a dozen cpu's on a chip

It actually doesn't matter much to me. I just find it amusing, in a lot of different situations, how status-quo most engineers are, how unwilling they are to speculate or brainstorm, how wedded they tend to be to how things are now, and how they are predisposed to slap down ideas, rather than play with them.

Where I work, we brainstorm constantly, toss good and silly ideas in all directions, argue them over hill and dale, and wind up with products we can sell for 10 times cost, because nobody has tried it before. It takes a peculiar kind of arrogance to do that, an unwillingness to be intimidated by "authorities" who slap down ideas on contact. SED is on average a pretty effective idea killer.

It's interesting that some of the best engineering has been done by non-engineers and some of the best programming has been done by non-programmers. Sort of like music.

John

Then don't let that happen. That part is easy and obvious.

Of course it's not mainstream. The question is whether "mainstream" will ever change. A lot of people are arguing that it never will. I suppose there were people who thought that console hi-fi and black-and-white TVs and portable manual typewriters were the ultimate in home automation.

John

How can making the whole system more efficient, reducing cache and main memory traffic, and eliminating context switching cause bottlenecks?

All I hear is criticism; nobody picks up on the fact that the chip manufacturers *are* building or planning 32 and 64 core processors, and that *might* really change the way OS's are designed.

What's really interesting here isn't the technology, it's the psychology.

Intel has an impressive record of a) investing in the status quo and b) wildly missing the mark on everything else. They sure have an engineering mentality!

John

Cores only need to be fed if they have something to do; if not, they can idle with zero power consumption [1] and zero bus traffic. For a given amount of computing, N cores do it just like one core, same nanojoules per megaflop, but with less shared cache and memory bus traffic, and no waiting and no context switching overhead. Why does everybody seem to assume that all the cores must be run full-blast?

John

[1} as Phil notes, silicon as we know it suffers from static leakage as geometries scale down, so even an idle nanometer-scale CPU has high static power consumption. A similar amount of compute power distributed among less agressively designed cores has a great power advantage. That's one reason that everybody is backing off on GHz and bulking up on cores. For Pete's sake, extrapolate that trend and *think* about where it might lead. It would take a new, probably non-silicon, process to change that trend. But you still have to make things out of atoms.

John

The Cell CPU is a good contender to be one of our next PC CPUs.

I already have it in the Sony PS3, and they disable the graphics too. That's why nobody cares it runs Linux. It is a real dog right now.

If we had access to the whole CPU and GPU, it would be a serious contender. Big mistake on Sony's part.

That's not true. IBM's fab has 100% yields. Sony is the one making the loose acceptance through less utilization model. Even Toshiba has a

100% yield Cell fab line.

No. Testing for 100% and only incorporating a fully functional CPU into your system is "too expensive"... for a consumer product... from Sony's POV.

They didn't have time to get a fully tested fab line into the delivery schedule for the product. So the quick tested, one SBE down acceptance scheme went in.

Hell, the friggin' PS3 has thermal problems at the CPU cite even with one less SBE.

Nobody has a 100% yield on a complicated chip of this sort. It might be 90 or even 95 or a bit more, but surely not 100.

M

On a sunny day (Fri, 16 May 2008 22:20:18 -0700) it happened Archimedes' Lever wrote in :

mmm ;-)

2006: IBM Claims 10% - 20% Yield Lucky for the Cell Processor. Cell Chips Continue to Have Yield Issues

See links above.

Not anything like the Microsoft console 'red ring of death'. Actually there are newer versions of PS3 with a new cooling system in the make:

In German, run through translater: Neues Kühlsystem für die Playstation 3

Those are more than 2 years old, dipshit.

IBM has blades that are DUAL Cell powered. It is a product, and you can bet that despite the article claiming that it is the "same" as what is in the PS3, it is not. It has ALL of it SBEs functional and in use. That is a tighter testing spec than the cell dice that are considered "good" for a PS3.

No, I will not be going to your German link.

Any idea of what the various processors would typically do on a blade?

John

On a sunny day (Sat, 17 May 2008 13:24:00 -0700) it happened Archimedes' Lever wrote in :

From your style I see that you are the nutcase in a different incarnation. Go away. You are too stupid to see the reasoning, do not know what 'yield' is, too stupid to learn, and generally just make an unpleasant noise. Do not bother to post here again.

What a slob, asking for information you would not use when you could probably use any search engine to find it.

So, why have a newsgroup at all?

John

Not "various". TWO IBM CELL CPUs, each of which consists of ONE POWER main core, and 9 SBE sub units.

Each blade has two CPUs, and two pairs of 2GB RAM banks.

They are in a "blade" form factor so that one could populate an entire rack with blades and create a small super computer cluster.

Jeez... the mentality of the so-full-of-self fucktards that populate these groups is astounding!

I don't have any of those myself--my cluster is made of dual-Opteron pizza boxes, which I got for a song as refurbs (don't tell anyone). A friend two labs down has a BladeCenter with a bunch of Cell blades in it, though. (He mainly works on liquid metal cooling at the moment--the new Apples are shipping with his liquid metal cooling technology.)

The programming tools look a lot like vanilla Power Linux, but with this magical vector capability. They get used for rendering, ray tracing, some HPC jobs, stream processing, video encoding, and suchlike--with a SIMD architecture you need vectorizable work. The compilers are smart enough to vectorize a lot of apparently iterative work too, I believe, so it isn't just a Power 6 with a bag hung on it.

When I was a kid right out of school, building satcom equipment, I was being really surprised by how demanding telecom performance specs were--I had this idea it was like AM radio, I suppose. It's the same with games machines--the technology drivers of the 2000s.

Cheers,

Phil Hobbs

As if you were any different.

Oooh, snappy comeback. I can see that you really have this usenet insult cycle thing down. That will be a useful skill when you grow up.

John

Just for fun, the Intel salesdruide mentioned some killer applications that would require his 80 core... well to look at a few that may come to mind: Voice recognition: Well, user independent voice recognition works fine on a 1GHz Duron with perlbox-Voice in Linux, perlbox voice is based on the Sphinx2 sofware from Carnegie Mellon University. I can just say: 'Show BBC', 'show ITV', 'show CNN', and it will pop up full screen on the monitor. So that saves 79 cores. Then there is perhaps object or face recognition. Imagine your computer purring aways for 1 minute and then saying: 'Good morning John Larkin, what are your orders for today?' I do not have this yet, but say we add a one extra core... or Cell SPE? Then of course you can build on that, one of the advantages of Linux, if the wrong person appears in view, by water hose, gunfire, what not. Today I watched Mythbusters launch a 200 liter water boiler through the roof, but then, oh yes, guidence and control... I once wrote a program that does option trading with the data from the last say 20 years and then finds the best strategy, Unfortunately it runs on slower single core too. So what sort of application would it need, on an 80 core (10 by 10 cm ;- ), chip? There are things like automatic steered cars that come to mind, but I am sure others have many more ideas... Let's have those :-)

What would really be interesting is multicore embedded processors, something like Arm or Coldfire chips with, say, four or eight cores, a couple of them with ethernet mac/phy and at least one with floating point. Thet would nicely partition hard realtime stuff into dedicated, reliable processors, safe from user interface and tcp/ip stuff.

Dang, a minute with google shows that lots of people are on to this already...

John