How to Design a Self-Balancing Multicore CPU for Fine-Grain Parallel Applications

formatting link

From the Blog:

The Real Purpose of the CPU

The most important question a CPU designer must ask himself or herself is, what is the purpose of the CPU? Most people in the business will immediately answer that the purpose of the CPU is to execute sequences of coded instructions. Sorry, this is the wrong answer. This definition is precisely what got us in the mess that we are in. In order to figure out the answer to this crucial question, we must first grok the true nature of computing. And the sooner we learn that a computer is really a collection of concurrently behaving and communicating entities, the sooner it will dawn on us that the CPU is a necessary evil.

Necessary Evil

Why necessary evil? Because, ideally, each and every elementary entity in a computer should be its own self-processor, not unlike a neuron in a biological nervous system that sits around most of the time waiting for a signal to do something. In other words, ideally speaking, additions should add by themselves and multiplications should multilply by themselves. Unfortunately, we have not yet progressed to the point where we can build super-parallel systems like the brain with its tens of billions of tiny, interconnected, self-executing processors. That will come one day, but not today.

CPU as Simulator

Fortunately, what we lack in numbers, we make up in speed. We can have a single fast CPU do the work of thousands or even millions of elementary processors (cells). Ideally, adding more CPUs or cores into the mix should improve performance linearly. At least, that is the goal. So this, then, is the true purpose of the CPU, to simulate the behavior of a large number of behaving and inter-communicating cells. So, essentially, the CPU is a cell simulator, or cell processor if you wish. If the reader disagrees with my argument so far, there is no point in reading any further because what I am going to say below depends on what I said above.

Reply to
Traveler
Loading thread data ...

I thought Flying Skybuck had bought a grammar checker........

Martin

Reply to
Martin Griffith
[crap]

Yo, Griffith. Are the asshole in Berlin who's been visiting my blog everyday? ahahaha...

Reply to
Traveler

You are about 30 years late. Try reading something a little less flaky:

Dataflow Architecture

formatting link
formatting link

Artificial neural network

formatting link

then use google to educate yourself from qualified sources.

Eric

Reply to
Eric P.

Email addresses don't mean shit. You are the asshole using the "Individual Network Berlin" ISP. You don't fool me. ahahaha... If you don't like what I write, stop reading my blog. It's not meant for you. There are plenty of other blogs out there, ASSHOLE. ahahaha... AHAHAHA... ahahaha...

Reply to
Traveler

Come on, who on earth would want to read your blog, as you say there are much better blogs out there.

I suggest that you get on the waiting list for the Total Perspective Vortex

Martin

Reply to
Martin Griffith

You do. Asshole. ahahaha...

Reply to
Traveler

I see. You're just a know-it-all senile old fool. You old washed up geeks got us in the mess that we're in, in the first place.

Reply to
Traveler

Hey - don't disparage all the geeks here - as far as I know, I'm the only one to ever program an analog computer (at any rate, no one else ever owned up to it).

Perhaps you would feel more comfortable in a group like sci.physics? They are always interested in new concepts that push the envelope. Take Skybuck with you too.

And don't let the doorknob get stuck in your butt crack on the way out.

Eric

Reply to
Eric P.

You are not alone.

Ignore the maroons. "You old washed up geeks got us in the mess that we're in, in the first place" is a classic. Actually I had nothing to do with the mess you're in today there fellow traveler, unless you are running an AS400 or i-series.

Reply to
Del Cecchi

I'll wager that a large percentage of readership of comp.arch.embedded have programmed analog computers; I happen to have two of them upstairs...

Regards,

Michael

Reply to
msg

Besides the two analog computers upstairs, there is a a little AS/400

9402 downstairs ;)

Regards,

Michael

Reply to
msg

As one of the ancient geeks here I can also own up to having done a bit of analogue computer programming (back in 1970). Then, in my career I have done a great many things, all very interesting, I have also followed the work of Professor Igor Alexandra who was heavily into neural networks.

I was tempted to point traveler in the direction of one company that is doing a very fine grained 24 processor chip but having seen his general attitude and demeanor I am not sure I would want to now.

To conclude with an observation, it seems to me that we will see much more of mesh computing concepts coming in and improving from now on. There will be a range of capabilities across the market place and a few favourites will emerge and be widely used. I look forward to seeing how it all develops and getting my hands on some of this new hardware.

--
********************************************************************
Paul E. Bennett...............
 Click to see the full signature
Reply to
Paul E. Bennett
+--------------- | Hey - don't disparage all the geeks here - as far as I know, | I'm the only one to ever program an analog computer | (at any rate, no one else ever owned up to it). +---------------

Me too, me too!!! Circa 1965, multiple racks full of Philbrick integrators, programmed with plug-in capacitor & resistor "blocks" and interconnected with patch-cord wires with stackable banana sockets/plugs on the ends. Ahhh... The good old days!! ;-}

Of course, then we had to spoil it by connecting an Adage D/A converter and multi-channel A/D converter between all the Philbrick kit and an IBM 1410, which let us run some really fun hybrid analog/digital computations. But that's another story...

-Rob

----- Rob Warnock

627 26th Avenue San Mateo, CA 94403 (650)572-2607
Reply to
Rob Warnock

Really... wow. Commercial systems right (not something custom)? When were they built, and what are they used for now?

Eric

Reply to
Eric P.

Speak for yourself.

I've got to find a Systron-Donner for CHM (for chaos theory history). But we don't usually pay for artifacts.

Reply to
Eugene Miya

Hey, me too, in about 1963, a room-sized one with a zillion patch cords and banana plugs. We didn't have any fancy A-D's, just really really good XY pen and ink pin-feed plotters. I recall a "function" generator which used diodes and resistors to allow you to program arbitrary piece-wise linear functions.

Reply to
Tom Knight
[geriatric crap deleted]

Another senile old geek heard from. ahahaha...

Reply to
Traveler

They're just the Heath EC-1 (good for basic training but of very limited accuracy of course) ;) I believe the EC-1 was first offered around 1955. A synopsis is here:

formatting link

At least they're more accurate than this (photo I saved from a few years ago):

formatting link

Regards,

Michael

Reply to
msg

One of the most impressive rooms was the Hybrid Computing Lab at the University of Minnesota: twin EAI 630(?) analog computers connected to twin CDC 1700 systems with a Digigraphic Display in the adjacent darkened room (approx. 30" flat circular screen). A favorite pastime was Spacewar (modified from the original which IIRC came from GM) with a rich accompaniment of sounds made by the patches on the analog side ;)

Regards,

Michael

Reply to
msg

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.