Re: If you could design a $300 homecomputer today, what would you choose?

[David Ogg <davidogg@nc.rr.com>]

In article <tln0ubc06v1g89@corp.supernews.com>, tthai@mindspring.com 
says...
> 
> "David Ogg" <davidogg@nc.rr.com> wrote in message MPG.13e2c36ea28bbffc989686@news-server.nc.rr.com">news:MPG.13e2c36ea28bbffc989686@news-server.nc.rr.com...
> > In article <tlcvenh9gb1d9a@corp.supernews.com>, tthai@mindspring.com
> > says...
> > >
> > > "David Ogg" <davidogg@nc.rr.com> wrote in message MPG.13dd9c209377d2c989684@news-server.nc.rr.com">news:MPG.13dd9c209377d2c989684@news-server.nc.rr.com...
> > > > In article <PXD47.853$bl.47003@news.siol.net>, brankob@avtomatika.com
> > > > says...
> > > > >
> > > > > >
> > > > > > Besides, isn't the 6502 limited to 64kB chunks.  Apple simply
> > > > > > implemented some sort of paging mechanism to access the 128kB found
> > > > > > in later models (IIe/IIc).  Likewise for the 192kB card which I
> > > > > > have lying around here, somewhere.  (Though my understanding is
> > > > > > the 1MB card for the 8-bit machines used some sort of window into
> > > > > > memory.)
> > > > > >
> > > > > > Byron.
> > > > >
> > > > >
> > > > > Why not using 80386, then ? It has it all- MMU and segmented adressing, and
> > > > > one
> > > > > even gets a several 100's bytes of cache ? ;o)
> > > >
> > > > How about Transmeta's Crusoe? This would outsell everything mentioned
> > > > thus far I guarantee it. It can emulate anything, including x86 at
> > > > native speed!
> > >
> > > I dont think thats true. The transmeta chip only deliver a fraction of the performance
> > > of a core like the thunderbird at the same mhz.
> > > And you can go in a shop and buy a single chip for under 90$ (the 1.2 gigahertz part).
> > > But then you can get a 750mhz similare core (Duron) for 25$... thats your price.
> >
> > A notebook CPU costs more than that, heck I can't even find a price for
> > comparison. You just couldn't use one of those furnaces in a book-size
> > unit.
> 
> Actually the 750mhz part should run 'cool', also you can run it at 400mhz and remove on chip active cooling?.
> I put 533mhz celeron in booksize computer and they where fine. a 750mhz duron would be cake.
> 
> >
> > > I dont see the point in using a much slower chip like the transmeta crusoe for a desktop/booksize PC.
> >
> > But it would be a HELL of a lot faster than the 68000/Timex
> > Sinclair/6502 relics others are suggesting, and it runs without a
> > heatsink. You ever felt the CPU in a notebook? imagine the temp a
> > thunderbird would be!
> 
> Yes, I would choose a rusoe over a 68000, no problem here.
> For low power... I kind of like the strongarm.
> I dont really have a sweet spot for transmeta effort, I dont see where it really fit in the market.
> I also see that they are not doing well at all... Could be riding the economic wave, or just problem
> grabing some of the cpu market.

The problem is it's a niche market, albeit an expanding one. They don't 
even plan to break even for 5 years. I don't ever see them taking over 
any desktop market, however, for low-power applications with platform
legacy requirements, you can't beat em.

> > > just to give you an idea, a crusoe TM5600 600MHz running at max power usage deliver
> > > 46 sysmark2000 , a 500mhz P3 laptop deliver 115 sysmark2000.
> >
> > Thats under the x86 codemorphing layer with the optimizations set to
> > "don't cache code", The Crusoe is a tricky beast to benchmark,
> > especially against it's own codemorphing layer. The codemorphing tends
> > to cache the benchmark loops giving unrealistically high scores, so for
> > simplicity, it's turned off.
> 
> In the end , MMX/SSE/multimedia/3d stuff will probably run many fold faster on a slower mhz p3
> then a crusoe chip.

Oh, yea, no question about it, but then, this isn't a desktop machine 
we're talking about. x86 on the desktop is king.

> At this point I would guess that a mobile p3 can render more then a crusoe using the same amount of
> power... Using less power ut performing less is not always an advantage.

Not on a desktop or server machine, but then thats not where any sane 
person would dream of sticking one of these anyway. The goal is a 
slightly higher platform abstraction, and slightly lower power 
consumption.

> > > > Can you imagine UAE ported to Crusoe as its code-morphing layer?
> > > > incredible.
> > >
> > > No so sure about that...
> >
> > It would be wild, a codemorphing layer that abstracts a 68040, and the
> > chipset! come to think of it, I wonder if UAE would even be needed?
> 
> Their is a diference betwen reformating code (UAE is already doing its own type 'codemorphing')
> and doing cycle precisise timing on virtual register set of custom HW on top of that.
> Their is alot more stuff in UAE then what transmeta is doing, thats why I never caught on the hype.

( %me% knows nothing about the internals of UAE so I'm not going there )

> > > > Modules could be made for just about ANY old school CPU/platform, and an
> > > > OS wouldn't need to be developed for it, as it could run these platforms
> > > > OS's natively, maybe even in addition to a custom OS.
> > >
> > > Their is already so many core running on the x86... From the most obscure to
> > > C64 & amigas.
> >
> > But those are "emulators", Crusoe's codemorphing is a much different
> > beast, even x86 is just "codemorphed". c64/Amiga's are running on top of
> > Windows/linux on top of x86. The Crusoe wouldn't need a host OS.
> 
> Does codemorphing support handling of strobe registers? I dont think transmeta
> support emulation of something as complex as the OCS HW register set.
> The floppy controler register would need 'HW logic' emulated... etc... Not an easy task.
> So you would need to show big revenue to spend all that R&D to emulate decade old HW.

Not any more than it took them to write the codemorphing abstraction for 
x86, from what I hear, the x86 architecture was a real bitch with all 
the registers they had to emulate, I think the reason they emulate a 486 
is because those registers were documented. That was the reason NASA 
gave for using the 486 as well, access to documentation. Transmeta then 
had to provide a BIOS as well, as part of the morph abstraction.

A big part of the R&D has already been worked out by others already, I'm 
sure AROS/UAE technology could be used. I know UAE is free under the GNU 
GPL license, not sure about AROS.
 
> > > > Computer do alot nowdays in term of multimedia because of the power they have.
> > > > I'm not so sure intel/amd etc.. just develope for people to run half life at 500FPS.
> >
> > On the contrary, they'll market them for dumber things, like Intel
> > saying it's "New Pentium 4 processor will enhance the web" and going
> > on about how the internet will be faster. I always get a kick out of
> > that :)
> 
> It run flash smoother, it handle better codec like mpeg4 and real, complex web page scrool
> smother, it will give higher frame rate in complex java gfx applets, etc, etc...
> So in effect, yes the web to an end user look smoother/faster... He can navigate a bit faster,
> scroll complex page with ease.
> Use a DSL conection on a pentium clasic and then a pentium4. If you visite media rich web
> page you will see a Dramatic diference.

Sure, but what they said was that the SIMD instructions new to the P3 
would speed up DSP code, making software modems faster. I think alot of
people somehow thought that they could upgrade their dial-up connection 
by upgrading their processor. ;) It's true the SIMD instructions do 
include DSP specific instructions, but I doubt anyone would notice on a 
dial-up software modem anyway, and I don't think the network card 
drivers will benefit from SIMD at all, since there's no DSP taking place 
anyway.

> So I can see people getting a new computer when they get braodband.
> 
> >
> > > > Linux Geeks would love it! (if only because Linus works for transmeta).
> > > > Slashdot would pimp the hell out of it, you wouldn't even need to market
> > > > it :)
> > >
> > > > Hell, I'd buy one and just run it as my mail/webserver and
> > > > router/firewall. transmeta already has a custom linux distribution
> > > > available for it. Not only that, but it could be offered in several
> > > > packages, you want WinXP on it? No problem!
> >
> > > Why would you want to pay for low performance?
> > > Run x x86 version of the major OS... run Ti calculator or c64 class stuff under emu
> >
> > Because emulation would /be/ low performance, the Crusoe running a
> > platform under codemorphing is faster than any emulator I've ever seen.
> > Show me an x86 emulator that can run Windows faster than the Crusoe can
> > under codemorphing/x86. Remember, this is a hobby box, if I want to toy
> > with an OS, I want to run it natively, (or a codemorphing facimile
> > thereof.)
> 
> What OS worth anything doesn't run native on a x86? :)
> And how much power do you need to get z80, 6502, 68000, etc.. emulation going on
> a gigahertz clas thunderbird?

;) If it's meant to run Windows/Linux/Solaris ect. and sit on my desk, 
give me my Athlon 1400/266 768MB DDR MSI-K7T-Master w/raid box anyday, 
but this is a smaller, possibly battery-only portable hobby thing here
meant to be multi purpose webpad/PDA/portable what have you, it would 
have to be very flexible in what you could do with it and where you 
could bring it, so as not to become so "niche" that noone buys it.

 The reason I see Crusoe as a good match for a project like this is that 
it sort of combines a dozen or so pre-existing niche markets into one 
market that could be sold the same hardware. There's already a market 
for a low-power webpad, and people are willing to pay $2300 for them 
(check out Slashdot's submissions for saturday) there's already a market
for PDA's, and Crusoe is faster than anything in most PDA's right now. 
portable units for inventory and UPS/shipping are already being sold, 
for anywhere between $400-$1000.

 Hell, Transmeta's Crusoe, running at 633MHz, is already being used in 
very-low power servers, running on 7 watts (idle) to 15 watts (max load)
http://www.transmeta.com/everywhere/products/servers/rlxtechnologies/rlxtechnologies_system324.html

(this link may wrap, my newsreader sucks)

> > Plus it's upgradeable (the codemorphing), meaning the platforms it can
> > morph will increase over time.
> 
> I dont think they have yet to emulate a pentium MMX, they just emulate the basic 486?

I think your right.

> > > > The market could even be expanded to commercial vendors, it would be a
> > > > viable replacement for say, the hardware TiVo uses now. Cable
> > > > descramblers, Sattelite receivers, hell, I can think of very little
> > > > you couldn't make this thing do.
> >
> > > You would need native code to do software encoding on that chipat a decent speed.
> > > Who would write all that code?
> >
> > In the case of the Tivo, it's already been written. the Linux/x86
> > package is already available customised for the Crusoe, tailored by
> > Linus Torvalds himself. In the case of the Satelite reciever, you can
> > find the GPL'ed code from a link off of www.satmatrix.com. Kiosk
> > software has already been made available for the Crusoe under GPL, so
> > has PDA codemorphing for the Palm platform.
> 
> What codec did Linus code nativly for the crusoe?!!?!?
> Do you have a link on what is in a Tivo box?

http://www.tivo.com/linux/index.html

They released their power-PC linux mods as per the GPL.
Since the mods were developed by TiVo themselves, they already have the 
codec source, all they would need to do is recompile for Linus's 
"Transmeta distrobution" (maybe not even).

If Transmeta releases a PowerPC codemorphing layer (or someone makes 
one) they wont even need to do that. This would probably interest TiVo
greatly, as they already stated they wish to make their money from 
selling the subscription, not the hardware.

> > The beauty is, if code isn't available, as long as you can produce the
> > codemorphing layer, you just reuse your existing code! Case in point,
> 
> Thats the whole problem... The chip is just to slow.

yes for a desktop, but for the applications and the specs of this 
project, it far exceeds what the alternatives can do, and will still be 
in the "desktop speed ballpark", unlike a z80/68000 ;)
 
> > Windows. Windows doesn't run on the Crusoe, because the Crusoe isn't
> > x86, it simply pretends to be through abstraction via the codemorphing.
> > Hence Windows can run "natively" on a CPU it isn't written for.
> 
> Like UAE convert 680x0 to x86. I can take a fresh copy of Amiga OS from a box and
> install it on a PC.

You mean you can install windows, install UAE, install AmigaOS, if you 
were buying something specifically to do things like run AmigaOS, why 
would you fudge around with windows? Heck, the best way to have a 
desktop Amiga is to get an Amiga desktop anyway ;)

> The thing is windows will expect some kind of common PC chipset/bios support if
> its to ru nout of the box... I dont see the point. Get a mobile P3 to run windows, it
> will be faster. Support native MMX/SSE at full speed.
> 
> >
> > > > R&D required? very little, the hardware is ALL standard, and already
> > > > made, produced, and sold, not to mention it already has HUGE "geek
> > > > mindshare."
> >
> > > Cant that be said for the x86 platform?
> >
> > Only for the case of x86 hardware meant to run x86 software, the Crusoe
> > isn't limited to x86.
> 
> Aside hacking a new processor, or fiddle around... why would anyone actually
> morph to something else then a x86?
> For example, can the sony laptop also run MaxOSX with a PPC version of the crusoe code morphing?

The Sony laptop was made as an x86 laptop, this project isn't about 
making a Windows98 laptop, they exist already.

> > As for the "Geek Mindshare" I think a Crusoe captures more mindshare
> > among the HW geeks than x86 does.
> 
> I'm a programming geek.. and I got loads of tools for the x86. I also love processing power.
> The crusoe is a just a processor that got its core designed to be emulator friendly... But aside
> emulation is doen't have much to it.
> 
> Again, I'm not a fan of that chip because I see no use for it in all common computing form.
> PDA, laptop, bookside, desktop, workstation, server... All have better processor already there for
> the picking.

The project needs something that can do them all, at whim. No general 
purpose device will ever outperform a specifically desined for a purpose
device, thats why we use 1400MHz Athlons to play games and 33MHz 
Dragonballs to sync our palms. but neither one of those would be very 
good at both ;)

 You can play Quake on a Crusoe, and still remain at only 2 watts. The 
same beast will run linux, and can scale down to less and be walked 
around like a PDA.

 It's not something that would replace the desktop, but an extension of 
it, replacing possibly our webpads, mp3/Rio's, PDA's, pagers, inventory
interface units, ect with a common device/platform. It might even 
slightly eat into the laptop market as alot of people only "need" a 
laptop for trivial things anyway.

> 
> Stephan
> 

--David
But then again maybe not.
Boycott Adobe! www.eff.com