@iggy

Do you thimk it is the moment to male public the specs we discussed yesterday on a mail amd check what are the opinions of community about it?

_________________
"When the Amiga came out, everyone [at Apple] was scared as hell." (J.L. Gassée, former CEO of Apple France and chief of devs of Mac II-fx, interviewed by Amazing Computing, Nov 1996).

@Raffaele

No, we are still discussing options.
The specs I sent you were sent to an interested party that asked for a single statement/document/message with a concise list of options.
So that is just the base.
Feel free to keep throwing out additional idea.

If we have everything winnowed down by Friday, we'll see what everybody thinks about categorizing (or grouping) the different specs.

@thread

And then take a look at THIS. Scroll to the bottom and take a close look at the price. Make sure to be seated properly. What the heck are we talking about?

(And I still come to love that Mezzanine board concept even more and more: what NVidia shows is quite what I have in mind as a viable base for an Amiga-platform!)

From what I gleaned in this thread, some parts of a design need multilayered boards. I read about 7 layers and beyond. Which parts would benefit from that? I mean, when referring to 'my' cut-up designproposal, which parts would only need a less layered design and which parts would benefit from an extensively layered design?

OldFart

Last edited by OldFart on 11-Nov-2015 at 01:08 PM.

_________________
More then three levels of indigestion and you're scroomed!

@OldFart

Nice ARM board. 64bit, fast.
I don't think it makes sense, consider the prices and power available to consider anything other than 64 bit (even if we aren't there yet).

But right now we are running on PPCs, and what we have been discussing is an alternative PPC board.

MorphOS eventually will move to X64 (AROS is already there), OS4 could be moved to ARM (eventually (but who knows).

Anyway...right now...

Looking at the costs and benefits, I'm beginning to think that those pointing to the e6500 cored CPUs might be right.
In particular the T2080 and T2081.

They are more expensive than the P1022 or T10XX CPUs, but less than the P50XX CPU.

Four core, eight thread, up to 1.8 GHz, but with a faster core that has both a standard FPU and AltiVec instructions.

And a T2080 could be the base of a system without a Southbridge, or either could be used with one.

@iggy

Quote:

Nice ARM board. 64bit, fast. I don't think it makes sense, consider the prices and power available to consider anything other than 64 bit (even if we aren't there yet). But right now we are running on PPCs, and what we have been discussing is an alternative PPC board.

Oh, sure! What I meant was two-fold: both the mezzanine-concept which I hold in high esteem and the incredible high price for this system: 600 USD now and half that next year. I was merely looking at the conceptual properties and not so much (read: absolutely NOT) proposing the NVidia board as an alternative for OS4's PPC base.

Quote:

MorphOS eventually will move to X64 (AROS is already there), OS4 could be moved to ARM (eventually (but who knows).

I, for one, am simply not interested in x64. Never been, never will be. Not that I think it is 'bad' per se, but I do not feel any enthousiasm for such a system.

Quote:

Anyway...right now...

You know, I like this thread and am eager to see where it is going to. I think it heartwarming to see how much specific knowledge is hanging around here. Would love to see more of them to chime in!

OldFart

_________________
More then three levels of indigestion and you're scroomed!

@OldFart

You can't route 789 wires out of a 23x23mm package without using multiple signal levels.

Anyone who is the LEAST interested should watch ALL of this:
https://mentor1.adobeconnect.com/_a781163502/p24156661/

_________________
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.

@OldFart

Quote:

OldFart wrote: And then take a look at THIS.

Thanks for the link. Not cheap, but it sure looks like FUN.

I was hoping to do a BeagleBoard on a card and turn the X1000 into a carrier for it. Just doesn't seem to be any interest if it's not 3D or old games.

Too bad Freescale or Acube don't do something similar with PPC.

Yeah, think I'll get me one of those in the Spring.

Sorry for the off topic guys. Please continue.

_________________
Tinkering with computers.

@olegil

Quote:

Anyone who is the LEAST interested should watch ALL of this:

I took a look at that and i must say it is very informative, although i was intrugued at first with the slider at the bottom: what the heck was that doing there and for what reason? Until I got it was a reading time indicator. Knowing that made me rather less nervous...
I'll give it a good look when I've got some more time and admittedly I will need to go through it more then once to grasp the fuller picture.

Thanks for the link!

OldFart

_________________
More then three levels of indigestion and you're scroomed!

@OldFart

Quote:

And then take a look at THIS. Scroll to the bottom and take a close look at the price. Make sure to be seated properly. What the heck are we talking about?

Cool. I might save up for one as a toy to tinker in ARM stuff with. The comments at the bottom have people saying its too expensive, but I think your intention was to show us Amiga people that somehting can cost only that much, which is a ways below what we're used to for several years. I'd be more interested if this was using a standard module format. I haven't read far enough to know if that is or not, I just don't know yet, but I don't recognize it. If yes, then an even bigger "Neat!". I can get student pricing for now, so I may try for that. I'd like to get an Nvidia Shield for my TV at some point.

My obsession with ComExpress is because it's a popular standard, it's essentilly a full-blown PC on the mezzanine module, and they planned for SODIMM slots, which other module standards did not. The SODIMMs are important to me for expansion options, as I'm very interested in possibility of upgrading everything possible, and as little as possible soldered down as is reasonable (ie don't solder X MB or GB of RAM and not let me change that). Freescale even has their own flavor of the module pinout, though I'm not yet sure if that's as useful as a standard Com-Express type-6 module goal, should we go in that direction. I've followed this one since 2004, when it was called ETX-Express at Kontron before it became a PICMG standard. (I'd previously followed Kontron's previous ETX format as a similarly neat idea, but ETX-Express/COM-Express are indeed a better standard) Others are Core-Express (similar to COM-Express but not as popular), Qseven, and some others I had on a wiki page that needs put on a new server so I can remember them again. (Curse you WikiSpaces for changing policies!)

As Olegil has mentioned by now, you need layers to get to all the BGA balls or pins of the bigger chips. (Even smaller chips now, as the BGA balls are getting too close together to run many traces between them)

The CPU I currently obsess over, t4240, has almost 2000 BGA balls (somewhere more than 1900 of them). You need layers to get to them all. That does not mean dozens of layers, but some. T2080 is less severe of the same issue. Southbridges are less severe of the same thing. T1040 same thing... etc.

Last edited by billt on 11-Nov-2015 at 04:28 PM.
Last edited by billt on 11-Nov-2015 at 04:27 PM.
Last edited by billt on 11-Nov-2015 at 04:20 PM.

_________________
All glory to the Hypnotoad!

@iggy

Quote:

Looking at the costs and benefits, I'm beginning to think that those pointing to the e6500 cored CPUs might be right. In particular the T2080 and T2081.

T2081 loses half of the Serdes on the T2080. The T2081 loses the SATA entirely if I read things correctly, so would have you needing a Southbridge more than wanting it, to be able to use . The T2080 may be a doable thing without Southbridge, though I still encourage it.

My philosophy, if you HAVE to add a SATA controller to T2081 and use PCIe to connect it, then why not chose one of the Southbridge parts to be that, and then get more than SATA as part of the deal. I think you'd be crazy not to, even if a dedicated SATA-only chip might be a couple US$ cheaper. IMHO, it's a better use of that one PCIe port (of however many lanes it gets to be).

For those who argue against Southbridge, then go with T2080 or a T10xx part that lets you keep the SATA connections, and forget about T2081.

Last edited by billt on 11-Nov-2015 at 04:36 PM.

_________________
All glory to the Hypnotoad!

@billt

Quote:

The SODIMMs are important to me for expansion options, as I'm very interested in possibility of upgrading everything possible, and as little as possible soldered down as is reasonable...

Well, that 's as close to my reasoning as possible when I mentioned modularising the design by cutting it up: replace/add/remove/upgrade/downgrade/wreck as much as you can and have lust for; you'll end up with something suitable at all times.
I can imagine your obsession as you call it with ComExpress modules.

With regard to the topic of layering I take it that the part around the CPU and memory modules is the toughest one, but is a PCIe bus equally complex requiring equally complex layering or can it be made to function with substantially less?
These thoughts keep me busy (in a manner op speaking, that is).

OldFart

_________________
More then three levels of indigestion and you're scroomed!

@OldFart

Quote:

The SODIMMs are important to me for expansion options, as I'm very interested in possibility of upgrading everything possible, and as little as possible soldered down as is reasonable...

And now that I've said that, I also say other things to the effect of maximizing the features soldered on, so let me try to explain that duality.

I want as many opprtunities to update or expand as possible. So don;t solder RAM down and tell me that's the only allowed amount. Give me an SODIMM or a desktop DIMM slot instead.

I do want a lot of features, and part of that is my wish for a laptop, where later addons are harder to do for many things.

If a connection must be consumed for some fundamental feature, then get as much as possible out of consuming that connection. So that if a CPU is selected which does not provide its own SATA, and a PCIe chip must be added to gain this fundamental feature, then go further than SATA and maximize what we get out of that consumed PCIe connection. If it's not a slot to change things later, then don't minimize what we get from it to only one feature.

So I am encouraging to solder in that Southbridge, but that is kindof a special case to save a handful of slots for other things not in a Southbridge.

So my ideal is a combination of maximum features and maximum expansion/updatability as possible.

Then, we minimize what we cannot change to expand or update later. With these CPUs, we can't really change them out as they are not in a socket like they had been in ages past. A mezzanine helps a little with that, but you'll see me wanting that mezzanine module to also contain the southbridge chip and SODIMM sockets, so all of that changes out together. That isn't necessarily bad. And it does let the motherboard to become simpler, and hopefully cheaper, as a module carrier that mostly only connects ports and slots to the mezzanine card.

And toward my wish for a laptop, which is a somewhat different discussion with different requirements and goals as a desktop, I would of course want as many as-modern-as-possisble features built-in as possible, since we really cannot add them later, but still have Expresscard, mini-PCIe/Msata or M.2/ngff slots as can be reasonably provided for some future changes, and get Olegil shaking his finger at me again.

_________________
All glory to the Hypnotoad!

@OldFart

Quote:

With regard to the topic of layering I take it that the part around the CPU and memory modules is the toughest one, but is a PCIe bus equally complex requiring equally complex layering or can it be made to function with substantially less? These thoughts keep me busy (in a manner op speaking, that is).

PCB layering is mostly related to how many pins there are, how they are organized at the chip/PCB interface, and where they need to go, and then a little by what they are.

What the yare (PCIe, SATA, DDR, USB, etc) will define any impedance requirements to work well, and that impedance requirement, along eith your board thickness and space between layers, defines the width of each trace for that connection. The wider the trace needs to be, the more space it needs to route and thus more likely to increase number of layers to accomplish that for all pins.

Wider busses, such as DDR, classic PCI, etc. are a little harder, as you need things to all arrive at the same time. The more wires that need to match up, the harder that is to do. SERDES such as in PCIe, is a method to make life easier on the PCB designer. Each lane has I think 4 major signals, and those are grouped in pairs. Each signal pair must match within itself, but one pair does not necessarily need to match another pair. So the RX differential pair mus tmatch itself, and the TX pair must match itself, but the RX pair can be different than the TX pair. And then one lane can be different from another lane. So PCB designer must make each diff pair work well, bu tbeyond that things are less difficult than in a high-speed parallel bus. So PCIe x16 is probably easier to do than PCI-X parallel bus at 133 or 166MHz, or PCIe x16 is easier than a big DDR bus. Or doing several PCIe slots is easier than doing two DDR busses.

_________________
All glory to the Hypnotoad!

@billt

Nice to see you guys still bantering this one about.
So far, it looks like three tiers.

First, a T1022 orT1042 based board with no Southbridge.
Second, a T1022, T1042, or T2081 based board with a Southbridge.
Or, a third option, with a T280 based board with or without SB.

And on the first two boards a mezzanine layout would allow upgradability.

@iggy

I'm all in favour of upgradeability, but in this case I personally feel the cost and hassle is not worth the benefit. Maybe a simple BGA socket adapter can be used.

But you generally don't want to route high speed PCIe and DDR3 through extra connectors, so I would really advice against it (not that I'm getting much attention here )

_________________
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.

@olegil

Quote:

olegil wrote: @iggy I'm all in favour of upgradeability, but in this case I personally feel the cost and hassle is not worth the benefit. Maybe a simple BGA socket adapter can be used. But you generally don't want to route high speed PCIe and DDR3 through extra connectors, so I would really advice against it (not that I'm getting much attention here )

I think that this first motherboard should be really minimal to keep low prices in prototyping, manufacturing and mainly end-user price.

If we will succed in creating a well balanced and good funcioning motherboard, and if it will be a best seller (in Amiga market numbers), then we could think to realize another model, more upgradable than the first one.

We must be patient and fighting with the weapons we have available.

Last edited by Raffaele on 12-Nov-2015 at 12:46 PM.

_________________
"When the Amiga came out, everyone [at Apple] was scared as hell." (J.L. Gassée, former CEO of Apple France and chief of devs of Mac II-fx, interviewed by Amazing Computing, Nov 1996).

@olegil

Quote:

But you generally don't want to route high speed PCIe and DDR3 through extra connectors

If one were to choose to go mezzanine, then one would choose to put DDR3 memory on the mezzanine module along with the CPU. No extra connector there, either SMT soldered or in a DIMM/SODIM socket it was designed for. DDRx would not go through the Mezzanine connector to the carrier motherboard.

Various such mezzanine modules already do put PCIe through such connectors form CPU down to the carrier then to the slot or chip that it connects with. So do SATA, USB3.0, SGMII, and various others. Nothing new or special about that. Plus we have PCIe cables that take a slot to a card to a cable to a card to another slot to whatever is there, and also have cable-bsaed slot "riser" adapters for extending to a mounting or out to benchtop debug access etc, and those work.

_________________
All glory to the Hypnotoad!

@billt

Well, some of the CPUs previously mentioned support DDR3/DDR3L (T2) while the others (T1) support DDR3L/DDR4.
So the commonality is DDR3L, which is fine by me since it will generate less heat.
As time goes by, laptop memory prices (especially second hand, thanks to laptop failure) tend to fall to below desktop prices, and it takes up less real estate.
But I still rather like desktop memory.
And I'm not sure what makes more sense, a mezzanine board or a BGA to pin adapter.
But placing the memory on the CPU board seems to complicate it.

My problem in a nut shell?

I can see an argument for each approach to this project.

@iggy

Well, then, for the problem at hand, the subject of this thread has "cheap" in it. So ponder what would be the most cost-efficient thing.

Messanine and BGA pin-sockets cost money. Mezzanine could be reused in other things. Does that get you anything... If no Mezzanine card, then go desktop memory. If Com-Express or similar them probably go SODIMM per that standard.

If you do a prototype, where do you go from there. Do you sell the simplest board, which is essentially a CPU with only a few connectors attached to it and as little else as possible? Does essentially that plus a mezzanine connector give you any additional benefit? If it's too simple, then a mezzanine may not give you any benefit over an SMT assembly option between pin-compatible CPUs. Mezzanine basically takes very different CPUs and allows you to plug them into same carrier. Then, does it make sense to plug a more complicated CPU into a too-simple carrier made for a simpler CPU?


If Mezzanine is more expensive now, does it save you anything in the long run if you do additional modules, saving you the carrier motherboard redos? Hard to say as it's possible that the smaller module pushes you to more PCB layers, but it's smaller so you can get more modules onto a PCB manufacture panel than a monolithic motherboard.

Last edited by billt on 13-Nov-2015 at 10:17 PM.

_________________
All glory to the Hypnotoad!

@iggy

I didn't even think there WERE any desktop DDR3L memory. But apparently that is not true. Anyway, it's not always about what you would like to have, but what you have space for. DIMM takes more space -> rest of design becomes more cramped -> need more layers.

_________________
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.