Sun Feb 3 19:56:44 2002
Valid topics are things like I/O, MPU's, Video PU's, Mem Types, Etc.
This is an attempt to interject some tagnutry into our beloved OSY which we so much love.
x86:
This things going to be around awhile. IA-64 won't find it's way into the desktop until at least 2006 IMO and right now it is slated only as an Enterprise Server type solution. It will probaly give Intel and Windows a decent foothold in the 64-bit market as players like Alpha and HP dry up.
It will also become the most popular platform for scientific computing utilizing both Win64 and *nix.
i don't see it making all that much headway versus Sun and IBM initially; rather, it will fill the void left by the absense of the PA-RISC and the Alpha.
x86-64:
Hammer will sell, but it's new extentions to the ISA will be virtually ignored. There is simply little need for a 64-bit address space in desktops. Intel is probably quitely working on a 64-bit x86. if so, it will most likely be released far behind hammer and could end up being a huge marketing advantage for AMD due to a pending misconception that 64-bits is faster than 32.
That could potentially push IA-64 down to consumer levels faster than everyone's speculation.
PPC:
Motorola is in trouble. Not only is it presently dying, but it may be held liable for Billions for its involvement in Iridium. It has competition in it's bread and butter MPU market in embedded devices. It sounds to be horribly mismanaged and it's process technology sounds to be mediocre.
What happens to the Motorola PPC is anyone's guess. It could just die with other manufacters picking off whatever tech they choose. It could survive with serious cutbacks in R&D.
I think the rumors of Apple buying the PPC is a bit too hoopefull. Apple has experience with developing some VLSI circuits and could take the G4 and improve it, yet I doubt they could keep it competitive with Intel/AMD. They may have problems finding foundry resources to produce chips and will probably pay a high royalty for such a privelidge.
Going with IBM is a much better economic solution, but they offer nothing at the consumer level better than Motorola. Plus Apple would loose it's Gigaflops and suffere a Huge PR barage by going back to the G3.
It doesn't look good.
Sun Feb 3 21:25:47 2002
duals or better becoming more common, on account of number 2
Hyper transport. Packet switching on the mobo makes me hot.
No more old-school BIOS. Open Firmware or something similar, fully 32-bit, and a new boot process that makes some damn sense.
Serial ATA, or the return of commonplace (and more phool-proof) SCSI
The end of the old ways. No more serials, no more parellels. Just USB2 or FireWire for connecting everything. And, if you have enough channels, there's no bandwidth sharing.
x86-64, so that if you need it, you've got it.
DDR or MRAM (MRAM would rock, but as it's at that "5 years to production" type status, it's a pipe-dream.)
Organic LCDs that can handle refresh rates of standard CRTs, no less than 19 for $400.
I'd like the "next PCI" to just be an external extention of Hypertransport, and add-on cards to interface with the PC the same way as Cisco Router add-in cards. (they just slide in the back and have a couple thumbscrews. The case doesn't come off unless you want to mess with the memory) Of course, this could mean the return of the NLX form factor, and who really wants to see that ;)
I'm ready for some hyperthreading. just 2-way would be fine.
I want everything video to use something digital. cameras and whatnots have FW, Monitors should all use something like DVI. Not the huge connector like on my Radeon VE, but something smaller like a smaller version of the 80 pin SCSI connectors. (only 20 or so pins, 80's nuts.) Since video is inherntly serial, there's no need for so many pins. Plus, how many damned signals can you need?
ISA is done. Let it be. (you think "oh, I don't have any ISA ports, whee!" You are wrong sir. The integrated floppy controller, the math coprocessor, and DMA are all on the ISA bus, running a blazing 8 or 16 bits. Huzzah. This requires some real changes, so like many of my other requests, don't hold your breath.
PPCs: PPCs as a desktop processor will die with Apple. Anything else that happens is inconsequential.
Sun Feb 3 21:33:27 2002
What is "Packet Switching"?
Sun Feb 3 22:43:06 2002
No more old-school BIOS. Open Firmware or something similar, fully 32-bit, and a new boot process that makes some damn sense.
The whole boot process should be at 75hz so I don't have a disco flashbacks.
SMT and Dual processing is going to be niche until multithreaded proggies become the norm. I don't see this happening until compiler tech is such that it can derive multiple threads at compile time.
Don't expect that anytime soon. It will probably require a new language paradigm (or a revisiting of an old paradigm like Fortran?) to implement this.
Sun Feb 3 22:46:21 2002
how much longer till we see an ORganic LED display hit the market??
Mon Feb 4 00:37:15 2002
If it's gonna be truly IA32 backwards compatible, it's still gonna boot in fucking real mode, it's still gonna have a poxy instruction set. AFAIK, it'll add another 8 general purpose registers. So I'm presuming it still won't have nice three operand instructions, and will still be generally crap.
IA32 needs to die.
SMT and Dual processing is going to be niche until multithreaded proggies become the norm. I don't see this happening until compiler tech is such that it can derive multiple threads at compile time.
Threads are high-level things; automagically creating new threads won't happen. It requires too much knowledge of how the program will work. But better extraction of instruction-level parallelism will hopefully be realized, particularly for IA64.
The big problem is, lots of algorithms can't readily be done in parallel, so a single processor at 2X MHz will almost always be faster than two at X MHz, assuming, at least, PC-style architecture (where adding more processors doesn't do much in the way of adding memory throughput, so if it's memory bottlenecked at 2X it'll be memory bottlenecked at 2 * 1X).
Mon Feb 4 01:50:17 2002
There are problems with the serial nature of a lot fo code. And not withstanding that, Compiler tech is not likely to come around any time soon to where parallelism can be extracted from suitable places to any extremely significant extent.
It kind of defies the current paradigm I think. Compilers can be relatively crude as hardware reordering is fairly promenint in the x86.
Not to say that compilers aren't or haven't made headway recently, it's just not as important with such things as OOE.
So I'm presuming it still won't have nice three operand instructions, and will still be generally crap.
Probably will be the same old accumulator style stuff. I mean, we wouldn't want to give anyone reason whatsoever to actually use them.
Judging by a quick glance at task manager, they already are
Yes, but that is just OS stuff for the most part isn't it?
Mon Feb 4 04:23:57 2002
Yes, but that is just OS stuff for the most part isn't it?
The rest are not multithreaded:
*Uptime2K
*Office text services thing
*Intel tray icon thing
*4NT
*srvany
*QuickRes
*srvany
Idle process
Of the multithreaded tasks, exactly half of them are not part of Windows.
Mon Feb 4 07:21:45 2002
I like that because it could greatly simplify drivers because they wouldn't have to deal with the different kinds of hubs like Via chipsets and all that crap.
SMT on consumer level procs would be nice too.
Mon Feb 4 14:20:45 2002
What is "Packet Switching"?
HT also has the lovely quality of being a lot like ADSL. It can have differeing upstream and downstream speeds. So, for a fat burst read, it can use 12 of the 16 data lines for upstream and 4 for downstream (control and whatnot) And when things cool back down it can revert to its "standard" of 8/8 or whatever it's programmed to. Really cool.
how much longer till we see an ORganic LED display hit the market??
DrPizza has some good points, but what can IA-32 be replaced with? I suppose they could start with the functional units from the PIIIs and 4's and just strip out the real mode stuff and more or less end up with a new processor that's already fast but I don't imagine that will work. I think Intel is probly stuck with it for a while sadly. (that real-mode booting thing sucks donkey nuts. )
Mon Feb 4 14:55:28 2002
Of the multithreaded tasks, exactly half of them are not part of Windows.
interesting, so maybe SMT and dual processors would have a better impact than I thought?
Mon Feb 4 15:38:27 2002
from AllYorBaseRBelong2Us posted at 2:55 pm on Feb. 4, 2002Of the multithreaded tasks, exactly half of them are not part of Windows.interesting, so maybe SMT and dual processors would have a better impact than I thought?
Mon Feb 4 16:19:56 2002