Quote:

True but ARM doesn't try to compete on the desktop - at the moment.
They're moving up very rapidly though.
Currently, the highest end stuff is based on phone chips but that's changing as tablet chips become available. It'll get really interesting when server chips appear.

Quote:

Since when has cheap, available, new hardware been a disadvantage?

Quote:

Actually, at 6 billion a year they're pretty hard to avoid.

They're not exactly well represented in the desktop space (in the west) but there are some and there's more due at the end of the year.

If you just want a cheap ($149 - $199) dev board there's a few here:
http://www.linaro.org/low-cost-development-boards

Quote:

Can't argue with that.

Quote:

Good. Neither am I.

Quote:

You base this on what exactly?

Quote:

Quote:



Of course you can't be referring to me because this is one area where I can't be described as blind...

>> There's no ARM machine that can compete with a QuadG5
>> machine in price/performance right now.

> That'll change next year. The year after, the opposite will be true.

Don't be so sure about that. While it's true that ARM performance will increase over time and actually make a jump next year with the advent of Cortex-A15 based chips, leading also to an improvement in price-performance ratio, used quad-G5 Macs are becoming even cheaper, improving their price-performance ratio as well, all the time. Currently, the average price that a (fully functional) quad-G5 Mac is sold for on eBay Germany is about 530 EUR, whereas 16 months ago it was still 910 EUR, which is a reduction by 42%. If that trend continues (and I guess it does) new machines with future ARM chips will definitely have a rough ride beating used quad-G5 Macs in terms of price-performance ratio for some time to come.

Addendum:

> Page 12 of FTF11_NET_F1176 document mentions "Improved load /store
> to ease bit alignment" and "New instructions for video analytics".

Found another interesting comment on the new AltiVec in the QorIQ AMP:

"An important note regarding AltiVec is that some users have previously found the ISA difficult to use. AMP marks its relaunch in a form that allows a far easier definition of parameters."
http://www.techdesignforums.com/embedded/embedded-topics/freescale%E2%80%99s-qoriq-amp-series-illustrates-growing-demand-on-the-control-and-data-plane/

> The PA6T CPU paves the way for the introduction of SMP with AmigaOS4.x.

OS4 core developer tonyw:

"I would expect OS5 will be 64-bit and SMP."
http://amigaworld.net/modules/newbb/viewtopic.php?topic_id=33968&forum=33&start=20#622636

Seems SMP got postponed to "OS5".

Update:

> First solid information on POWER7 preceded its official release by 1.5 years.
> Going by that, 1 year from now plus 1.5 more years would mean early 2014
> for POWER8 release. But as we know that POWER8 is scheduled for 2013
> (and rather early than late that year), solid information may be imminent ;-)

Found three interesting files:

http://www.csiltd.co.uk/PDFS/ISD/Itanium%20-%20Reasons%20to%20Act%20Now.pdf (May 2011, page 10)
https://www-950.ibm.com/events/wwe/grp/grp006.nsf/vLookupPDFs/POWER7/$file/POWER7.pdf (June 2011, page 6)
http://maben.homeip.net/static/computers/aix/brochure/IBM%20Roadmap%20cf%20sparc%20itanium%20201106.ppt (June 2011, page 2)

They list as a POWER8 feature among others something called "Accelerators". And what is the real surprise for me here, those "Accelerators" are listed for the POWER7+ as well.


Edit:
Found a PDF where the POWER8's "Accelerators" became "Imbedded accelerators" (and also "Dynamic TurboCore" was added to the chip's features):

ftp://ftp.software.ibm.com/la/documents/stg/z/itso-mx/executive_briefing/power_systems_soluciones_para_un_planeta_mas_inteligente_freddy_alves.pdf (October 2011, page 27)

[ Edited by Andreas_Wolf 21.03.2012 - 10:21 ]

Power8
Accelerators+?

> Power8
> Accelerators+?

Yes, that's exactly what I'm referring to :-)

Good find!

Look up the PowerEN and that might have some clues. Probably special blocks for crypto, network etc.

Another thing that IBM actually said in the past is Cell will merge into things like POWER. So don't be surprised if SPEs turn up at some point.

What I thought interesting was the mention of a very large cache, as if 32MB wasn't big enough already!

That said the IBM mainframes can have 196MB cache...

Of course being a "BAF" I had to mention this.

Netbooks were an early skirmish. Now things are about to get really interesting...

> Probably special blocks for crypto, network etc.

I hope not, but rather this:

> don't be surprised if SPEs turn up at some point.

...as that's exactly what Jim and I are talking about here in this thread regarding POWER8 all along. So no surprise from us I'm afraid. Seems you lost track a bit ;-) To see where this sub-discussion started in this thread three weeks ago refer to:

https://morph.zone/modules/newbb_plus/viewtopic.php?forum=3&topic_id=7183&start=294

So I think that the mention of "Accelerators" in the PDF files I linked to might refer to SPEs rather than crypto or network accelerators.

Quote:

You expect people to read the entire thread before posting???

Think of it as a drive by posting.

> You expect people to read the entire thread before posting???

No, but I expect people who took part in a discussion before to not act as if it didn't happen. At least I wouldn't know why you say I shouldn't "be surprised if SPEs turn up at some point" in IBM POWER when three weeks ago you, Jim and me discussed exactly that and you even got to know that Jim and me had discussed this long before even.

> Think of it as a drive by posting.

Hard for me considering the history of this thread :-P

Was that only 3 weeks ago. Wow.

I couldn't tell you what I posted tonight never mind 3 weeks ago!

Really - I visit a lot of different sites...

> I couldn't tell you what I posted tonight never mind 3 weeks ago!

So my assessment that you lost track was correct obviously. I'm looking forward to you telling us to not be surprised if IBM POWER will come with SPEs and acting as if this discussion never happened, again, and again, and again... :-P

Update:

>> The PA6T CPU paves the way for the introduction of SMP with AmigaOS4.x.

> OS4 core developer tonyw:
> "I would expect OS5 will be 64-bit and SMP."
> http://amigaworld.net/modules/newbb/viewtopic.php?topic_id=33968&forum=33&start=20#622636
>
> Seems SMP got postponed to "OS5".

Hyperionmp says:

"The way in which it will be implemented however is already clearly defined and was subject to peer review by other developers. Obstacles to an efficient implementation were removed (e.g. the use of Forbid) and replaced in many OS components over the years (e.g. DOS).The foundation for SMP support was put in place, a clear picture exists what needs to be done to accomplish it and how. I'm willing to take a bet that it won't take 2 years ;)"
http://amigaworld.net/modules/newbb/viewtopic.php?topic_id=34171&forum=33&start=40#627520

>> https://morph.zone/modules/newbb_plus/viewtopic.php?topic_id=6570&forum=3

> It's not quite the same. As I understand it, on ARM the problem would be
> intermixing big endian code (68K, PPC) with little endian code (ARM). In
> both cases the data is all big-endian. On IA32 everything is little-endian,
> code and data. What I had suggested is essentially 2 A-Boxes. One includes
> a PPC emulator to run PPC and 68K code. The second A-Box (B-Box?)
> would only run natively compiled ARM code. No intermixing.

Zylesea has now put his thoughts on the matter into a nice article:

http://via.i-networx.de/q86.htm

Quote:

Interesting, a couple of points though:

Quote:

Full Memory protection these days is really best described as "basic" memory protection, desktop OSs these days are starting to run things in sandboxes. I can see eventually everything running in a virtual machines.

What he means by cost? There is a cost in terms of API changes, but a well designed memory protection system wont be noticeable to the use performance wise.


Quote:

For backwards compatibility you use an emulator. Anything else, modify as necessary and recompile. It works for AROS, why shouldn't it work for MorphOS?

> What he means by cost? [...] a well designed memory protection system wont be
> noticeable to the use performance wise.

I think that copying the contents of a memory area is always slower than just passing a pointer.

>> For MorphOS to keep backward compability a true bigendian mode is required.

> For backwards compatibility you use an emulator.

Yes, that's what's done in MorphOS (and OS4). And for such transparent emulation you need a matching byte order, as Zylesea correctly points out.

minator,
Quote:
Andreas said the things about MP already. But as I said in the article,I think MP is debatable. I don't say MP is evil, but I think the equation MP=good is too easy, too. Depends a lot on the way of implementation at least. But I am not an OS designer, it's just about what I read and the conclusions I drew from that. I haven't exercised it myself.

Quote:

Every new feature has a cost Vs benefit. In this case all mainstream desktop OSs, most smartphones and pretty much all tablets have MP, it's obviously not that much of a cost. MP needs to be set up by the OS it'll mostly be done by the hardware afterwards. It also shouldn't involve much copying of data, you can do that by remapping pages.

It's not exactly a new problem so there's plenty of examples out there to look at. I doubt the MorphOS developers will have much problem figuring it out

> all mainstream desktop OSs, most smartphones and pretty much all tablets
> have MP, it's obviously not that much of a cost.

From my own experience with MorphOS vs. Linux on a 600 MHz G3 and MorphOS vs. Mac OS X on a 1500 MHz G4 I can say that there's a vast difference in general OS responsiveness. In my humble layman's opinion this should in large part be attributed to the much lower overhead of the message passing concept that MorphOS' ABox is built upon. I'm not sure I would like MorphOS the way I do if it felt as little responsive on my gear as the mainstream desktop OSs do ;-)

> I doubt the MorphOS developers will have much problem figuring it out :-)

"Will", or rather "would"? ;-)

Again Andreas wrote much of the things I'd also say. However it is also true that there are OSes out there that are actually *very* snappy and lightweighted and come with MP. Most prominently and from my 1st hand experience qnx neutrino springs to my mind. That was impressivly fast and rock solid. So I know MP is possible on a lightweight and fast OS. But I also think it is often overrated. I mean most crashy apps are caught already and don't shot down the OS - not much different as if there were MP. Only difference: A program *can* shot the OS dead. But so can apps on OSes that offer MP (I quite well remember the issues I had with a program for data acquIsition I wrote for windows 2k/XP that locked the machine dead again and again - calling a faulty dma routine from driver was the issue... Okay on a micro kernel OS like QNX where all drivers are separated tasks this may be a different story.
Anyway, MP is definitely not a bad thing per se, but many OSes with MP are dog slow (for many reasons) and still not 100% stable against full locks.
I rather see the bigger demand for a change in the field of resource tracking, removing dead tasks and freeing all resources would be big plus. But I imagine work on resource tracking and MP are often going hand in hand anyway. But I emphasize it again MP is at least debateable - it's not the holy grail nor the devil.

Quote:
And this is coming from same people, who claimed AmigaOS4 would be done in a few months, and would include some features it still doesn't have (many pointed out to be impossible back then (auto-expanding stack, anyone?), but dismissed by Hyperion as FUD)

So at least I have some reservations whenever it comes to deadlines from Hyperion

Also, in no point does he state, it will NOT be postponed to OS5.

After all, it's only a matter of what do you want to call that particular release (personally, I can't see the big difference between, say, OS4.5 and OS5, they are both likely paid updates anyway)

> in no point does he state, it will NOT be postponed to OS5.

Yeah, it's quite possible that it's the release of "OS5" that is supposed to not "take 2 years".