Future: Pocket 3D hardware

[Guest]

This is not quiet Pocket PC stuff, but it didn't seem to fit into a "anything" discussion forum with threads names like "Libary girl thread" and "Rice or not" ;) (I'm not having a go at those threads). I'm just going to share a few interesting technologies I know of for your amusement.

A few years ago I was intently waiting for the release of a 3D parrallel processing array chip by PixelFusion. Then they scrapped the project and announced they were taking the product towards networking solutions. They have renamed themselves Clearspeed and have just announced their first chip the CS301 Multi-Threaded Array Processor.

The press release is:

http://www.clearspeed.com/news.php?pr=17

Now you thought the Arm was low powered this does 10 GFLOPS per watt of power (25 in total, twice that of competitors and a fractiion of the power/cost). I think the Playstation 2 does 6, or was that 3.6. Even though it doesn't mention 3D applications, and I haven't read the data sheets yet, I think it is likely it would still be suitable for that purpose.

Now imagine a Pocket PC at even a twentifith of that performance at just over 100mw of extra power consumption. Even a smaller down clockable 1 Watt version would do wonders when combined with a Arm controller and a suitable 3D/graphic interface. Isn't Power VR in the same country and have there own parrallel techniques, surely if great minds can think alike they can get a license and think together.

I've known that this sort of power has been achievable for years. I was a member of the Minimal Instructions Set User Group mailing list for years. The head chip designer has a parrallel processor design that has power and power consumption that would make a xscale blush. The 25 Processors with local memory would achieve 60 Billion instructions persecond on an older, non state of the art, chip process. The cost would be 1 or 2 dollers per chip. The instructions though are to simple, and unfortunately it is crippled, for general use, by lack of general controller and restriction on processor unit local address space. He has withdrawn it from his web site and has left only a basic processor design up. So he might be working on it under contract and should not be disturbed. It is interesting that his first microprocessor came out shortly after the Arm and was clearly better, even though the Arm based Acorn computer was reveiwed as the fastest personal computer on Earth (a long fall to fastest Pocket PC on Earth). That the processor had a long instruction word that ran a group of instructions in parrallel, sound familiar, he then moved onto a better scheeme, shame on you Intel. I wish I had the money to hire him to do a volumetric rendering chip.

www.colorforth.com

Otherwise I have heard information on a company working on some near term optical computing stuff that promises Star Trek like processing power. This maybe related to the group that came through my local region some time ago displaying privately working free float 3D images (like that TV show again). I sort of made the deduction back then that it was using optical computing (though I don't have absolute confirmation). We will see how much is true and how much comes through eventually.

So at least we can dream for now.


Wayne

[sponge]

Interesting. Don't have the willpower to read through it all, as most of it is above my head. Any developers have any comments?

[wayne]

Yeah your right, it is simplified but still has technical jargon.

To sum it up it's way more powerfull than a Playstation 2 in floating piont (mathematics used in conventional 3D), it is also cheap and extremely good on the batteries. A theorectically reduced Pocket computer version would give a big boost in 3D to handheld computers for little price and little reduction in battery life. If you did this with PS2 parts at the moment you could expect little battery life.

The Misc design simply reinforces the piont that present designs could have been done a lot faster, lower power and cheaper (though the Misc design is too simple in present form).

The optical stuff is potentially Millions of times faster then present (I don't know what they are going to use for memory to keep up with that speed though) and is due out in the not too distant future (years). So you could theorectically run true Star Trek "Holodeck"/Matrix type 3D programs without solid light of course (which deosn't yet exist). My calculations (guess) are that just conventionally following Moores Law (speed doubles every 18 months) with conventional technology advancements, a basic (small) "holodeck" environment is possible by the end of the decade, and large ones over 10*10KMs by 2036-2050. The parallel processing designs needed though are going to be quiet different from what we presently have, which are the end product of the 3D API I want to eventually implement for my OS.


Wayne.
----------------
Any spelling mistakes are entirelly my own tiredness.

[Guest]

[sponge]

[Jinks]

[deciever]

[Jinks]