@ackcontrls

Wow, the high end machine would then be a dual-core, 64 bit machine running at 2GHz, with a DDR2 memory controller, PCI express (supporting x16 - full bandwidth for DX10 capable gfx cards!) and gigabit ethernet. All that consumes just 25W of power. A board based on this chip would catapult amiga back into the running, putting it on the same ballpark as current PCs and Apple Intel boxes. All we'd need is for OS4 to be ported to 64 bit and we'd even be on the same level as OSX Tiger. As it is, it would already beat the Apple and Windows systems on power consumption: 25W for the PASemi chip vs 65W for a core 2 duo (read: much quieter computer). I'm dying to see what benchmarks would show for one of these things vs an Intel chip.

If this is real and not just Adam yanking our collective chains, it makes it extremely important for Amiga Inc and Hyperion to settle their differences in such a way that OS4 doesn't get gutted.

@Insanity

Ok, I've managed to find some more recent information about SMP in Amiga OS4.

4 March 2005, EntilZha said:
Quote:

No, there's no support for multiprocessing yet. There are several unresolved issues regarding the semantics of multiprocessing. For example: What does Forbid do with more than one CPU ? Forbid is said to forbid task switching, but that would mean two CPUs would continue to run concurrently. However, it's most of the time used to achieve single threading, i.e. the task calling Forbid will assume it's the only entity running. These are things that will have to be cleared up, and this will happen in 4.1.

So the biggest problem is the forbid()/permit() semantics. Leaving it in there could be a real performance killer, particularly given the amount of forbid()/permit() abuse some software have. Changing the behaviour of forbid()/permit() will cause old software to break unless they're run in a sandbox of some sort.

Then 2 April 2005 Rogue said:
Quote:

Even a non-SMP aware CPU could make use of it by running two programs at once. SMP is definitely high up on the todo list. In fact, the new memory system is already geared towards multi-CPU support.

So I'm guessing that partial SMP support meanse something like:
- The necessary functionality, such as spinlocks, has been implemented into ExecSG.
- The new memory system is geared toward multi-CPU (as stated above)
- There is no API available to use SMP, and hence, SMP is disabled

It would be interesting to hear if there has been any progress since then.

Hans

Last edited by Hans on 05-May-2007 at 07:30 PM.

_________________
Join the Kea Campus - upgrade your skills; support my work; enjoy the Amiga corner.
https://keasigmadelta.com/ - see more of my work

@Hans

When OS4 gets ported to 64 bit, all old software is going to break anyway, because there are many, many places in AmigaOS where there is an assumption that the size of a pointer is 4 bytes, whereas with a 64 bit address space, a pointer is 8 bytes wide. There is no solution for that except some sort of sandbox that emulates a 32 bit environment.

The Forbid/Permit issue is a brilliant example of how one bad decision you make today can totally screw you over years from now. Carl Sassenrath must have known Forbid/Permit would never work on an multi-CPU system, but he probably figured "what are the odds of a multi-proc Lorraine anyway" and went ahead and did it.

Last edited by CodeSmith on 05-May-2007 at 09:41 PM.
Last edited by CodeSmith on 05-May-2007 at 09:40 PM.

@CodeSmith

I think that AmigaOS4 should have been all-new-spangly (but based upon AmigaOS's design, architecture and principles, of course) with OS3 running in a sandbox from the beginning. All cruft and legacy could have been dumped. Hell, if AmigaOS3 wasn't sourceable, AROS could have been used for providing libraries for legacy applications.

Of course, this is with the benefit of hindsight, as it took 7 years to get to AmigaOS4.0 (albeit a very refined AmigaOS4.0, far better than originally envisaged). A lot of work done for backwards compatibility could have been avoided, although work for a sandbox would have been needed (although emulation of the 68k was done, and let's not forget that UAE and other applications could have been of use too).

I expect that this is what AmigaOS5.0 will do, in 2012, once the court cases are over.

@Hattig

I seem to recall the very original plan (for the old Escena bridge boards) to have been exactly that, but it was discovered to be too slow. Then again, at the time Petunia didn't exist and the fastest PPC CPUs were 7410 G4s running at 500MHz. With the more powerful CPUs available today, that idea could be revisited.

[edit]Hmm, just found out that the PWRficient chip has support for virtualization. That should make creating a 32 bit sandbox a lot more efficient.

Last edited by CodeSmith on 05-May-2007 at 09:41 PM.

@ironfist

*Falls over*

Well, unfortunately there is not much from PA-Semi available so far.
They are sending out the first samples and hope to start
the production sometime in Q4/2007.
That is the 2 GHz dual-core.

Two weeks ago I received a mail from one of our distributors at work
about PA-Semi and asked for more information.
And without having any details here at hand I only can tell
you that my reaction was that at the targeted price
I thought this will be a fail - and available way too late.

And if not designed by a top-notch company I do not believe
it can be taken off ground.
This requires a PCB with about twelf layers,
nothing you come up with overnight.
This monster has 1156 solder-balls in 1mm pitch.

And you can't even "just" take the eva-board and modify it.
The "Electra" has 1 IDE, 2+1 USB2, 1 PCIe x16, 1 PCIe x4, 1 PCIe x1,
1 PCI, 2 GB Ethernet, 4 DIMM sockets, 2 serial.
To make this a mainstream mainboard it would need sound, 4+ SATA-2, 4+ USB2.

I can only hope someone has pockets deep enough to make this happen
but for now I am afraid that it won't.

@Shadowolf

Sure you can. Add a PCIe SATA card and a PCI USB card, you're all set. Not saying it's ideal, but it is possible.

@CodeSmith

This would be a bad joke for a supposed to be high-end system.

@Shadowolf

Quote:

I can only hope someone has pockets deep enough to make this happen

Maybe Amiga Inc or ACK or both have managed to get subsidy or funding from the semiconductor manufacturer. After all, with the falling interest in PPC Linux and Apple saying bye bye to PPC, the amiga community is the last hope for these desktop orientated processors.

@Colin_Camper

Only that these processors are *not* desktop oriented at all.

@Shadowolf

On Monday you will see the actual specification for the board. Most designs lately with high ball count need at least 8 layers, some 10....12 isn't a stretch, especially in the telecommunications industry. Also, the ball count at 1156 isn't much higher than the ball count of the Tsi108/109/110, which is only a host bridge. For all the functionality in this chip, the overal ball count is reduced, along with total component count.

It must be the North American "can do" attitude that enables us to accomplish goals that others believe impossible.

Adam

@Shadowolf

Quote:

Shadowolf wrote: @Colin_Camper Only that these processors are *not* desktop oriented at all.

2 GHZ, dual core (and up), VMXunit, 2 MB cache, PCIe controler......What is oriented for? Microwaves oriented? Wake up, please

_________________
No PowerPC, No Fun
Make Amiga Great Again

@A3000T

Quote:

Bring on the High-End board. I'll buy it if the OS4 development team will support it.

I hardly dare say this but at the back of my mind I've been wondering if Pentti Kouri has a notion that OS4 devs headed by Friedans plus Indian programmer stable might make for accelerated progress sans Hyperion - Amiga Inc conflicts.

Noel

@Shadowolf

As I said, it's not ideal...

If the board didn't support PCI express (like a bunch of existing designs that only support PCI or PCI-X), then it would be impossible. Remember though, Adam says that Amiga Inc want to sell complete boxed systems. There are enough slots in the board you described to add the few remaining pieces to make it into a home computer; no-one said anything about the design being open. Apple has been working like this for ever, and most people don't seem to mind too much (in fact, it allows their software people to concentrate on making good drivers for a small set of cards, rather than mediocre drivers for a large set of cards). Given the time constraints (Amiga Inc saying "this summer" and PASemi saying "Q4"), I suspect Amiga Inc will just cut a deal with PASemi for a few thousand demo boards at a reduced price, get some addon cards for the missing functionality, add a box and PSU and sell it as the Ack-One. Some might call it cheating, I'd call it making the deadline for once

@ackcontrls

Quote:

It must be the North American "can do" attitude that enables us to accomplish goals that others believe impossible.

This is not something what can be done solely on motivation alone.
This requires a lot of money to pull thru.

The small scale of the operation is what I see as the main problem.
If you are ASUS or GigaByte or whoever and need to throw a new
Pentium-board at the market you "just" spent 100+k relying on
the expertise of your team that went thru this a couple of times.

But if the target is only in the range of 1000 boards
you have a problem unless you win the lottery.

Still, I would be pleased to see it happen.

cant you just wait to monday when the specs are released. this is just mumblejumble ;)

@Fransexy

My hotdogs were cooked with altivec. Mmmm

@Kicko

Quote:

cant you just wait to monday when the specs are released.

No

@Fransexy

Quote:

2 GHZ, dual core (and up), VMXunit, 2 MB cache, PCIe controler......What is oriented for? Microwaves oriented? Wake up, please

Applications: "embedded networking", "telekom", "military", "storage" and "pervasive computing"

http://www.freescale.com/files/32bit/doc/fact_sheet/MPC8641DDLCRFS.pdf

PA-Semi mentions "Enterprise-Routers" and "Modular Storage Platforms"
and yes, "Supercomputing"

Desktop is something different.

Still I would love to put something like this in a tower under my desk.