geekbench. So when it comes to disingenuous comparisons of products that aren't really in the same market segment, Microsoft Google and Apple all lose!

Supercomputing? HAH!Linux is so fucking weak that it can't even support my super-server with 12 GPUs PER BOX.

http://www.top500.org/statisti...

In 2011 Cray deployed a Linux supercomputer with 9600 GPUs and from that point on Linux has been able to handle essentially infinitely many. I think you may want to read up a bit.

The Europeans, that's who. So it's not surprising to find the US in second place...

Behind China, but ahead of Japan and European countries.

Better than 2000's ASCI White, but worse than 2002's Earth Simulator. 13 years back to the past!

Or maybe the actual performance is 16 PetaFLOPs, as the linked article states.

Better than 2000's ASCI White, but worse than 2002's Earth Simulator. 13 years back to the past!

Or maybe the actual performance is 16 PetaFLOPs, as the linked article states.

Well, congrats to NOAA for the milestone...but isn't a peta-scale system a thing of the past for something as computationally intensive and critical as weather prediction? The prediction model seems to have no scalability issues: "The Hrrr model produces output from the model every 15 minutes versus the previous hourly rate,..", meaning 4x speedup for a 4x upgrade in peak performance. Why didn't NOAA go for a peta-scale system straightaway?

The upgraded systems are already placed way down the list of world's fastest 500: http://top500.org/system/17783....

The article says a peta-scale system is under way but I'm afraid by the time it starts ringing its bells, we'll all have entered exa-scale. Is NOAA always going to be catching up?

I notice that of the top 500 fastest computers in the world, 97% run Linux. http://www.top500.org/statisti... That would include the $390 million Tianhe-2. Oak Ridge National Laboratory spent $60 million to upgrade the $104 million Jaguar to become Titan, both running Linux. So yeah, if you're definition of "cheap" includes "the most expensive and powerful in the world", I guess you're right.

It's not that specialized. It's just plenty of DSPs strapped together on a torus.

Actually Anton uses ASICS, their cores are specially geared at MD codes. This goes way beyond just "strapping together DSPs". They have IIRC ~70 hardware engineers on site. (Source: I've been to DE Shaw Research last year).

Unlike what wikipedia claims, you could probably achieve comparable performance using a more classical and general-purpose supercomputer setup with GPU or Xeon Phi accelerators, provided the network topology is well tuned to address this sort of communication scheme

No, you can't, and here is why: Anton is built for strong scaling of smallish, long running simulations. If you ran the same simulations on a "x86 + accelerator" system (think ORNL's Titan) then you'd observe two effects:

• The GPU itself might idle a lot as each timestep only involves few computations, leaving many shaders idle or waiting for the DRAM.
• Anton's network is insanely efficient for this use case. IIRC it's got a mechanism equivalent to Active Messages, so when data arrives, the CPU can immediately forward it to the computation which is waiting for it. That leads to a very low latency compared to a mainstream "InfiniBand + GPU" setup.

(most recent supercomputers don't use tori)

Let's take a look at the current Top 500:

• #1 Tianhe-2: Fat Tree
• #2 Titan: 3D Torus
• #3 Sequoia: 5D Torus
• #4 K Computer: 6D Torus
• #5 Mira: 5D Torus
• #6 Piz Daint: 3D Torus
• #7 Stampede: Fat Tree
• #8 JUQUEEN: 5D Torus
• #9 Vulcan: 5D Torus
• #10 nn: 3D Torus

So, torus networks are the predominant topology for current supercomputers.

> The government could probably mine 30000 coins in thirty minutes on spare CPU power on some defense computer system.

No they can't. The combined hashing power of the bitcoin network is 1.15 million Petaflops. The top 500 supercomputers in the world have a combined power of 250 Petaflops ( http://s.top500.org/static/lis... ). Even if there are hidden NSA machines or whatever, they don't account for that big a discrepancy.

The reason the bitcoin network is so much faster is they use custom chips that do nothing but the SHA1 hash calculations used in mining. It's directly wired into the transistors, they can't be used for other math. But you can pack tens of thousands of copies of that algorithm on one chip, and mining rigs have boards with ~36 chips each, and multiple boards.

Also, the network difficulty adjusts every 50,400 coins (2016 blocks @ 25 coins per block), so if someone *could* mine that fast, the difficulty adjustment would bring it back to the normal rate (300 coins per hour) rather quickly. Lastly, that high a mining rate would be considered a "51% attack" on the network, and the rest of the network would rapidly reject the blocks and fork the history.

You guys keep working on that. Meanwhile Apple will continue selling millions more Macbooks and Mac Pro's to hard core developers, scientists and engineers who have work to do and need a computer to get it done with.

You do realise that almost all of the top 500 supercomputers run Linux

You do realize that even if all of the top 500 supercomputers ran Linux, it would still only be a mere 500 computers doing real work, compared to millions of Macs?

You guys keep working on that. Meanwhile Apple will continue selling millions more Macbooks and Mac Pro's to hard core developers, scientists and engineers who have work to do and need a computer to get it done with.

You do realise that almost all of the top 500 supercomputers run Linux

In other words, unlike GPL, it's not viral. It's this viral aspect of GPL that is turning people against it, and towards the more permissive BSD and MIT licenses.

Which is why we just heard Linux running out of funds? Oh wait, we didn't. And WebKit is LGPL (and not an Apple creation but a fork of KHTML - undoubtedly refined since though).

(For the record, I've nothing against BSD-licensed software, but people seem to be fine with GPL and its derivatives. Linux seems to be the platform of choice for most of smartphones and completely owns supercomputing. The desktop part is missing, insert compulsory joke about "YEAR OF LINUX DESKTOP" here - but generally it seems GPL is not scaring people away. And yes, even you, running Safari on your i-Device - you're running LGPL software.)

• * The per-clock throughput of CPUs has been on an exponential increase curve, per last decades, ie. last 20-30 years matter a lot
• * The proliferation of silicon compute devices has been on the increase for decades, ie last 10-20 years matter a lot (from the very small to the very big)
• * The duty cycle of most silicon devices is not 100%; there are two notable exceptions: infrastructure (embedded) systems and supercomputers
• * Parallel computation implies fast interconnects for quick message passing; for some initial info, ref. https://computing.llnl.gov/tutorials/parallel_comp/
• * Despite regular interconnect technology upgrades, MPI has been the agreed standard for message passing since the early 90s

ref. http://www.top500.org/ ->Statistics for more details; so, here we talk about big machines, of high duty cycle, of using mostly a uniform API for synchronization.

Given Weather, Climate & Computational Fluid Dynamics codes (and some more), the temporal density of such calls is pretty good.

Concluding, MPI should be the most common *API* being called nowadays per unit of time;
there is still room for challenging this though: MPI Send/Recv calls have a few variants plus,
MPI stack implementors may have fragmented the codebase, to declare a clear winner...

Apparently you haven't heard of an operating system called Linux then. Say you had a tiny little smart phone with a low power chip and wanted an operating system for it. Or say you wanted the fastest supercomputer in the world and wanted an operating system for it. The little smart phone lets call Android and the supercomputer lets call Tianhe-2 . Then you would be informed to know these operating systems are built with exactly the same LINUX code! Its not a different operating system for the phone or the supercomputer, its.the.same.code. Now you know. Oh, and I left the link up so you can see for yourself. Are you ready? If you go to the statistics tab, then "list statistics" and select "operating system family" and click the big 'submit' button, you might get skooled about Linux on supercomputers. Or you can be chicken.

That'd be the value over which some country's national labs or universities turn on their supercomputers and mine almost everything

The current Bitcoin mining network operates at twenty times the aggregate processing power of the entire TOP500 supercomputer list.

For comparison, today's most powerful commercially available ASIC-based Bitcoin mining rig, the KNC Neptune, pushes 3000GH/s. Using the semi-official conversion of 12,697 Flop/Hash (somewhat fuzzy given that BTC mining involves no actual floating point operations, but that number gives a fair estimate of the processing power involved), we come up with a whopping 38 PFlop/s. That puts the Neptune - a single unit, all by itself - solidly above the #2 supercomputer in the world (and above #1 by some metrics, since the Neptune can actually sustain that while the #1 can only sustain 34 PFlop/s).

No government - Not even the US government - Has the ability to flip a switch and monopolize BTC mining. They would need to buy (or make) rigs just like the Neptune and compete will all the other people running them (admittedly rare, but 1/5th that much power, 600+ GH/S ASIC rigs, have become relatively commonplace). At best, a government willing to throw infinite resources at the project could push the cost of mining above the cost of the electricity needed... Not much of a victory, there, since that government would itself need to use the electricity needed to go block-for-block vs the entire rest of the network, at a net loss.

> Once the last coins are mined, what happens?

Miners also get transaction fees included in the block, so they still have incentive to search for block hashes.

> how do you set fee levels such that people will validate little transactions as well as big ones?

Fees are *user defined*. If you are in a hurry, slap a big fee on it. If you can wait, put a small fee on it. The bitcoin software allows zero-fee transactions if you meet some conditions, and some miners will include zero fee transactions in a block if there is room. In the long run, many small transactions will move "off chain". For example, Coinbase processes bitcoin transactions for merchants, accepting BTC payments on their behalf and depositing local currency to their bank account. Conversely they sell BTC to individuals and pull money from bank accounts to pay for it. When a Coinbase user uses their online wallet to pay a Coinbase merchant, that can be all internal to Coinbase, and never reach the Block Chain. Eventually such processors will arrange to settle with each other in bulk transactions, gathering up many little ones and posting it as one big transaction on the block chain. That both avoids bloating the block chain, and makes small transactions easier to process.

> the value over which some country's national labs or universities turn on their supercomputers and mine almost everything

Custom chips (ASICs) that do nothing but the particular calculation for bitcoin mining are 100 times more efficient than GPU's, and 1000's of times more efficient than general CPUs. Supercomputers are useless for bitcoin mining. The world's top 500 supercomputers combined process 250 million GFLOPS ( http://www.top500.org/statistics/perfdevel/ ). The Bitcoin Network at the moment is running at 78 billion GFLOPs, which is over 300 times faster. Custom hardware and monetary incentive wins big time. The downside is the ASICs are absolutely useless for any other calculation, because it is hardwired into the chip.

It wouldn't be very different than the most powerful supercomputer in the world: http://top500.org/system/177999