@KimmoK

Quote:

KimmoK wrote: " is a PC with just the CPU (a RISC!) replaced..." RISC means more work in assembly level,

Exactly. That's also why the RISC code density is poor, whereas CISCs do much better (and the 68K was and is one of the best).

Quote:

but native AOS4 can execute 68k code as well (almost as native).

Every CPU can emulate the 68K, even little-endian ones (with a little extra conversion work, when needed).

AOS4 "promoted" 68K applications to work as first class citizens, like PowerPC ones, but it's a "side effect" of the 68K layer implementation (e.g.: works only because of using a 32-bit, big-endian, architecture).

But it's nothing special. The same thing can be done with other architectures, even little-endian ones.

Quote:

Times when Amiga SW+HW combination could do a lot of unique things is over for now.

You can find a good approximation with the consoles.

Quote:

(and we could do a lot same+new stuff in SW now, with 1Ghz+CPUs, but we lack manpower) My most enjoyable CBM Amiga was/is Amiga4000, already it had parts like chunky pixel GPU that had nothing to do with "AmigaHW", etc...

But lacking manpower makes it difficult to address the full potential of a GPU, even a very old one with only the shader model 1.0.

Quote:

I learned that the OS is more important than A4k HW.

OK, it's true because the Amiga 4000 hardware is very obsolete, but the same applies to the o.s. too: it's very outdated, and unable to make a good use of the new hardware.

Quote:

So, for me, "Amiga" is where the current OS is. And AmigaLike can also be as nice to use (or nicer) than the real thing.

It's clear that anyone here has a personal idea of what the "Amiga" word represents now, but once people communicate they should agree on the terms they use and on their meaning.

While I think that there's a general consensus on the "Amiga" term related to the Commodore history, it's highly questionable what it means after this period.

That's why I prefer to use words like post-Amiga or AmigaOne: because they express exactly the meaning of what we are talking about, without falling in the personal side.

For the same reason, I certainly don't use the "Classic" label, because there was no "Classic" Amiga, but just and simply the Amiga(s).

Quote:

btw. I would not mind being able to do some special things on NG AmigaHW ... if it had things like DTV OUT (broadcast quality), it would enable some cool stuff...

I don't understand. Are you referring to something like the genlock?

Quote:

NutsAboutAmiga wrote: @cdimauro Quote: As coder, it was a pleasure to program thanks to it's 68K (CISC rulez!), custom chipset with ad hoc coprocessors, and the possibility to directly hit the hardware. In all other operating system you put hardware specific stuff in a driver, so you can replace the driver, when you get better hardware. if you want to hit hardware on a modern computer, you will need to scan the PCI bus, find the product and manufacture id, then ask for access to the hardware, then you get a memory IO, and port IO address of PCI bus, now that you have this you can poke the hardware.

Which is something that you also needed for the Amiga, unless you are using only 512KB of chip ram and the OCS, where you already know all the you want, and can use it immediately.

Beside that, if you wanted to use more than the 512KB and/or something different than the OCS, you also needed to make some tests and retrieve the required additional information to accomplish the task.

Quote:

As for writing 68k assembler I agree it's easy because of naming of the instructions, they are easier to memorize then for example powerpc assembler, where you have to memorize more names, 680x0 assembler uses its syntax to differentiate between operation code, powerpc does not, the name of instruction is the only thing that decides what operation code to be used (1 to 1). a "Word" is also not the same for powerpc as 680x0, a word is 32bit on powerpc, on 680x0 its 16bit, a half word is 16bit on powerpc. Almost all of PowerPC instructions have 3 registers as parameters where 1 of registers are optional (or the 2en parameter). An example: 680x0: move.b PowerPC: lbz and stz 680x0: move.w PowerPC: lhz and sth 680x0: move.l PowerPC: lwz and stw I think once you memorize how the names are logically organized it can be as easy to program as 680x0 assembler.

It's not only a question of mnemonics. Yes, they are important, but a CISC let you use complex memory addressing modes for many instructions, whereas with the RISC you have only the load/store instructions.
The same limit applies to the immediates, where to use them you need more instructions on a RISC to load a generic (and not "short") constant.
And we can put on the table even the complex instructions which a CISC generally makes available.

CISCs and RISCs are very different, and if I've to write low-level code, I generally prefer to do it with one of the latter.

Strange People we are, indeed.
This topic is more present now than there was a (real?) possibility to have an amiga laptop.
@cdimauro:
Hi, pal. Remember tha Cisc and Risc are two squirrels verrry friendly to Donald Duck.

Last edited by rebraist on 07-Nov-2013 at 10:56 PM.

@cdimauro

Quote:

For the same reason, I certainly don't use the "Classic" label, because there was no "Classic" Amiga, but just and simply the Amiga(s).

I understand what your saying but if you do not want to talk about one model, but any computer from Commodore, ESCOM and VisCorp, then I think it makes sense, Commodore was not the only company making Amiga computers.

Computers whit classic chipset, is easier to say then computers whit AGA or ECS, PAULA, DENICE, ALICE and so on.

Quote:

Which is something that you also needed for the Amiga.

Well is that true?

Most things writen in assembler for classic Amiga, pokes the Joystick ports direct.
They never check if there is gameport.resource you can open or anything like that.

The same goes whit Paula, you do not check if its there, you know its there so you start setup hardware registers whit out knowing if there or not.

And also, if you have program look for cia.resource, no program check what address hardware register are at, thats hard coded in the code.

AHI was introduced in 1992, but most games do not use it.
VHI is some thing people have forgotten about.

Every thing is hard coded, you can't replace or remove anything if you wanted too.
All the important stuff, like graphics and audio is mounted on mother board, and expected to be there.

Quote:

but a CISC let you use complex memory addressing modes for many instructions

But that's mostly ineffective, memory is way slower then registers, that's way the powerpc has 32 registers, and they can be used as you like, they are not restricted to data or address register, they all are address and data registers at the same time, use the registers as you like.

As for complex memory addressing I do not think there is anything missing.

I think if you where able to write PowerPC assembler whit the same syntax as 680x0 assembler I do not think we be having this argument now.

@olegil

Quote:

address1 to address two would take a load AND a store, With a spare temp-reg.

that's a case where most likely be using altivec to boost performance.
That's how apple does it, so you be using a vector register.

Last edited by NutsAboutAmiga on 08-Nov-2013 at 01:45 AM.
Last edited by NutsAboutAmiga on 08-Nov-2013 at 01:40 AM.
Last edited by NutsAboutAmiga on 08-Nov-2013 at 01:21 AM.
Last edited by NutsAboutAmiga on 08-Nov-2013 at 01:13 AM.
Last edited by NutsAboutAmiga on 08-Nov-2013 at 12:42 AM.
Last edited by NutsAboutAmiga on 08-Nov-2013 at 12:32 AM.
Last edited by NutsAboutAmiga on 08-Nov-2013 at 12:30 AM.

_________________
http://lifeofliveforit.blogspot.no/
Facebook::LiveForIt Software for AmigaOS

Quote:

rebraist wrote: Strange People we are, indeed. This topic is more present now than there was a (real?) possibility to have an amiga laptop.

The real possibility could be using one of the SAM boards for it, but unfortunately the problem is that there's no "standard" case for a laptop, and producing one would be too much expensive.

Quote:

NutsAboutAmiga wrote: @cdimauro Quote: For the same reason, I certainly don't use the "Classic" label, because there was no "Classic" Amiga, but just and simply the Amiga(s). I understand what your saying but if you do not want to talk about one model, but any computer from Commodore, ESCOM and VisCorp, then I think it makes sense, Commodore was not the only company making Amiga computers. Computers whit classic chipset, is easier to say then computers whit AGA or ECS, PAULA, DENICE, ALICE and so on.

Call them Amiga, and you'll see millions of people immediately recognize of what we are talking about.

Call them "Classic Amiga", and only a few thousands will...

Quote:

Quote: Which is something that you also needed for the Amiga. Well is that true? Most things writen in assembler for classic Amiga, pokes the Joystick ports direct. They never check if there is gameport.resource you can open or anything like that. The same goes whit Paula, you do not check if its there, you know its there so you start setup hardware registers whit out knowing if there or not. And also, if you have program look for cia.resource, no program check what address hardware register are at, thats hard coded in the code. AHI was introduced in 1992, but most games do not use it. VHI is some thing people have forgotten about. Every thing is hard coded, you can't replace or remove anything if you wanted too. All the important stuff, like graphics and audio is mounted on mother board, and expected to be there.

You quoted only the first part of the sentence which was inserted in a specific context, so it completely changed the meaning.
Here's it in its complete form:

"Which is something that you also needed for the Amiga, unless you are using only 512KB of chip ram and the OCS, where you already know all the you want, and can use it immediately."

But I've to add that before directly hitting the hardware you had to acquire the specific resource or the whole system before doing that.

Anyway, a lot of software didn't follow the Commodore's guidelines, as we know.

Quote:

Quote: but a CISC let you use complex memory addressing modes for many instructions But that's mostly ineffective, memory is way slower then registers, that's way the powerpc has 32 registers, and they can be used as you like, they are not restricted to data or address register, they all are address and data registers at the same time, use the registers as you like.

You forgot the data cache and the processor pipelining, that make cached data access as quick as accessing a register from a global performance point of view.

x86 lead the performance crown by many years, and it had only 8 general purpose registers, 8 FPU registers, and then 8 SSE registers. If you take a look at the generated x86 code, it's full of pushes, pops, and accessing to stack frame variables... but it's still executed fast. The reason stays in what I've explained above.

With the x64 the situation greatly improved, because of its 16 g.p. and SIMD registers (FPU ones remained 8), so even the ABI changed, pushes and pops almost disappeared, and a great part of instructions work with registers only, but there's still a good part that uses the memory (again, primarily the stack).

Quote:

As for complex memory addressing I do not think there is anything missing.

Are you joking? The PowerPCs have very limited addressing modes. Take at look a this: http://www.mactech.com/articles/develop/issue_21/21balance.html which made a (rough) comparison with the 680x0.
The 680x0 had far more addressing modes, some of them very complicated.
The PowerPC has base update modes, but it lack the index scaling, which is quite useful.

Quote:

I think if you where able to write PowerPC assembler whit the same syntax as 680x0 assembler I do not think we be having this argument now.

In the end you're suggesting to use a compiler, not an assembler, since the 680x0 instructions are so complicated that they require several PowerPC instructions and some temporary registers to get the same behavior.

Quote:

@olegil Quote: address1 to address two would take a load AND a store, With a spare temp-reg. that's a case where most likely be using altivec to boost performance. That's how apple does it, so you be using a vector register.

I disagree. olegil showed a common instruction in 680x0, which needs at least 2 PowerPC instructions and a temporary register to do the same work, so even introducing a dependency that can hurt the performance in some implementation.
The situation is even worse if we look at instructions which stores the result in the memory...

@cdimauro

Quote:

The PowerPC has base update modes, but it lack the index scaling, which is quite useful.

lbzx, lhzx, lwzx

all load and store instruction for index addressing ends whit x.

http://www.mactech.com/articles/mactech/Vol.10/10.09/PowerPCAssembly/index.html

Quote:

I disagree. olegil showed a common instruction in 680x0, which needs at least 2 PowerPC instructions and a temporary register to do the same work, so even introducing a dependency that can hurt the performance in some implementation.

If you use altivec I do not think is going to hurt speed at all,

http://www.oreillynet.com/pub/a/mac/2002/04/05/altivec.html

AltiVec can offer 4-way parallelism by simultaneously operating on four elements of a vector with each instruction.

So basically 4 times faster then normal code, if you use vector register, so even if takes two instructions, its going to be twice as fast, if you doing 4 moves.

Anyway there other ways to do it, you do not need the CPU to do it at all, get the north bridge chipset to do it, for you.

Quote:

In the end you're suggesting to use a compiler, not an assembler, since the 680x0 instructions are so complicated that they require several PowerPC instructions and some temporary registers to get the same behavior.

if you convert the code into different format, and then assemble that or if assemble the ascii text file into machine code direct does not matter.

(But I guess its easier to convert the 680x0 code into PowerPC code, before the assembler is the easier way to go.)

Because the move.l command does load, store, move between register, direct, indirect, and indexed, and set register values.

This where 680x0 assembler is easier to remember, as powerpc has in most cases a equal.

So its possible to hide li, mr, stw, lwz, stwx, lwzx, under the move.l command.

Last edited by NutsAboutAmiga on 19-Nov-2013 at 12:38 AM.
Last edited by NutsAboutAmiga on 08-Nov-2013 at 08:32 PM.
Last edited by NutsAboutAmiga on 08-Nov-2013 at 08:23 PM.

_________________
http://lifeofliveforit.blogspot.no/
Facebook::LiveForIt Software for AmigaOS

Quote:

NutsAboutAmiga wrote: @cdimauro Quote: The PowerPC has base update modes, but it lack the index scaling, which is quite useful. lbzx, lhzx, lwzx all load and store instruction for index addressing ends whit x. http://www.mactech.com/articles/mactech/Vol.10/10.09/PowerPCAssembly/index.html

I know it, but T talked about index scaling, not simply index. Something like this:

move.l d0,(a0,d1.l*4)

Quote:

I disagree. olegil showed a common instruction in 680x0, which needs at least 2 PowerPC instructions and a temporary register to do the same work, so even introducing a dependency that can hurt the performance in some implementation. If you use altivec I do not think is going to hurt speed at all, http://www.oreillynet.com/pub/a/mac/2002/04/05/altivec.html AltiVec can offer 4-way parallelism by simultaneously operating on four elements of a vector with each instruction. So basically 4 times faster then normal code, if you use vector register, so even if takes two instructions, its going to be twice as fast, if you doing 4 moves.

I know SIMD units, but olegil and I were talking about a completely different thing, which is related to general purpose code which makes (single) moves between memory. Something like this:

move.l (a0,d0.l*4),(a1,d1)

Here there's no Altivec or SIMD that can help, and RISCs require at least two instructions to accomplish the same task, wasting at least one register to hold the temporary, and eventually even intermediate(s), result.

Quote:

Anyway there other ways to do it, you do not need the CPU to do it at all, get the north bridge chipset to do it, for you.

And you'll go even much slower than...

Anyway, the northbridge is a very old concept: it's integrated into the CPU; several years ago. Despite this, may be you are referring to the classic DMA, which needs to be programmed first (and it takes time), adding a lot of latency to complete the operation.

Quote:

Quote: In the end you're suggesting to use a compiler, not an assembler, since the 680x0 instructions are so complicated that they require several PowerPC instructions and some temporary registers to get the same behavior. if you convert the code into different format, and then assemble that or if assemble the ascii text file into machine code direct does not matter. (But I guess its easier to convert the 680x0 code into PowerPC code, before the assembler is the easier way to go.)

I don't understand what is that doesn't matter. It's not clear to me. Can you give more details, please?

Quote:

Because the move.l command does load, store, move between register, direct, indirect, and indexed, and set register values. This where 680x0 assembler is easier to remember, as powerpc has in most cases a equal. So its possible to hide li, mr, stw, lwz, stwx, lwzx, under the move.l command.

Only some 680x0 instructions can be directly remapped to PowerPC ones. Take a look at my examples, which also aren't that much complicated.

P.S. PowerPC mnemonics are horrible...

@cdimauro

General cases are available. If you don't believe me, look at the cases System76 uses for its custom Ubuntu laptops. All made from standard produced cases which they happen to buy and then fill 'em up with compatible hardware. If they can do it, then any company can.

_________________
Proud user of AmigaOS 4.1 on an AmigaONE 500. This is the first Amiga I've ever had so I don't know all the ins and outs of AmigaOS yet, so I'm sorry if I'm asking noob questions and stuff.

@cdimauro

Quote:

That's also why the RISC code density is poor, whereas CISCs do much better (and the 68K was and is one of the best).

While CISC architectures can "do more with a single command", as there are loads of different variants of each command, that approach has its problems.

1)Some variants of a command might be practically useless. Sure, you might save space, but some variant might be SLOWER than doing the practically same thing with more commands. Still useful for 4k intros I guess...
2)Some less used variants might be dropped from future models, and emulated... Slowly... That's what happened with 68040 and 68060. Of course this could be done on ANY architecture.
3)There can be variance of what you can do with what commands. Some addressing modes might be only available to some etc.
4)Even if amount of commands is smaller, the amount of space needed might not necessarily be. See ARM Thumb, for example, which basically uses 16, rather than 32 bits per command (and obviously offers more limited set of commands).
5)There are things beyond "CISC vs. RISC" that can reduce amount of code needed. For example ARM has conditional execution on most (all?) commands, which seriously reduces amount of branch commands. Basically you can set any command to execute f.ex. only if the zero flag is set. No need to add a branch command to skip a single instruction.

Of course 680x0 has also its benefits, but I'm honest. I don't know that much about it on the "lower level", but a few things about ARM and a lot more about 6510.

And in the end, I guess 680x0 was the best option available at the time Amiga 1000 was in development, so it's rather pointless to argue which more recent option might be better

Quote:

Jupp3 wrote: @cdimauro Quote: That's also why the RISC code density is poor, whereas CISCs do much better (and the 68K was and is one of the best). While CISC architectures can "do more with a single command", as there are loads of different variants of each command, that approach has its problems.

Yes. I never stated the contrary.

Quote:

1)Some variants of a command might be practically useless. Sure, you might save space, but some variant might be SLOWER than doing the practically same thing with more commands. Still useful for 4k intros I guess...

Yes, but in the end it depends on the micro-architecture (architecture implementation), and by the transistors budget.

Quote:

2)Some less used variants might be dropped from future models, and emulated... Slowly... That's what happened with 68040 and 68060. Of course this could be done on ANY architecture.

Of course it can happen, and happened with the 68K and some PowerPCs.

But x86 has never dropped instructions. x64, since it was a new ISA, removed some "legacy" opcodes from the x86, even if it was built on top of the latter.

In the end, removing instructions and emulating them can be a feasible way, but it doesn't mean that the overall execution will suffer. It depends entirely on the kind of the removed instructions. For example, legacy ones are and can be very rare, so removing and emulating them will have less or no impact on the performance, whereas removing the long MUL and other FPU instructions from the 68060 or another instruction (I don't remember the mnemonic now; sorry) from the PowerPC G5 severely hurt the performance.

Quote:

3)There can be variance of what you can do with what commands. Some addressing modes might be only available to some etc.

Sure, but compilers can deal with it. For humans it's a bit more difficult, but they can survive taking care of these asymmetries.

Quote:

4)Even if amount of commands is smaller, the amount of space needed might not necessarily be. See ARM Thumb, for example, which basically uses 16, rather than 32 bits per command (and obviously offers more limited set of commands).

Thumb (and Thumb-2 & ThumbEE) is... a CISC. It has a variable-length instruction set (16 and 32-bit opcodes), and that's why it achieves a very good code density.

In the end, (some) RISCs have dropped the simpler fixed-length opcode length to compete with CISCs in the code density area, which is still a very important factor even on multi GBs systems, borrowing one of the most distinctive characteristics of the CISCs...

Quote:

5)There are things beyond "CISC vs. RISC" that can reduce amount of code needed. For example ARM has conditional execution on most (all?) commands

Yes, for all. But... see below.

Quote:

, which seriously reduces amount of branch commands. Basically you can set any command to execute f.ex. only if the zero flag is set. No need to add a branch command to skip a single instruction.

I know it very well, but it can work only if you have very little pieces of code that are executed conditionally, otherwise the performance will be greatly affected (because you have to execute ALL the instructions of both code-blocks).

It means that it needs to add two instructions to "emulate" the same thing for a processor which hasn't such feature, one conditional branch and one unconditional branch. But only the former can create performance penalties if the true and false conditions have almost the same probability to be executed.

Anyway, it seams that implementing the conditional execution of every instruction can make problems. That's why ARM removed it from its brand new 64-bit ISA (ARMv8 aka ARM64); instead, she added only some very specific instructions that can be executed conditionally, such as the well known conditional move.

It was a wise decision, since the conditional execution can only be used for very small code blocks, as I said; modern processors have branches prediction units that can give very good results as well, with the enormous advantage to don't steal 4 precious bits from the opcode table...

Quote:

Of course 680x0 has also its benefits, but I'm honest. I don't know that much about it on the "lower level", but a few things about ARM and a lot more about 6510. And in the end, I guess 680x0 was the best option available at the time Amiga 1000 was in development, so it's rather pointless to argue which more recent option might be better

Who knows. May be an enhanced 68K ISA will come from some FPGA guys.

@cdimauro

Some discussions about architectures and internal workings of computer systems is appropriate in just about any Amiga forum message, but please try to keep any off-topic, or side-topic discussions brief, or start your own forum thread to discuss the finer points that you seem so passionate about debating.

All I am saying is that we like to keep most postings on topic of the original thread posting, though some thread creep into related side topics will always happen.

_________________
Amiga! The computer that inspired so many, to accomplish so much, but has ended up in the hands of . . . . . . . . . .

I apologize. I invite who wants to continue the discussion about architecures to open a new thread.

@KimmoK

Quote:

using 1Ghz $30 PPC chip

Which chip do you think of?
The ones that I know cost more than $100

Last edited by elwood on 20-Nov-2013 at 10:42 AM.

_________________
Philippe 'Elwood' Ferrucci
Sam460 1.10 Ghz
AmigaOS 4 betatester
Amiga Translator Organisation

@elwood

IIRC APM86791 has been available for about USD30 (now it seems to start from 55).

I think there exists several 1ghz PPC chips below 100 eur.

And if extremely low cost board is designed, it's production run can be above 1000 units in one go.
(via crowd funding or some business case for it, it could be 10 000 units... still not cheaper than Pi but several times more powerfull)


For the low end I think T10xx chips are also interesting. Several times the performance of APM parts and possibility to use simple 2D gfx from the soc. (quad core seems to become available for about USD160 if I'm not mistaken.)

Last edited by KimmoK on 20-Nov-2013 at 02:16 PM.

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

@Vistaus

Just had a look at the System76 cases and they seem quite deep, possibly deep enough for a Sam440 or Sam460 board.

I'm willing to offer my Sam440ep, currently working and has OS 4.1.6 installed, up for a good modding, anyone interested?

_________________
1 AmigaOne G4XE (OS4 Pre-Release Update4)
Minimig
Sam440ep + OS4.1FE
Sam460cr + OS4.1FE

@elwood

I didn't say this, but:
Keep an eye out for T1023. Core from P5020 and the peripherals of P1025. Good match, awesome price as far as I've heard.

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@1Mouse

While looking for laptop cases, I stumbled upon an article from Computer Shopper about building your own laptop...by Denny Atkin....hey, wasn't he an Amiga guy who wrote books about Amigas back in the day?

http://www.computershopper.com/feature/build-your-own-laptop

Scott

_________________
AmigaOne uA1-c 512M RAM - Only Amiga Makes It Possible!
Check my blog AmigaOne Computing

@scabit

Quote:

scabit wrote: @1Mouse While looking for laptop cases, I stumbled upon an article from Computer Shopper about building your own laptop...by Denny Atkin....hey, wasn't he an Amiga guy who wrote books about Amigas back in the day? http://www.computershopper.com/feature/build-your-own-laptop Scott

This article is really about customizing a barebones/whitebook laptop. What our goal of an Amiga laptop is, is much more than that. The whitebook comes with the motherboard, you just fit which processor, memory and hard drive you want into that. Our biggest problem is there is no "standard" laptop shell to target a Amiga "standard" laptop motherboard design to.

So we're left to hack about with what we have. Find a laptop shell that coincidentally can sortof, do a good enough job of taking in a small Sam motherboard. Or make a new shell to fit one of the availabel small motherboards. With the time I spent imagining that for a MicroA1 board, I personally feel that the best that can be done is still big and clunky. Or, we can design a new motherboard to fit into some designated PC laptop shell as a direct motherboard swap, and hope that shell stays on the market long enough to get the Amiga motherboard done and for sale. Expensive and tricky, though likely the most satisfying result.

If I were to "build" an Amiga laptop, I would choose the specs and features required (which CPU, do we want Expresscard slot or not, SDHC or SDXC, etc...) first. Then, do the schematic design, no real layout yet. Once the schematic design is very well along, look at available whitebook/barebones etc. laptops for what fits with the chosen specs/features. Buy a sample or two and tear it apart, measure out the motherboard dimensions and component placement. Now that you know what shape it is and where important items are located, start implementing the PCB layout of your schematic design. Now, hopefully the chosen laptop shell is still buyable after the motherboard implementation design is done, tested, proven, and sellable. Hopefully the heatsink kit can be reused, if not then that needs redone as well. Ideally, one could reuse a good bit of the original PC laptop, whatever memory modules it comes with, perhaps separate button/LED and power PCBs for example, MXM graphics card if done that way, if not all integrated in the motherboard. And whitebooks can help reduce the waste, if you can buy them without CPU, Windows license fee, different type of memory modules etc.

Now, I myself prefer high performance, and I'll say that I'd be a bit of a snob about things looking very nicely done, so I wouldn't likely be fully satisfied with a Sam board screwed inside of an old briefcase. I realize that what I would do is a huge undertaking and I obviously haven't turned that into a successful business myself, but I do remain hopeful that such a thing finds a way to happen.

_________________
All glory to the Hypnotoad!

@billt

Or, you can buy a few thousand units of an obsolete model off of eBay for peanuts. They'll usually be missing the harddrive mounting hardware and need new batteries. Both of those are issues we could solve.

Example: 125 results for "d630 ram hdd" (Dell Latitude Core2Duo). Only three of which are not for complete laptops.

Not saying this idea is PERFECT for your ideals, but it would be a valid choice that noone would jump off a cliff because of. Probably. Whether WE would be PUSHED off a cliff I'm not so sure, though. Inside it you'll need graphics (I've stopped believing in MXM, I'm afraid).

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.