User avatar
kusti8
Posts: 3439
Joined: Sat Dec 21, 2013 5:29 pm
Location: USA

Re: Moore's law & the Raspberry Pi

Tue Dec 15, 2015 10:40 pm

http://www.nei.org/Knowledge-Center/How ... ctors-Work
First paragraph, reading in between the bias, nuclear reactors do use steam. I can't think of a way that they wouldn't. It's all the same principle. Something heats up water, it spins a turbine and makes electricity. The only thing that is different is what heats up the water.
There are 10 types of people: those who understand binary and those who don't.

BMS Doug
Posts: 3824
Joined: Thu Mar 27, 2014 2:42 pm
Location: London, UK

Re: Moore's law & the Raspberry Pi

Tue Dec 15, 2015 10:51 pm

kusti8 wrote:http://www.nei.org/Knowledge-Center/How ... ctors-Work
First paragraph, reading in between the bias, nuclear reactors do use steam. I can't think of a way that they wouldn't. It's all the same principle. Something heats up water, it spins a turbine and makes electricity. The only thing that is different is what heats up the water.
I don't think that anyone was denying that it was steam or offering other choices?
Doug.
Building Management Systems Engineer.

drgeoff
Posts: 9926
Joined: Wed Jan 25, 2012 6:39 pm

Re: Moore's law & the Raspberry Pi

Tue Dec 15, 2015 10:51 pm

W. H. Heydt wrote: Isn't there a project in Scotland to install a tidal power system? Or is that somewhere else and I have a memory leak?
Power from tides and power from waves are two different approaches.

W. H. Heydt
Posts: 11109
Joined: Fri Mar 09, 2012 7:36 pm
Location: Vallejo, CA (US)

Re: Moore's law & the Raspberry Pi

Wed Dec 16, 2015 5:32 am

drgeoff wrote:
W. H. Heydt wrote: Isn't there a project in Scotland to install a tidal power system? Or is that somewhere else and I have a memory leak?
Power from tides and power from waves are two different approaches.
I'm familiar with the difference. The recent article was using underwater turbines using tidal flows. The mentioned the idea of using them in the Bay of Fundy as well. There may be a project on someone's wish list to use the area under the Golden Gate Bridge, too. (Though I think Raccoon Straights might be a better choice.)

gsh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 1445
Joined: Sat Sep 10, 2011 11:43 am

Re: Moore's law & the Raspberry Pi

Wed Dec 16, 2015 7:21 am

Yes, precisely....

Intel didn't care about power because they could dissipate it away (using more complex system as time went on) but they never tried to change their process to reduce power. It just didn't figure in their calculations, which is why they have been unable to upset the ARM apple cart since they do not have and have never had any experience with reduce power processes or design rules to reduce power (clock gating, block gating etc).

I recently did a paper on this and come to the same conclusions as I did four years ago when Raspberry Pi first started. Moore's law has stopped so you are no longer seeing an automatic reduction in transistor size (previously the reduction in transistor size would actually reduce the device size enough to reduce it's cost enough to outweigh the extra costs for the wafer plus the design costs, with the move from 28nm to 20nm this is no longer true.)

So in future advancements are going to come from people throwing away those terribly inefficient languages (javascript, java, python (although maybe a compiled version will be able to make it).

Also I think we'll see a change in the standard software architecture model to help developers move away from thinking of their programs in a uni-threaded model. But overall I think everything will become more parallel and less serial by necessity

Of course there is another way of getting electricity out of nuclear, which is thermocouples as in an RTG (check wikipedia for atomic battery)



Gordon
--
Gordon Hollingworth PhD
Raspberry Pi - Director of Software Engineering

User avatar
r3d4
Posts: 968
Joined: Sat Jul 30, 2011 8:21 am
Location: ./

Re: Moore's law & the Raspberry Pi

Wed Dec 16, 2015 8:34 am

Woll wrote:And no, I didn't think they relied on steam in nuclear power plants. The program was very interesting.
If you want more variation on that theme look up the "windscale documentary" :!:

jamesh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 24188
Joined: Sat Jul 30, 2011 7:41 pm

Re: Moore's law & the Raspberry Pi

Wed Dec 16, 2015 9:38 am

W. H. Heydt wrote:
jamesh wrote:
timrowledge wrote: I think that's actually not entirely true, since IIRC all those nicely radioactive elements were created in supernovae and so doin a sense stem from starlight.
But the starlight came from fusion....everything came from the big bang really.


And I should add that I completely forgot about Hydro, wind turbines, solar, and certain gas power stations when I said all our electricity came via stream turbines. Apologies.
Isn't there a project in Scotland to install a tidal power system? Or is that somewhere else and I have a memory leak?
AHhhh. I forgot tidal and wave power AS WELL!
Principal Software Engineer at Raspberry Pi (Trading) Ltd.
Contrary to popular belief, humorous signatures are allowed. Here's an example...
“I think it’s wrong that only one company makes the game Monopoly.” – Steven Wright

W. H. Heydt
Posts: 11109
Joined: Fri Mar 09, 2012 7:36 pm
Location: Vallejo, CA (US)

Re: Moore's law & the Raspberry Pi

Wed Dec 16, 2015 4:40 pm

gsh wrote: So in future advancements are going to come from people throwing away those terribly inefficient languages (javascript, java, python (although maybe a compiled version will be able to make it).
Maybe.... It's more likely to come from redesigning the run time engines (such as just in time compilation) to run as a separate process on a different core than the actual process than it is from abandoning interpreters and intermediate code systems.

The history of programming has been to move the programming farther and farther away from the machine. When I started out, a lot of business programming was done in assembler languages. The first shop I worked, not only were we writing in assembler (as well as other languages, mostly COBOL), but there was "legacy" code running on the IBM 1401 emulator on a System/360. That code was written the assembly language of the 1401.
Also I think we'll see a change in the standard software architecture model to help developers move away from thinking of their programs in a uni-threaded model. But overall I think everything will become more parallel and less serial by necessity
Perhaps. Not all tasks can be made parallel--easily, or at all. The more likely scenario is where, unless absolute performance is crucial, is to just run a lot more stuff at the same time to use the additional hardware threads that are possible.

And, by the way, this brings up the possibility (particularly with ARM SoCs) of going back to a trick CDC used in the 1960s on their 6000 series machines. To take the full example...the CDC 6600 had a fast 60-bit processor (the "CP") and 8 somewhat slower 18-bit processors ("PPs"). The main task of the PPs was to handle I/O so the CP didn't have to. However, PP0 ran the OS, so all of that was off-loaded from the CP. Suppose one did a LITTLE.big configuration and ran the Linux kernel on a "little" core, leaving ALL of the big cores to run user processes?
Of course there is another way of getting electricity out of nuclear, which is thermocouples as in an RTG (check wikipedia for atomic battery).
Yeah...but it doesn't appear to scale very well, though it's a quite good method for a system that is hard to get at to effect repairs for long periods of time.

Woll
Posts: 472
Joined: Mon Jul 06, 2015 2:14 am
Location: Cloud Cuckoo Land

Re: Moore's law & the Raspberry Pi

Wed Dec 16, 2015 4:43 pm

What about graphene? Are all the pridictions so far based around current technology. Graphene is supposed to revolutionize the future.
Who ate all the Pi's? Who ate all the Pi's?
Boris Johnson, Boris Johnson,
Who ate all the Pi's?

mikerr
Posts: 2789
Joined: Thu Jan 12, 2012 12:46 pm
Location: UK
Contact: Website

Re: Moore's law & the Raspberry Pi

Wed Dec 16, 2015 4:47 pm

gsh wrote:Also I think we'll see a change in the standard software architecture model to help developers move away from thinking of their programs in a uni-threaded model. But overall I think everything will become more parallel and less serial by necessity
Been thinking this for the past 20 years, but apart from the likes of CUDA, not a lot of general purpose computing is done greatly parallel.
Android app - Raspi Card Imager - download and image SD cards - No PC required !

User avatar
Burngate
Posts: 6100
Joined: Thu Sep 29, 2011 4:34 pm
Location: Berkshire UK Tralfamadore
Contact: Website

Re: Moore's law & the Raspberry Pi

Wed Dec 16, 2015 4:49 pm

Don't know whether you'd count Ffestiniog among your sources of electric power in the UK (it's a sort of secondary, rechargeable battery)
Built in 1963. Interesting that it's only about 75% efficient, but that's enough to make money.

Also interesting is that one of the earliest terrestrial Uranium reactors was in Gabon, Africa. About 1.7 billion years ago, 100kW, light water as the moderator, so there'd be at least some steam.

User avatar
Burngate
Posts: 6100
Joined: Thu Sep 29, 2011 4:34 pm
Location: Berkshire UK Tralfamadore
Contact: Website

Re: Moore's law & the Raspberry Pi

Wed Dec 16, 2015 4:51 pm

Just a thought: doesn't having a larger word-size effectively speed things up - in which case when will we see a 128-bit Pi?

User avatar
rpdom
Posts: 15604
Joined: Sun May 06, 2012 5:17 am
Location: Chelmsford, Essex, UK

Re: Moore's law & the Raspberry Pi

Wed Dec 16, 2015 5:10 pm

Burngate wrote:Just a thought: doesn't having a larger word-size effectively speed things up - in which case when will we see a 128-bit Pi?
I guess that would depends on when someone makes a 128 bit ARM (or perhaps switch to a different CPU?).

64 bit ARMs do exist, but doubling the data bus width also doubles the number of traces you'd need to route between the CPU and RAM, adding complexity and cost to the PCB design.

Also, increasing the word size only speeds up certain operations. If you're just prodding one bit it doesn't matter if your CPU handles 4, 8, 16, 32 or 1024 bits in one go. It can help with a lot of maths functions though.

W. H. Heydt
Posts: 11109
Joined: Fri Mar 09, 2012 7:36 pm
Location: Vallejo, CA (US)

Re: Moore's law & the Raspberry Pi

Wed Dec 16, 2015 6:01 pm

Burngate wrote:Don't know whether you'd count Ffestiniog among your sources of electric power in the UK (it's a sort of secondary, rechargeable battery)
Built in 1963. Interesting that it's only about 75% efficient, but that's enough to make money.

Also interesting is that one of the earliest terrestrial Uranium reactors was in Gabon, Africa. About 1.7 billion years ago, 100kW, light water as the moderator, so there'd be at least some steam.
There is--or was, I don't know if it's still in use--a facility in California in which the station could pump water to a high up lake when there was a surplus of other power being generated, and then run the water back down through turbines to generate power when needed. A "secondary" power source, but on a scale that would be extremely difficult to match with batteries.

W. H. Heydt
Posts: 11109
Joined: Fri Mar 09, 2012 7:36 pm
Location: Vallejo, CA (US)

Re: Moore's law & the Raspberry Pi

Wed Dec 16, 2015 6:10 pm

Burngate wrote:Just a thought: doesn't having a larger word-size effectively speed things up - in which case when will we see a 128-bit Pi?
I don't know if *anyone* is doing 128-bit processors (ignoring for the moment double-precision arithmetic on 64-bit CPUs, which does exist). The possible exception might be GPUs, but there, the real effort to get around bottlenecks is to increase *memory* bandwidth by expanding the bus--and trace count--between the processor block and the RAM. This has gone to at least 512-bit fetches.

Having a different memory fetch size and processor word size in nothing new. IBM was doing it in the mid-1960s. The 360/30 had an 1 byte fetch, the 360/40, 2 byte, and on up. IIRC, the top machines,--360/95, 370/195--had an 8 byte fetch. This was driven by the mismatch between memory speed and processor speed as they were *all* 32-bit machines. Intel did something similar on the low end. The 8088 (think original IBM PC) was really a 16-bit CPU, but it used an 8-bit memory bus. It was designed that way to ease the transition from 8-bit to 16-bit designs. It was really just a modified 8086 chip.

timrowledge
Posts: 1290
Joined: Mon Oct 29, 2012 8:12 pm
Location: Vancouver Island
Contact: Website

Re: Moore's law & the Raspberry Pi

Wed Dec 16, 2015 6:17 pm

W. H. Heydt wrote: There is--or was, I don't know if it's still in use--a facility in California in which the station could pump water to a high up lake when there was a surplus of other power being generated, and then run the water back down through turbines to generate power when needed.
Yup, that's the same thing as the Ffestiniog water pumping facility mentioned above. There's also Dinorwig not far away which does the same thing but about 4x bigger. The California facility I know of is the Big Creek one, which is about 3X bigger than the Ffestiniog. Right now it apparently has problems because there is so little water in the dams that it can't really run at all.
Making Smalltalk on ARM since 1986; making your Scratch better since 2012

timrowledge
Posts: 1290
Joined: Mon Oct 29, 2012 8:12 pm
Location: Vancouver Island
Contact: Website

Re: Moore's law & the Raspberry Pi

Wed Dec 16, 2015 6:56 pm

gsh wrote:So in future advancements are going to come from people throwing away those terribly inefficient languages (javascript, java, python (although maybe a compiled version will be able to make it).
It's not so much the language as the actual executed code that is the issue. Even COBOL could be executed efficiently with good enough compilers and given how important it still is in business I'd be startled if there hasn't been a lot of effort. Then again, logic in the business world is a rare beast.
As an example of this we could consider Smalltalk; the original version was written as a very simplistic interpreter implemented on an interpreted BASIC. Unsurprisingly it wasn't terribly fast but then nothing was back in 1970. Then a rewrite to build a compiler of Smalltalk (written in Smalltalk, of course) plus a new virtual machine written in Alto assembler (because there was no C back then) made things much faster. Then new hardware was built - one advantage of a research lab with a big budget and a lot of smart people - and the vm rewritten to take advantage of the stunning performance of a 70MHz 16 bit cpu. In '84 the concept of the dynamic translator was born and Smalltalk got much, much faster.
We recently implemented the latest iteration of that on the Pi of course and the most visible result for Pi users is a large increase in Scratch performance. Another improvement is likely to arrive soon when the SiSTA optimiser is released; one of the nice things about a properly object oriented and reflective language is that you can do things like writing the compiler in the same language in the same running system and thus have access to the running code to analyse and optimise at run-time rather than statically. We think it ought to roughly triple performance, without making for undebuggable systems that are pickled into automatic obsolescence.
gsh wrote:Also I think we'll see a change in the standard software architecture model to help developers move away from thinking of their programs in a uni-threaded model. But overall I think everything will become more parallel and less serial by necessity
Actual parallelism is really important; not the half-hearted 'threads sharing a few cores with a single memory bus' stuff that has been thrust upon us by a combination of intel's desperate thrashing to keep getting faster without catching fire, and the software world's unwillingness to consider anything new.
The tragedy of the mid-80s was the failure of the transputer, smashed by intel's huge investment bucket making single thread cpus faster every few months. That appears to have lead to further tragedy as software writers gave up on trying to write efficient code and simply drank the Moore's Law kool-aid and assumed that by the time their code hit the streets it would run ok on the faster machines sure to be out there. Aside from making the Pi Smalltalk use dynamic translation (which gave us ~4X performance) I spent much of the last couple of years simply re-writing bad code in Scratch to be more effective and that produced another 5X to 10X depending on the exact tests; a good recent example of how careful engineering can produce performance independent of the language.
One of the things I like about Scratch as an educational system is that kids will have experience of thinking about parallel systems working by communicating rather than tired old purely procedural models. A future Pi based on dozens or hundreds of ARM cores *with separate memory spaces* would be a wonderful thing.
Making Smalltalk on ARM since 1986; making your Scratch better since 2012

W. H. Heydt
Posts: 11109
Joined: Fri Mar 09, 2012 7:36 pm
Location: Vallejo, CA (US)

Re: Moore's law & the Raspberry Pi

Wed Dec 16, 2015 10:01 pm

timrowledge wrote:Even COBOL could be executed efficiently with good enough compilers and given how important it still is in business I'd be startled if there hasn't been a lot of effort.
There have been optimizing compilers for COBOL around for decades. Something to keep in mind about business programs (where you are going to find COBOL) is that it tends to be I/O intensive (indeed, COBOL programs are generally "I/O bound" rather than "compute bound") and proper design of I/O operations and attention to how and where the data is stored (I could give you some real horror stories) will have a greater impact on how a program runs than most choices about the code.

Beyond that, I can personally attest that some COBOL programmers do appallingly stupid things in their code because they fail to understand what the compiler and CPU will do with what was written. This usually can be seen in how is held internally, something the compiler can't really do anything about. The common bad example I've seen over the years is to use a DISPLAY (that is, data held in human readable form--ASCII or EBCDIC characters) variable for subscripts (which *must* be binary values). This cause the compiler, (a) pack, (b) convert to binary, and then use the value *every* *time* the variable is used as a subscript, and to at least pack and unpack it every time it is changed. This is terribly inefficient, but many COBOL programmers do time and again and then wonder why their programs run slower than the ones written by programmers that don't do that.

I should note that, back when I was working on IBM mainframes, I would fairly often tell the compiler to produce a pseudo-assembly listing so that I could check the generated code to be certain it was doing what I wanted it to, and to look for any data optimizations that I might have otherwise missed.

User avatar
rurwin
Forum Moderator
Forum Moderator
Posts: 4258
Joined: Mon Jan 09, 2012 3:16 pm
Contact: Website

Re: Moore's law & the Raspberry Pi

Thu Dec 17, 2015 12:37 am

jamesh wrote:But the starlight came from fusion....everything came from the big bang really.
Incorrect. The big bang created only hydrogen. Stars create the next few elements and in their red giant phases create everything up to iron (IIRC) which can be distributed when the red giant collapses into a neutron star. The only way to get heavier elements is with a supernova.

IIUC, current thinking is that the solar system is unusual in having so many heavy elements when the universe is as young as it is. There must have been a fortuitous supernova close by very early on. The sun is 6 billion years old and the universe is only 13.82 billion years old. That leaves just 7.82 billion years for the big bang to result in galaxies and stars and for one of them to live its life and blow up.

Whether a supernova counts as sunlight is another argument.

User avatar
rurwin
Forum Moderator
Forum Moderator
Posts: 4258
Joined: Mon Jan 09, 2012 3:16 pm
Contact: Website

Re: Moore's law & the Raspberry Pi

Thu Dec 17, 2015 12:47 am

It may be that the data flow architecture could make a return.

Instead of n separate cores, you have n ALUs. The data drives the program. So if you have C=A+B, when A and B are both available they are added together and C becomes available. It's a very different architecture that requires very different thinking to work efficiently. Scaling it is not easy, which is probably why it got nowhere in the 80's.

User avatar
expandables
Posts: 654
Joined: Fri Jun 27, 2014 7:34 pm
Location: Neverland with Michael Jackson

Re: Moore's law & the Raspberry Pi

Thu Dec 17, 2015 12:49 am

rurwin wrote:
jamesh wrote:But the starlight came from fusion....everything came from the big bang really.
Incorrect. The big bang created only hydrogen. Stars create the next few elements and in their red giant phases create everything up to iron (IIRC) which can be distributed when the red giant collapses into a neutron star. The only way to get heavier elements is with a supernova.

IIUC, current thinking is that the solar system is unusual in having so many heavy elements when the universe is as young as it is. There must have been a fortuitous supernova close by very early on. The sun is 6 billion years old and the universe is only 13.82 billion years old. That leaves just 7.82 billion years for the big bang to result in galaxies and stars and for one of them to live its life and blow up.

Whether a supernova counts as sunlight is another argument.
Ah we got a lot of creationist on the forum :shock: .
:twisted:
By thinking like an engineer you can create a raspberry pi.
Michael Jackson enthusiast.
I got the PI model B, B+ and PI 2 model B.
When will I get the A? I don't know.

stderr
Posts: 2178
Joined: Sat Dec 01, 2012 11:29 pm

Re: Moore's law & the Raspberry Pi

Thu Dec 17, 2015 4:22 am

rurwin wrote:. The only way to get heavier elements is with a supernova.
Because anything heavier requires additional energy to create while the lower elements are created and liberate energy.
IIUC, current thinking is that the solar system is unusual in having so many heavy elements when the universe is as young as it is. There must have been a fortuitous supernova close by very early on. The sun is 6 billion years old and the universe is only 13.82 billion years old. That leaves just 7.82 billion years for the big bang to result in galaxies and stars and for one of them to live its life and blow up.
The bigger the star, the more quickly it lives out its life. If early stars were massive, they could've exploded very rapidly. And because they were so big, provided a lot of material for more star formation. I think I recall that galaxies are being found much closer to the Big Bang than previously thought.

dancytron
Posts: 56
Joined: Sun May 06, 2012 11:09 pm

Re: Moore's law & the Raspberry Pi

Thu Dec 17, 2015 5:17 am

W. H. Heydt wrote:
Burngate wrote:Don't know whether you'd count Ffestiniog among your sources of electric power in the UK (it's a sort of secondary, rechargeable battery)
Built in 1963. Interesting that it's only about 75% efficient, but that's enough to make money.

Also interesting is that one of the earliest terrestrial Uranium reactors was in Gabon, Africa. About 1.7 billion years ago, 100kW, light water as the moderator, so there'd be at least some steam.
There is--or was, I don't know if it's still in use--a facility in California in which the station could pump water to a high up lake when there was a surplus of other power being generated, and then run the water back down through turbines to generate power when needed. A "secondary" power source, but on a scale that would be extremely difficult to match with batteries.
That's what the Taum Sauk power plant in the Missouri Ozarks does. https://en.wikipedia.org/wiki/Taum_Sauk ... er_Station

Ten years ago (hard to believe its been 10 years), it failed due to the negligence of the operators and dumped a billion gallons of water on a State Park. They got the largest fine for a dam breach in U.S. history.

Every engineering student should read how the idiots managed to design it without spill way and then pump water over the top of the wall.

It has since been rebuilt with a much safer design, but the same idiots still run it. They talk on and off about building another one nearby. Hopefully the Tesla battery and other recent battery advances will make it uneconomical.

Return to “Off topic discussion”