Supercomputer on a chip

[Taninim the Destroyer]

Those of you who are technology junkies will probably be happy to know about this cool new development. Sony has developed what they call the "cell chip" for the next generation of playstation (initially, of course). The chip boasts a parallel architecture of eight separate processors and high clock speeds making it essentially a cluster on a chip. I haven't seen many details yet, though I speculate that on chip cache memory may be an issue.

The revolutionary aspect is not the hardware configuration, but is actually the fact that the chip was designed from the ground up to work in a very efficient parallel architecture with cell software which locates and uses available resources.

Anyway, here is an article chosen at random off the net. See what you think:
http://www.theregister.com/2005/02/01/cell...lysis_part_one/ (http://www.theregister.com/2005/02/01/cell_analysis_part_one/)

[GreNME]

I think it's funny. The article begins with a warning that there is much speculation that needs to be taken with a grain of salt, and then goes right into rampant speculation.

However, once the following line came up in my cursory search:This time, Cell is aimed at a different market, one that Wintel has failed to conquer - the living room. ®
Thanks, but I'll wait until a more reputable, less biased (as in less zealous against "TEH EVAL M$") than the Register to come forth with information. They have been known to be quite unreliable when it comes to information dispersal (like the "3 Ghz Windows Malfunction" fiasco).

The dual cores (http://arstechnica.com/news.ars/post/20050207-4591.html) coming out later this year are going to be a bigger deal.

[Primal Curve]

I'm just waiting for some other company to come out of left field with a chip who's core processor exists on a seperate plane with different physics, allowing it to process data at speeds unimaginable in our universe.

Hmmm... sounds like an episode of The Outer Limits to me.

See, our hero invents the chip, but hasn't fully tested it. Someone in his lab leaks the information to the press, and the plans are stolen by a major chip manufacturer. After sucessfully completing several prototypes, the company begins a full production run. After several thousand are completed, however, it is discovered that, while one of these chips is okay, several thousand create huge rifts in the local fabric of space/time. Weird alien creatures come from the neighboring universes with varying goals- none of which relate to the goodness of mankind.

All hell breaks loose.

And the Ninjas retaliate because, while known of for many centuries, no one really knew the true purpose of ninjas- multi-dimensional proctectors of mankind. They've got the know-how and the grit to take on such bizarre and dangerous bad guys. Go Ninja Go!

Okay, I'm done with my weirness now.

<jumps onto another plane> :ph34r:

[Taninim the Destroyer]

Grenme-

there is actually a good bit out there on the net about this chip- all of it is speculation based on not much info, but it certainly sounds like a change is coming. Of course, the dual core intel chips are likely a reflection of this change. It makes sense that the next generation chips will essentially be clusters all running on a single machine. Of course current applications don't really need all of that raw computing power, but if it gets built then they will come.

[GreNME]

Well, the thing is I've been keeping up with the multiple core thing for a while, since the hub-bub really picked up last September. What I've noticed is that every chip maker has been doing their best to try to raise eyebrows and give just enough info to the public about plans outside of their typical "roadmap" to spur lots of speculation. They do this to spur investments, even though the risk of doing it is that they could hinder their current sales because of higher hopes for future product. IBM, Intel, and AMD have all been guilty of this. IBM has had the worst track record of the three so far (they promised a 3 Ghz PPC chip last year and never scaled above 2.5), and that they're working with Sony (who is notorious for making promises they can't keep) on this project gives me less enthusiasm than the speculation seems to permit.

But on the the technical aspects, since picking on IBM and Sony (not so much Toshiba) is like barrel-hunting.

The guessed-at technical aspects of the "cell" chip seem to promise a whole hell of a lot more than seems likely possible, especially giving people the impression of clustering on a single chip. There are already dual and quad systems out there, and if you go higher-end with servers you can get up to 32 or 64 processors working in tandem. Let's stick with dual or quad for the moment, though: just because these systems have twice or four times the CPU in the box does not mean the machine gives twice or four times the performance, even with software that is specifically designed to take advantage of more than one processor. Despite the hype, performance is often far from twice or four times as fast, closer to 40% at best on most software, less for others not designed for multi-threading. As far as I know from the details, dual core chips don't quite match the gains of two single core chips, but that's mostly speculation and I'll wait and see on my own, though the arguments for it make sense.

For more rampant speculation, you can check this (http://www.blachford.info/computer/Cells/Cell0.html) out, but I'm going to side with a more trusted source (to me) and agree with this little write-up (http://arstechnica.com/news.ars/post/20050124-4551.html) instead.
I should note that the author has published some "Clarifications" on his website, and he does back off some of the wackier claims. For instance, in response the criticisms of his claims about magical code parallelization, he says, "This is not true. You still have to break up problems into software Cells." Um, yeah. Precisely.

At any rate, if you have some intermediate level of computer science knowledge and you read the article with a critical eye, throwing out things that are obviously bogus and/or overblown, then you can actually pick up some information on the architecture. Mind you, there are no new revelations in the article (except for the stuff that's made up (e.g. SETI) and/or wrong (e.g. "check it out! no cache!")), but Blachford did manage to pull together a lot of what's already known into one place.
Grain of salt until otherwise shown proof.

[Taninim the Destroyer]

Grain of salt until otherwise shown proof

I quite agree. You raised some good points, and for this chip to be truly a quantum leap forward would require that essentially all of the supporting architecture (eg memory) would have to be optimized to work with this sort of scheme. Also, as you also mentioned, in order to get much bang out of something like this, the software would have to be written with this sort of processing in mind.

[Slash the Berzerker]

Yes, but will it make Half Life 3 look really cool?

[Taninim the Destroyer]

Slash- that is what uber gaming geeks are banking on- that all of that really expensive new hardware will somehow translate into better gameplay.

But in reality I think that unless the game programmers try to optimize for a given architecture then lots of that really cool hardware is just really fast spinning wheels...

[GreNME]

Not only that, but even what some are saying about the architecture just seems like theoretical-yet-ultimately-unrealistic stuff. For instance, the claim of all these FLOPs—or floating-point operations—is not directly translateable to overall system performance, unless we're talking strictly number-crunchers or calculators. Apple tried to play that game ("More FLOPS!") and when looked at under realistic load the theoretical superiority proved to be true only in theory.

The whole x86 architecture is a weird thing: it is nowhere near the most effective or technically superior architecture out there, but it still seems to dominate the market and still remain scaleable beyond what I think expectations would have been twenty years ago. In this respect I think it may be similar to the internal combustion engine (though I wouldn't say we know "superior" architectures to it, exactly).

My prediction: this is going to turn out revolutionizing appliance computing and create a much larger market of scaleable embedded computers for the home and other uses. Personal computing is going to require more than a processor architecture to bump the market into a new way of doing things.

[raventh1]

Here is some good information:
http://www.blachford.info/computer/Cells/Cell0.html
http://www.electronicsweekly.com/articles/...ArticleID=38754 (http://www.electronicsweekly.com/articles/article.asp?liArticleID=38754)
http://arstechnica.com/articles/paedia/cpu/cell-1.ars

I can't wait to see what happens.