Hi DrBombcrater,


Comes down to what I've been saying all along, ibum threw CPU's onto the market and said "Have fun boys", meanwhile the idiots at motoscum FINALLY realized "Hey, no one can use our CPU's on a motherboard, gee, why's that? Ohhhhhhh yeah, SUPPORT CHIPS, duuuuuuuh, Dilbert LIVES!!!!!!!!"



ibum! motoscum! "Duuuuuuuh, Dilbert LIVES!!!!!!!!"

@DrBombcrater

Quote:

DrBombcrater wrote: What I'd rather see is a board built around one of Freescale's new SoC chips. These have quite simply outstanding specs. Take the MPC8548E, for example. It has : * Core speeds up to 1.5GHz. * Support for 400MHz DDR and 667MHz DDR2 memory. * PCI-Express capability (8 lanes). * 2 x 32-bit PCI busses, or 1 x 64-bit bus. * 4 x Gigabit ethernet interfaces. All in a single chip. An A1 built around this would have a lower component count and simpler layout than an XC-style board, and thus be cheaper, and all the critical functions are designed by Freescale so there's less chance of bugs. If Eyetech are not planning such a system, I hope ACK, BPlan or somebody else is.

I have the high level plans for such a board already sitting on my HD

But you would want the 8545E chip, as it is the only one in the 8548 series that comes with a double precision FPU. Problem is that the series is based upon the e500 core. This is only user-mode integer compatible to the PPCs used in the A1 today. Supervisor-level software parts (HAL of OS4) would need adaption to this core.

And much worse I contatcted Freescale regarding the FPU implementation and is not opcode compatible to the normale PPC FPU They suggested to me using their new e600 based products (in fact the 8641, which is rumoured to be the basis for the Peg3), as these are compatible (they are more like G4s with periphal units attached). Problem with the 8641 is it samples end of '05 and is said to be shipping 1/2Q of 06... too long a wait.

The 8545 should be released any time soon. But FPU compatability would need a recompile of software or a kind of JIT-emulation of the PPCs FPU ops. Both should be possible, but would be either need all apps to be compiled in two modes or would maybe mean hard work implementing the JIT (emulation without JIT the execution speed would be too slow, I think, but would be enough for a start).

So is anyone interested in constructing, manufactoring and selling a board based on the 8545!?

My board spec is:
1.2GHz 8545E,
2x DDR(2) DIMM slots,
1 PCIe-Graphics (PEG) slot (4x == AGP4x),
2-3PCI slots,
1-2x GBit-LAN,
1-2x RS232 (for debugging you know ),
2x SATA + 2x PATA,
4x USB2.0
and maybe an onboard CMI sound chip.

(No PS2 or parallel ports, these would need more chips on the board and there exist cheap USBtoX adapters.)

And then we also need some software guys doing UBoot (Freescale has done it for the normal UBoot version), the OS4 HAL adaption and maybe the GCC backend adaption (should be available from Freescale, too) and an FPU-JIT. A damn lot of work. (BTW Freescale emphazises that the chip makes board design easier )

(Note: I am not working as a HW engineer by myself, just read a book about embedded HW design some time ago that got me started a bit in that direction. I could act as a consultant and SW guy on the project...)

Hi fisk,



to AmigaWorld.Net !!!


Some people are using their AmigaOnes at 1 GHz already, so it seems like a LOT of trouble to go through, just 2 get 2 1.TWO Ghz?


Not against it, but doesn't seem very "advanced" for the amount effort involved.



AmigaOne! AOS4.0! A highly advanced OS belongs on highly advanced HW!

@fisk:

Maybe you should start a new thread on the subject

_________________
If you need music for your productions, or graphics for your creations, feel free to contact me.
also check out my music at http://www.contrazt.no/records.html

@Atheist

Quote:

Atheist wrote: Hi DrBombcrater, TOTALLY AGREE! We need a full sized motherboard. An XC has to be just that! Minimum specification: -- Firewire2 * 2 -- USB2.0 * 2

I suggest 6 USB 2.0 ports i.e. this avoids purchasing yet another USB Hub.

Quote:

-- Slot for extra CPU, for dual operation, OR speed increase. If it had it's own memory slot, upto 1 Gig, it could run AOS4.0 separately from first CPU, and be "gotten to" through "screen" selection.

It may need some kind of virtualisation technology (i.e. similar to Intel's Vanderpool, IBM's Virtualization Engine) to run separate instances of AOS4 (assuming the machine is a dual CPU with intergrated memory controllers).

Quote:

It would communicate to the rest of the computer components, by making requests to the first CPU, as though it was an independent remote computer. Imagine, it could solely run dnetc RC5, or render, or run a message board, while you're using your "other computer".

Such multitasking would be OK if the OS can take care of itself i.e. implements full application memory protection and SMP support (PS I'm aware of AOS4's SMP support).

On the other hand, purchasing two uA1 could to trick.
Remember Siamese RTG system?

Quote:

-- Onboard RAID, with dedicated ram slot for 64 to 256 meg ram buffer.

Would you gain extra HD performance with 64MB over 8MB and within reachable cost?

PS; WinXP dynamically allocates large chunks of main system ram for such HD cache activities. This runs best on 1GB or greater RAM.

Quote:

-- PCI Express 1 or 2 slots -- SATA connector on MB -- Custom Case (A2000 like), with 2 compact flash card slots on the front (It would be cool if a keyboard had 2 compact flash card slots on it, and you could plug one in, and boot off of it, TAKE THAT pc users!!!

Note, HP's SR1000Z has built-in 9in1 card reader (connected to USB).

Refer to http://ucsu.colorado.edu/~shaher/Bootable_USB.html
Quote:

It would have to be a USB2.0 keyboard, though.) and IrDA sensors in the front.

Intel is already planning for desktop wide Centrino (going beyond IrDA and Ethernet) standard. This is related to Microsoft?s Athens PC reference. Refer the year 2000 era IBM T20 Laptop for "IrDA and Ethernet" standard.

Last edited by Hammer on 24-Jan-2005 at 09:38 PM.
Last edited by Hammer on 24-Jan-2005 at 09:35 PM.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@Atheist

Quote:

Atheist wrote: Some people are using their AmigaOnes at 1 GHz already, so it seems like a LOT of trouble to go through, just 2 get 2 1.TWO Ghz? Not against it, but doesn't seem very "advanced" for the amount effort involved. AmigaOne! AOS4.0! A highly advanced OS belongs on highly advanced HW!

I only have 800MHz-G3 A1 and it is damn fast using OS4. Yesterday I was playing 3 MPEG2 movies at once, with sound and all of them played without hickups )) Only for number-crunching or video encoding a faster CPU would be nice. For everyday use the 800MHz are more than enough on OS4 (did I mention I ran AROS demos side by side on A1/OS4 and my 1.6GHz Athlon PC/AROSmax? They seemed to run faster on the A1!!!!)

Maybe you have 1 GHz, but you only have a 133MHz RAM bus and all external I/O except graphics is going through the one PCI bus of the Articia You only have USB1.1 and PATA onboard, no GigE... Also you pay a lot for the chips and the socket on the current A1 boards, I guess (Articia is 50$ alone, the MegArray CPU socket is also not cheap).

A board based on the 1.2GHz 8545E (soldered to the board) and the features mentioned above may sell below 500$ when produced in same quantities as A1, I guess.

Did I mention the 8545E produces less heat than the G4!?

A 1.33 GHz version is also available (but initially too expensive, I think) and a 1.5 GHz on the horizon.

Also the 8545 has some special processing engines (one SIMD, sadly not Altivec) that can speed up e.g. TCP-IP a lot (would need special drivers, though; but Freescale provides example code anyway )

So I don't think the processing power is a problem. I more think the ports and busses on the current A1 are a problem. The CPU could handle the data faster as it is actually being delivered...

I really would like to see an inexpensive board like the one I described

@Atheist

Quote:

AmigaOne! AOS4.0! A highly advanced OS belongs on highly advanced HW!

Hyperion has to finish AOS4.0 first i.e. AOS4.0-RTM-OEM status.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@Agafaster

No, the REAL confusion begins here:

mini-ITX is smaller (by about a mile) than micro-ATX. Doh. Whoever thought of that one needs a beating. Badly.

I wish we would just adopt the BTX standard, as it has three major sizes available. Normal, micro and pico. Where pico is more or less what mini-ITX is today (just barely enough to fit the connectors ).

BTX also makes PCI/AGP chips sit on the UPPER side of the PCBs, so cooling would be a heck of a lot easier...

_________________
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.

@fisk

Not all software in the olden days needed FPU. Why would all software now need FPU?

_________________
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.

@fisk

Quote:

And much worse I contatcted Freescale regarding the FPU implementation and is not opcode compatible to the normale PPC FPU

Oh heck, none of the information I've seen on the e500 code suggested code incompatibilities. That's a pretty serious bummer.

Quote:

Problem with the 8641 is it samples end of '05 and is said to be shipping 1/2Q of 06... too long a wait.

Indeed, we need something that's available now, not next year.

Quote:

So is anyone interested in constructing, manufactoring and selling a board based on the 8545!?

I think the level of investment, in terms of both time and money, required to get OS4 booting on a processor not totally binary-compatible with the 750/74xx series is simply far too large to be considered. I imagine Hyperion would be reluctant to see OS4 on such a system anyway.

And the unexpectedly impressive specs of MAI's Articia-P board, the Teron PX2, mean the gap between that product and something based on one of the e500 chips would be too small to be viable. The advantages of PCI-Express, GigE, and perhaps a lower price are not enough to sell the product in the kind of numbers required to recoup the investned needed.

I proposed using Freescale's SoC chips because the PX2 (aka A1-XC) was supposed to be a micro-ATX board similar in design to the Pegasos2. In reality it's a very expandable, bandwidth-laden monster with some rather innovative features, and I think that pretty much kills any market gap for SoC boards except at the low end or, as ACK are doing, as A1200/A4000 accelerators.

_________________
Who do you serve, and who do you trust? - Galen

@olegil

Quote:

olegil wrote: @fisk Not all software in the olden days needed FPU. Why would all software now need FPU?

The old machines had no FPU, so software was adapted. Today most CPUs have FPU, so even compilers use them to optimize code that has no FPU instruction in it

But you are right, not all SW needs FPU; in fact in often you can even calculate with fixed point integer (scaled) and thus avoid FPU instructions This is how embedded controllers do their math. Only it can get quite complicated and error-prone to manually do the scaling and re-scaling Nothing a deskto developer wants to fiddle around with, I guess...

Of course re-compilation will always help, even if it would be using the math libs then.

@fisk

Quote:

(did I mention I ran AROS demos side by side on A1/OS4 and my 1.6GHz Athlon PC/AROSmax? They seemed to run faster on the A1!!!!)

Note that most of those demos are hardcoded to 50 FPS. You need to run a lot of them at the same time to get cpu 100 % busy. And the AOS4 porters cheated by changing them to use direct bitmap access, instead of the orig. WriteLUTPixelArray() calls AFAIR. Also I'm not sure if WaitTOF() which is used for 50 FPS speed limit, works same on AOS4. There it possibly is connected to real refresh rate of the screenmode, not the "virtual/assumed" 50 Hz in case of AROS.

@fisk
Quote:

did I mention I ran AROS demos side by side on A1/OS4 and my 1.6GHz Athlon PC/AROSmax? They seemed to run faster on the A1!!!!)

There are several Athlon XPs at 1.6Ghz btw (e.g. FSB133**/FSB266/FSB333/FSB400, Palomino, T-Bred-A/T-Bred-B/Thornton, Barton).
**On motherboards such as VIA KT133A.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@DrBombcrater
Quote:

Oh heck, none of the information I've seen on the e500 code suggested code incompatibilities. That's a pretty serious bummer.

That is one problem, but I think this is only minor, since it is isolated in the HAL of OS4. And that is adaptable in about a week according to a post of the Friedens, I have seen somewhere before here on aw.net.

Quote:

I think the level of investment, in terms of both time and money, required to get OS4 booting on a processor not totally binary-compatible with the 750/74xx series is simply far too large to be considered. I imagine Hyperion would be reluctant to see OS4 on such a system anyway.

I would run the project non-profit perhaps (I guess there are people here who have the skills; DigitalDisaster e.g.). That would be no development costs and only mean cost for prototypes etc. And then try to find someone who produces the board. This would ideally be Eyetech or ACK, as they have a license by Amiga. And I also could imagine Hyperion is interested in any board which sells bundled with OS4 During HW development a eval board of Freescale could be used to experiment with the chip and to get some basic stuff already ported (Kernel; UBoot with drivers etc., whereas there are already drivers for the e500 for UBoot).

Quote:

And the unexpectedly impressive specs of MAI's Articia-P board, the Teron PX2, mean the gap between that product and something based on one of the e500 chips would be too small to be viable. The advantages of PCI-Express, GigE, and perhaps a lower price are not enough to sell the product in the kind of numbers required to recoup the investned needed.

I am not really satisfied with the ArticaP's (current!?) specs. No PCIe, no DDR2 RAM. I would by an A1-XC based on it, though. It's better than the current A1, but it is behind already the day it appears. The PC world is moving to PCIe and DDR2 right now. Sales maybe a problem. So the board would also need to be sold as a linux product and/or as a cheap EVB for the 8545 itself (like Genesi is doing with the Peg2 for G4). That would call for a kind of dongle for OS4, though (USB preferred).

Quote:

I proposed using Freescale's SoC chips because the PX2 (aka A1-XC) was supposed to be a micro-ATX board similar in design to the Pegasos2. In reality it's a very expandable, bandwidth-laden monster with some rather innovative features, and I think that pretty much kills any market gap for SoC boards except at the low end or, as ACK are doing, as A1200/A4000 accelerators.

Question is, will the A1-XC ever exist!? Maybe we, the community, need to help Eyetech or ACK developing a new board without them paying too much for the development itself. And as the 8545 simplifies board design (a very huge plus compared to an Articia based solution!) it would be a good choice. Just the software side is a bit tricky (see my postings above). But with help of Hyperion should be no problem.

The 8641 would be even a better choice, but I guess it will be more expensive and it is ot available in the near future, as mentioned before.

@Hammer

Quote:

Hammer wrote: @fisk Quote: did I mention I ran AROS demos side by side on A1/OS4 and my 1.6GHz Athlon PC/AROSmax? They seemed to run faster on the A1!!!!) There are several Athlon XPs at 1.6Ghz btw (e.g. FSB133**/FSB266/FSB333/FSB400, Palomino, T-Bred-A/T-Bred-B/Thornton, Barton). **On motherboards such as VIA KT133A.

Athlon XP 2000, T-Bred(-A?) with KT400 and DDR-333 RAM (CL3) that is

Don't worry, I did not want to flame on the Athlon (better than P4), I just wnated to show that we do not need much GHz for OS4

@Georg

Quote:

Georg wrote: @fisk Quote: (did I mention I ran AROS demos side by side on A1/OS4 and my 1.6GHz Athlon PC/AROSmax? They seemed to run faster on the A1!!!!) Note that most of those demos are hardcoded to 50 FPS. You need to run a lot of them at the same time to get cpu 100 % busy. And the AOS4 porters cheated by changing them to use direct bitmap access, instead of the orig. WriteLUTPixelArray() calls AFAIR. Also I'm not sure if WaitTOF() which is used for 50 FPS speed limit, works same on AOS4. There it possibly is connected to real refresh rate of the screenmode, not the "virtual/assumed" 50 Hz in case of AROS.

Oh, I did not know that. I ran 3 demos at once (flamme, dawfire and metaballs; I like these the most ). At least the A1 did not show any slow down compared to the 2x faster clocked Athlon with faster RAM etc.

@fisk

About the 8641, one thing I've noticed about Amiga hardware is that it consistently shippes to late. To me this suggests that the development times are long for a company lacking in resources like Eyetech. So while the 8641 and 8641D might be far into the future, I wouldn't be suprised to see that the development time of a brand new A1 board based on it would be such that it'd be finished in about the same timeframe as when the first chips [8641s] would start showing up.

[Disclamer]I've never designed a mother board of anything like the complexety of a full fledged "PC" motherboard[/Disclamer]

@fisk

Quote:

I am not really satisfied with the ArticaP's (current!?) specs. No PCIe, no DDR2 RAM.

PCIe would be nice, no doubt about that, but it's probably just a marketing advantage right now rather than a technical one. Same for DDR2. There's no significant performance advantage over AGP and DDR, and currently AGP cards and DDR memory are easier and cheaper to source than PCIe and DDR2 stuff.

Given the complete lack of any ATX boards in the Amiga market right now a good machine this year is more useful than a very good one next year.

Quote:

Question is, will the A1-XC ever exist!?

That's up to Eyetech, but their AmigaOne products have previously just been copies of MAI's reference boards and the Articia P ref board, the Teron PX2, already exists.

Quote:

And as the 8545 simplifies board design (a very huge plus compared to an Articia based solution!) it would be a good choice.

I agree, but in this specific situation it's a moot point because the Articia based board already exists, while it would take a significant amount of time and work to get an 8454 based system ready.

_________________
Who do you serve, and who do you trust? - Galen

@Eric_S
Quote:

About the 8641, one thing I've noticed about Amiga hardware is that it consistently shippes to late. To me this suggests that the development times are long for a company lacking in resources like Eyetech. So while the 8641 and 8641D might be far into the future, I wouldn't be suprised to see that the development time of a brand new A1 board based on it would be such that it'd be finished in about the same timeframe as when the first chips [8641s] would start showing up.

I don't think development is the main problem. Production of the end user board is the problem, I think. Buying the parts, finding a manufactorer. All at reasonable costs... And maybe finding and fixing undocumented HW bugs for the next board revision

Quote:

[Disclamer]I've never designed a mother board of anything like the complexety of a full fledged "PC" motherboard[/Disclamer]

Me neither

Maybe we just need to point some people in the right direction

@DrBombcrater

Quote:

That's up to Eyetech, but their AmigaOne products have previously just been copies of MAI's reference boards and the Articia P ref board, the Teron PX2, already exists.

Ok, that makes sense. If the board already exists it's easier to use that. Still you would need to add at least USB2.0 on it. From the picture I can see it still uses the famous VIA southbridge (shudder).

Any A1-XC will be fine, I guess, at least if it's more advanced than the current models and maybe a bit cheaper I would not count on the latter one, though

A new design would take about 18-22 weeks, AFAIK. And then time for the software adaption (could be done partly in parallel, as noted above).