@Zylesea

Quote:

The e500v2 I'd skip completely, no binary compability is making the situation unnecessarily uncomfortable. With the logic of recompiling each progarm and picking the appropriate binary

Well I doubt the JIT compiler has lot of FPU code due the lack of it in most 680x0 chips, so if any assembler has to be changed its going to be only tiny bit, I think, the rest is C code and should fix it self whit a recompile.

Quote:

you could better make the big jump to x86 or ARM then (not that much more hassle included).

But if you go ARM you need a new JIT compiler, and on x86 its almost unthinkable to implement it due to endian issues.

Last edited by NutsAboutAmiga on 20-Nov-2012 at 11:01 AM.

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

@Crumb

"Most of amiga users I know are happy with their 20 years computers, I don't think warranties are their main cause of concern when talking about NG miggies."

Yes warranty is not a consern for all users. But for some it is.
And for professional use it's pretty mandatory.

"Pegasos2 is faster and was released in 2005."

I meant new HW is faster elsewhere. Like SAM460 is faster than Peg2 in everything except in raw CPU power. And I believe it's has faster interfaces than most y2005 MacHW. You can not use modern y2012 GPU on y2005 Mac in any simple way (except on the last liquid cooled G5 model).

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

@KimmoK

In my opinion if the support for this different FPU is hard to achieve (unless the kernel guy proves that is an easy job) the Amiga Hardware resellers should really consider to talk with the producers of these little ppc board, saying: we really want to focus on the board you produced, but we need a little hardware refresh for compatibility with our OS. Even add an extra 100$, the cost will be for sure under the 250€, the lower price we've seen till now.

_________________
retired

@NutsAboutAmiga

It's sad how no discussions can be about PPC without someone dragging in the old fights again.

I must say I had completely missed the P2040 (and 41) being e500mc, but of course it makes sense as the e500v2 only scales to dual-core systems because of the shared-bus design.

Now, this is an interesting part in that it would allow us to run 2 copies of AmigaOS (one with GPU and one for debug or render-farm purposes) plus 1 copy of Linux with dual-core CPU for those apps we just haven't gotten yet. As well as render farm purposes obviously .

From what I gather, Freescale has a hypervisor you can just start up and instruct which OS goes on which virtual machine with which peripherals. So not much work needed for that.

Is this interesting? The chip costs 100USD which means it's a 400USD end user price, plus taxes. But it's 1.2 GHz, quad core at that price. SMP might be cool and all, but with the hypervisor letting us run extra copies of the OS essentially for free, I don't see the current lack of SMP as a huge obstacle. And as I mentioned, if AOS4 gets a good dual-core solution we can still use this chip as a dual-dual with 2 virtual machines of 2 cores each.

Only one GPU in total on the system, as we're limiting the number of lanes severly, but I'm really not sure that's a huge problem either.

Does anyone think this sounds interesting?

Btw: Looking at the ball-out (there's always a source even if Freescale demands NDA ) I get the feeling that with micro-vias in the current stackup will lead to no more than 6 layers being used:

blind layer 1-2
buried layer 2-3
buried 3-4
blind 3-6 (large regular via for GND and PSU)

With the current layers:
1: GND
2: PSU
3: Routing
4: GND
5: Routing
6: PSU

(those last two are unorthodox but it means fewer so called stitching capacitors translating the reference of high-speed lanes from GND to PSU. Much easier if both routing layers are next to the GND layer). Besides, the back of the board is used for placing decoupling capacitors anyway.

With lasers, it's possible to drill to the glass-reinforced epoxy until copper is exposed (the laser used will not make holes in copper), so the whole process is fairly simple. Start with layers 4 and 5, laminate on 3 and 6, drill from 3 to 4. laminate 2 on top of 3 and drill again. Laminate 1 on top of 2 for the last run.

The technique is called HDI (high density interconnect) and anyone even remotely interested in electronics should read about it.

Last edited by olegil on 20-Nov-2012 at 11:15 AM.

_________________
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:

Now, this is an interesting part in that it would allow us to run 2 copies of AmigaOS (one with GPU and one for debug or render-farm purposes) plus 1 copy of Linux with dual-core CPU for those apps we just haven't gotten yet. As well as render farm purposes obviously . From what I gather, Freescale has a hypervisor you can just start up and instruct which OS goes on which virtual machine with which peripherals. So not much work needed for that.

The thing about running 4 programs at the same time, is that you can't have them write to same memory at the same time, nor can you have 3 programs reading some memory while an other one writing to it.

This the challenge is to serialize interaction between the CPU cores when its needed, now you do that is up to the kernel designer, we have a number of function calls that does it, permit(), forbid(), semaphore and mutxes.

semaphore and mutxes are smarter because they understand the difference between writing and reading, they do not protect the memory, but instead they prevent code from being executed that as intent to write or read from sheared memory, and also they return true or false allowing the program to do something else while it waits.

Permit() only gives one program the right to execute at given time, until premit() is called, it also prevents any program doing anything, I'm not shore it effects interrupts or not, but for interrupts you have enable() and disable()

in Linux word there is some thing called spin lock but its considered evil so maybe we should just forget about that, it is to prevent more then one core to run at the same time while doing something.

http://stackoverflow.com/questions/195853/spinlock-versus-semaphore

another issue is the CPU cache what you do when releasing the lock after a write.

The program does not know about low levels, its just ask the OS to do the right stuff, some variation the CPU design should not be a big problem.

Last edited by NutsAboutAmiga on 20-Nov-2012 at 11:45 AM.
Last edited by NutsAboutAmiga on 20-Nov-2012 at 11:42 AM.
Last edited by NutsAboutAmiga on 20-Nov-2012 at 11:38 AM.

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

@olegil

Just noticed I had done the math on 1.0mm ball pitch while the package has 0.8mm ball pitch, which sort of explains why I ended up not even using the second routing layer except in one or two places.

With 0.8mm pitch I would still most likely be able to do things with two routing layers, but it comes a bit close for the USB and DDR3 ports.

Maths:
for 1mm pitch, 0.1mm width and space, I get 9 wires on layer 3 and 6 on layer 5. So the micro-vias "wastes" the board space of one wire while the normal via from layer 3 to 6 wastes the equivalent of 4 wires.

For 0.8mm pitch with same width and space, start with 8 but reducing by 1 and 4 makes the end result 7 and 4 respectively, which is just a total of 11. I counted worst case 11, so that fit nicely. Wonder if they thought about that? The T4240 is very silly in that regard as it seems optimized for 1990's style circuit board techniques

_________________
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.

@NutsAboutAmiga

Arguing with you is a waste of breath. I am saying we can run 2 COPIES of AmigaOS AND 1 copy of SMP Linux at the same time AND THEY WON'T KNOW ABOUT EACH OTHER.

Try to read what I write here. I'm saying this can be done WITHOUT A REWRITE OF THE OS, as the hypervisor is supplied by freescale.

Actually, we should be doing this in Norwegian. You can call me, I'm in the Gule Sider

Last edited by olegil on 20-Nov-2012 at 11:57 AM.

_________________
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.

Quote:

Arguing with you is a waste of breath. I am saying we can run 2 COPIES of AmigaOS AND 1 copy of SMP Linux at the same time AND THEY WON'T KNOW ABOUT EACH OTHER. Try to read what I write here. I'm saying this can be done WITHOUT A REWRITE OF THE OS, as the hypervisor is supplied by freescale.

Somehow this text reminds my to Atheist2

_________________
AmigaONE. Haha. Just because you can put label on it does not make it Amiga.

I borrowed this comments from here (#27 & #28):
http://amigaworld.net/modules/newbb/viewtopic.php?topic_id=38873&forum=2&start=20&order=0

@olegil

Even if you use the Hypervisor you need to deal whit video, mouse, keyboard, Ethernet and so on.
You will need to implement some kind emulation layer taking care of visualizing hardware or assigning hardware in hypervisor the Root partition.

Anyway running AmigaOS two times is the same as running AmigaOS on two different machines, you can only really benefit from it unless you distribute work, like they do on super computers.

Not saying Hypervisor is not interesting or is not usable, its where usable for emulators because it removes translation obstetrical thanks to IOMMU.

Just think about rewriting the EUAE 680x0 CPU core so it does need to deal whit memory translation in EUAE and Basilisk II, and many of emulators we use, but it also enable running MacOSX and LinuxPPC in a virtual environment easily.

But more exiting at least for me, is support for multicore support in AmigaOS4, to get tasks to run on individual cores, this can really improve speed in many applications.

Not so interested in Hypervsior as emulator, but what can be done whit in AmigaOS to provide multicore support, and I think its more the possible to get AmigaOS to notice more then one core and to balance work load, so that we may decompress video and audio on different cores whit improved task scheduler.

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

@Virtualization

freescale virtualization infopage
"pick the right virtualization..."

Virtualization has been one "option" for multicore usage and some solutions require server code to be implemented before you can launch OS on top of virtualized HW.
I think PA6T might need Amigans to do the virtualization SW that would use the virtualization support of the HW ??? It would seem freescale have more ready made things to offer, for QORIQ chips.

Last edited by KimmoK on 20-Nov-2012 at 01:05 PM.

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

@NutsAboutAmiga

Go back a few posts and read my original idea.

One AmigaOS4 runs the graphics card and the mouse, the other (and the Linux) are controlled through VNC and/or X windowing system.

You really would say no to having an extra system "for free" for running suspect code?

_________________
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

Also, you can (again, using the hypervisor to split the machine up) set aside seperate cores for lightweight OS'es which basically take processes and data like "decode this frame of video using the following program". This should be as easy as pie to create. But someone would need to make a standardized framework where an app can request extra processing power.

All of this is possible with many cores, and there is a chip which has them for a low price. Have some vision, dude!

_________________
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

judging from this site, the interest was exactly zero. as usual. ok.

_________________
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:

Besides, I said P4080, that's the e500mc

I was referring to the part where you said "Hyperion should Support the e500v2 so we can make cheap hardware for entry-level."

Also you can't expect every developer to still be around to recompile their code, and to release multiple binaries of all their software all the time. You'd end up with only a limited number of software usable from the already very limited selection we have. Forget e500v2.

@COBRA

I already forgot it at least one page ago. Now I'm waiting for anyone to bother reading my ideas on running multiple OSes on a P2040 (or 41, if anyone wants to spend the extra cash on the higher speed). The board would be exceptionally simple, to keep costs down.

Freescales hypervisor would let us partition the CPU and peripherals into multiple virtual machines, only one of which would most likely have USB/SATA/PCIe, the others being connected by VNC and/or the X window protocol. Uses: safe testing of software you or others have compiled, rendering "farm" (easy to implement gcc compiling and blender frame rendering through distcc etc), running Linux software on the OS4 screens etc. If or when AOS4 starts supporting SMP (or even AMP) we can allocate more cores to that, removing cores from something else.

Once a 2040/2041 system works, I would move on to 3041/4040/4080/5010/5020 which are all pin compatible so it's just a question of what one wants to pay for.

All of these have the classic FPU so there need not be ANY worries about incompatibilities. AFAIK.

Please, anyone. Input?

_________________
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:

Please, anyone. Input?

The 2040 sounds cool, if you could build a cheap board around it. You should now change your signature.

_________________

afair, we just have managed to avoid to lose ressources on ultimate_ppc expansion for big box amigas, based on same incomtpatible concept. i guess better nothing than a thing that you cannot make usage of. what already has been said, even as small software base as this of os4, cannot be expected to get recompiled for every other cpu in sight, especially if its not open source. anyway, in advance to any approach like that i would secure the support of os4 deliverers.

@Troels

Well, I already have my 5121e board, so maybe adding to it would be a good idea rather than replacing it.

Unfortunately, me building a board doesn't mean Hyperion supporting it, which is sort of the big elephant in the room.

_________________
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

While digging info...
If you spot HW details for T series & developer board releases for those, please info also on Freescale PPC thread:
http://amigaworld.net/modules/newbb/viewtopic.php?mode=viewtopic&topic_id=35140&forum=33&start=80&viewmode=flat&order=0

(to my understanding no reference manual exist yet for T/AMP series)

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

@olegil

I think it sounds great . . . So, is it finished yet? LOL

If it makes it running a next-gen Amiga OS, I'll take 2 . . .

_________________
Avatar babe - Monica Bellucci.