a dozen cpu's on a chip

On a sunny day (Sun, 18 May 2008 11:15:54 -0700) it happened John Larkin wrote in :

I went through all links, but this one had that word 'paradigm' in it. I vaguely remember a discussion in Dr Dobbs (magazine) about software paradigms. Now, being rusty.. .I typed it in the English online dictionary, paradigm FORMAL a model of something, or a very clear and typical example of something

mmm, yes. Anyways sure assigning different parts of a signal processing chain to separate cores can make sense. As can vector operations, if the same operation has to be done on a stream of data. That is also where Cell comes in. But very often the data rates are so different for the different processes that a single core would work just as well. Not to mention, like for the ARM soft, 'will be coming in 2008?' The compilers and the OS lag behind so to speak. Or can not work for this new 'paradigm'?

Reply to
Jan Panteltje
Loading thread data ...

cores

data.

I've has situations where I really wanted to run a fairly simple process, say, 10,000 times a second, but I wanted it to run periodically, every time. So no other ISR or non-interruptable process could ever own the machine for more than about 40 microseconds. It would be so much nicer if that fast thing ran on its own core, and dumb stuff like ethernet and routine analog scanning and user interface had another processor or three to do their things in.

The compilers could be vanilla ansi C; a shared data region isn't hard to understand. And most realtime embedded systems don't need an OS at all. With multiple processors - no multitasking! - we have even less need of an OS.

John

Reply to
John Larkin

On a sunny day (Sun, 18 May 2008 13:33:40 -0700) it happened John Larkin wrote in :

Novell real time linux has a latency of 30us it seems (have not used it).

formatting link
I have played with preemptive and non-preemptive kernels though, and stability is not always the same (it is just a compile option for the kernel, but interrupting a driver interrupt etc is a bit scary, they modified everything but some drivers maybe not yet...???). Somebody correct me, but I disabled that option in kernel config, and have had fewer problems since.

Sure, but often you can simple use a separate PIC to do things that need be real time. The typical example is something on RS232 that logs data, and has its own buffer. A PIC, some RAM, but you could use part of a multi core chip too. Availability, price... you know.

You mentioned state machines. State machines can also work in a multitasker, for example SMTP client (the late Mr Postel's invention) is basically a state machine. I wrote one. This newsreader, in its essence, is a state machine, surrounded by a data base with data access and user interface.

But I know what you mean, no switching, dedicated chip. But maybe indeed, if no high speed is required (Internet is not really that fast, DSL), cheap small PICs could do the job with a lower budget?

Reply to
Jan Panteltje

But that's on a power-hogging Intel Pentium-level CPU. I'm talking about that sort of performance on a 1-watt, $10 Coldfire type chip, a typical embedded application.

kernel,

real time.

buffer.

If you need multiple CPUs, why put them in separate packages, with all the related interfacing problems? Not to mention cost.

Everything is a state machine. It's just a matter of who understands it.

fast, DSL),

Do you know of a small, cheap PIC that will do a foreground app, run something special 10,000 times a second, and run a couple of TCP/IP stacks simultaneously?

John

Reply to
John Larkin

But they aren't.

The set of problems the cell solves are smaller than those a GP computer solves. There is a reason IBM doesn't use the cell processor in its products.

No, perhaps that's what you meant but you said exactly the opposite. Cause effect.

Nonsense. The current CPUs look nothing like the 8008.

1) Because they can. 2) There will always be more things to do 3) Finer grained control allows a more responsive system.

Exactly my point. ;-) My vote is for PL/I (or perhaps even Ada).

--
Keith
Reply to
krw

Be careful WHAT you're trying to prove.

Nonsense (recent FPUs aren't significantly more complicated than the '87). Formal proofs save time and money but they're not the end- all. We used them to prove we hadn't installed bugs on an existing design but new designs weren't proved. Change the pipe and the proof gets ugly.

--
Keith
Reply to
krw

You're an absolute idiot!

Dimbulb, you take the cake!

Stupid bytestard.

Yet they're made on the same line, "with 100% yeilds". You're as stupid as a stone, Dimmie.

--
Keith
Reply to
krw

They only wish. Dimbulb is completely clueless.

--
Keith
Reply to
krw

It is more likely to change the way that application programs are designed to take advantage of the multicore environment. I doubt if home users will ever see boxes with N>4 in (apart from gamers).

You don't seem to appreciate why your "idea" will not work.

formatting link

Speculative thread execution goes way beyond anything you have imagined.

It is ironic that of all the things that would improve home and office experience of computing some extra hardware acceleration for the JVM used by clientside web applications would be more use than multicore.

Regards, Martin Brown

** Posted from
formatting link
**
Reply to
Martin Brown

On a sunny day (Sun, 18 May 2008 15:27:17 -0700) it happened John Larkin wrote in :

fast, DSL),

I was trying to say that is such a case use more then one PIC. But it is all theory, once we have a real application a choice can be made. I am not for or against one or the other, sure if you have the processor power and it can be done multitasking do it! full circle :-)

I suppose one could sell a PC with the application (happens), but could one sell a playstation 3 running YellowDog Linux with a scientific piece of equipment (as part of it)? Now that would require user education ;-)

Reply to
Jan Panteltje

take advantage of the multicore environment. I doubt

Again, whether it makes sense performance wise is irrelevant, >4 cpu configurations will show up in home boxes in 3-5 years.

Hell, how many buy quad cpu boxes today just because salesguy told them it was better. Will they ever need that performance? Almost certainly not.

M
Reply to
TheM

^^^^^^^^^^^

SALESDROIDS! ^

A real Druid would never intentionally have anything to do with the likes of Intel/MICRO$~1. >:->

Cheers! Rich

Reply to
Rich Grise

take advantage of the multicore environment. I doubt

Actually quads are showing up now.

Reply to
JosephKK

On a sunny day (Mon, 19 May 2008 17:35:02 GMT) it happened Rich Grise wrote in :

You are absolutely right, but I actually did mean 'druide'. Well, my only experience with druides is from the comic strips Asterix and Obelix, dunno if these exist in the US. Anyways, that druide brews a magic potion that gives those Galliers (part of France) incredible strength so they can beat up the Romans (Cesar).

Now the sales-druide does some magic too. You maybe be perfectly happy with what you have, but the real sales-druide (from here one SD) will make you think you need something to feel better. This is called 'creating a need'. For example you have good TV, the SD will make you think you need HDTV, even if you are short-sighted, have a too small living room, and there are no HD transmissions except for 30 year old noisy celluloid copies.

This is magic, same as a real druide,. The real druide used a potion, made of herbs... The SD uses advertising.

You may be happy, and not even use all 4 (cores ;-) ), but the SD will show you a picture of an 80 core chip, and sure soon you be having wet dreams about 300 cores and more.

And no truth involved.

I remember my first experience withe Intel, I needed a floppy controller chip for some project, many companies made those, but the Intel specified a feature where you could search for one byte in a sector. I thought : mmm maybe I can use that. ordered the chip, but could not get that feature working. Called the rep, 'we will send you an errata'. And in the errata it said: "This feature is not supported in the new silicon xxxA'. But now I had weeks of work invested in this chip, did it in software, and was stuck with Intel.

So no truth involved,

I dunno if all real druides cannot be made to bend the truth a bit for some profit... Human nature (are druides human? has it so you can buy almost anyone. And you can make almost anyone believe anything. Yes in a way selling is magic :-)

I looked up 'druid' too: druid noun a priest of a religion followed in Britain, Ireland and France, especially in ancient times.

Yet something different :-) Languages can be facinating.

Reply to
Jan Panteltje

This was announced recently:

formatting link

So Los Alamos will use Cell in the fasted computer in the world if I understand this right.

Reply to
panteltje

formatting link

It is a very enhanced Cell processor.

Reply to
Archimedes' Lever

ancient times.

Well, this doesn't really tell you much - the original Druids were "pagans", in that they worshipped trees. The tree, you see, is the heart of creation - it accepts energy from the Sun, and dirt from Mother Earth, and turns them into food, clothing, and shelter.

This is where the tradition of "knock[ing] on wood" came from - it invokes the good tree spirit, who protects you from the jinx. ;-)

I've started a new religion - I call it the "Neodruids".

Cheers! Rich

Reply to
Rich the Philosophizer

On a sunny day (Tue, 20 May 2008 21:47:46 GMT) it happened Rich the Philosophizer wrote in :

ancient times.

Yes, there clearly seems to be a parallel between the Dutch word 'druide', and 'druid'. I think it is the same. The 'druide' in the comic strip also is into trees and herbs, and wanders in the woods looking for those. Has a long white beard too.

jinx, an other word I have to look up: jinx noun bad luck, or a person or thing that is believed to bring bad luck: here's a jinx on this computer - it's gone wrong three times this morning!

jinxed adjective I must be jinxed - whenever I wash a wine glass, it breaks.

Well religions are a dime a dozen so to speak. Do neo-druids replace trees with computers? ;-)

Reply to
Jan Panteltje

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.