boglo wrote:
The new CELL cpu maybe your answer. It will run multiple os's at the same time.
This is a... weird extrapolation, at best. There's nothing about the hardware itself that supports it any more or less than a more conventional dual-core ("SMP on a chip") design, however my expectation is that the stream processors will probably run some sort of specialized "OS" (that won't look much like an OS, more like a bare copy of Elate at best, and now you know one of my stranger theories that probably won't pan out) to coordinate them and make life easier for developers. For all I know, the "ring topology" of the bus may mean that even that is unneeded overkill.
I know it will run linux but haven't heard if it wiil do windows/mac. This is a chip jointly develpoed by Ibm, Sony, and toshiba. First product out of the box with it will be PS3.
You'd kind of hope for Linux. IBM/Sony/Toshiba is right. PS3 is right (and a good question is how much of an OS, and what kind, the PS3 will ship with in-box or in-ROM at all).
Sony and Apple have been rubbing shoulders very visibly recently, but latest news suggests this may be for backing Blu-Ray in the new VHS/Beta wars (Blu-Ray on PS3 and Macintosh, HD-DVD possibly likely to overtake the cheap Chinese video player market, but I've lost track of what Microsoft and Intel and the big x86 vendors favor where, so it's not as clearcut a war; everyone seems to think Blu-Ray might be the 'nicer' medium for random-access data, so it may be that Blu-Ray becomes the next floppy replacement and something DVD-branded stays popular for canned video)...
If it can do windows then everything will change. Like ps3 games ported to desktops. Linux then has a strong software base to be real competition to windows. Ought to be fun to watch anyway.
This part appears to make no sense, to be honest.
Microsoft has IBM cooking something for XBox2, but last I checked, the take of the Britsites seemed accurate, and this revolves around using "PowerPC technology" to implement a 'proprietary-enough' Microsoft instruction set, as part of a "Well, why not, we have billions of dollars and no room left to grow" scheme to wrest control of the market from AMD and Intel. In MS fashion, this revolves around vector units or something that
could be broadly licensable -- so now your MS tax would come in the CPUs themselves -- but it remains to be seen if it will catch on. And it still doesn't automatically imply 'Windows-software-will-run-on-it,' especially without 'paying twice' (first in the price of the CPU for the units, then for your Windows license, assuming a port of something called Windows exists).
The XBox2 will certainly run a port of something Windowsly, but you'd still need the exact same CPU arch (or everything targeting the .NET virtual machine) to run the stuff elsewhere.
BTW it's not the motherboard that is important it's the cpu. The motherboard is just the supporting stuff you need for the cpu. So you should ask what is the mother of all CPU's? :-? :-?
*bong noises* :-P Nah, really, the reason we have 'motherboards' is that system tech has become a lot more "Amiga-like," with the same limitations -- getting a crapload of signals back and forth from CPU to memory and GPU and disk controller and gigabit network fast enough requires a lot of wires, and a lot of care in design. PCI-E solves a lot of this, as does Intel's new "FB-DIMM" spec for memory, but guess what -- both of those will either end up having their clocks bumped fairly rapidly, or more parallelism induced (think of what happened to SCSI -- Fast, Fast-Wide, Ultra, Ultra-Wide...).
Because many parts end up being "matched" to a particular level of CPU, there's not a lot of money in solving this problem; it's "too cheap" to just make a motherboard, and, when Moore's curve has been followed enough, drag that level of integration down into a "System-on-Chip" (which you'll surely see more of now that we can cram more transistors in at Moore's rate, but can't make individual ones switch faster so easily; integrating more 'motherboard' functions can make parallelism more convenient, as you see with AMD64 NUMA and such).
But otherwise, let's face it, this "motherboardless computer" already exists, in the form of Mini-ITX boards and smaller. Yes, those
are motherboards, but did you look at what some of those old 'swap the 286 for a 486' machines put on the card, even then? Come a bus like PCI-E, which can be fanned after the fact (or USB, or Firewire, already existing), and you've got the
equivalent of a backplane system, even if it happens to be strewn all over your desk. This is, in fact,
better than a backplane for a home user, because it doesn't all have to live in a rack, and you can put, y'know, the printer or external drive somewhere where you can reach the paper or the blinkenlights.
In supercomputer and blade applications... Yeah, the current level of integration can seem a bit wasteful, especially when you look at these clusters built with COTS parts and one power supply and disk per CPU. (How much fossil fuel are we burning to crunch SETI?) However, unless you've got a better idea than I do, the solution is probably miniaturization and "better use of Moore," rather than trying to devise purely-custom 50-CPU boards (for today's CPUs) that will be obsolete by the time you've finished routing the traces.
