> I still can't get data out of Applied Micro on the APM 83290 SoC [...]. Freescale has always
> been pretty open with me (even providing me the MPC8641D samples [...]). I'm going to
> check out the e5500. At 2.5 GHz, even without Altivec, it might be a match for the PA6T.

Just for the purpose of overview I compiled a list of the DMIPS/MHz per core figures for various PPCs in ascending order (AltiVec capability is generally ignored, obviously):

e300: 1.9...2.0 (depending on version)
PPC440: 2.0
PPC460: 2.0
Titan: 2.0 (presumably PPC450 based)
PA6T: 2.2
PPC750: 2.3
e600: 2.3
PPC470: 2.3 or 2.5 (varying with information source)*
e500mc: 2.5
PPC970: 2.9
e5500: 3.0

* <edit>IBM says the PPC476FP has a figure of 2.7</edit>

What seems really weird is the rather low value for the PA6T (which was provided by P.A.Semi themselves btw), especially compared to the PPC970, which I thought it should have been on par with. Or maybe the PPC970 figure is too high (i.e. with AltiVec)?

> its available from a company that doesn't make it difficult to obtain technical
> information or components.

According to Freescale first samples of e5500 based silicon will be available in Q4/2010, initially scaling up to 2.0 GHz, later up to 2.2 GHz. And while the e5500 core is claimed to be able to reach up to 2.5 GHz, no such part has been announced yet.

[ Edited by Andreas_Wolf on 2010/11/13 2:09 ]

Quote:

Freescale really is a true descendant of Motorola.

In the 80's when I was managing an independent Radio Shack (which also ran another business selling multi-user 68000 systems) Motorola announced it was discontinuing the 6829 (an MMU for the 6809 processor). Since I wanted one to experiment with, I contacted Motorola. They sent me out the XC labeled component (even though their reps and I had already discussed the Tandy developed VLSI that would become the core of the Color Computer 3 - and which made the 6829 obsolete).

I always had a good relationship with Motorola (their reps frequently gave me information before it was announced and samples were easy to obtain). Since I've managed a company that built computers based on their components, I never had a problem getting support from Motorola.

Believe it or not, Freescale (while it isn't quite Motorola) is also relatively easy to deal with. Unlike Applied Micro, I might add, which wouldn't give me an NDA unless I maintained a website, Freescale's reps and I have a good relationship that goes back to the Motorola period. Of all the companies currently making microprocessor, I think I'm most fond of Freescale. They really do try harder than most companies to promote their product.

I never posted anything about the MPC8641D project, because I didn't know if I'd see it to fruition. I did consult with Andreas because he has consistently offered me the best advice (of anyone on the web I keep contact with). I actually got pretty far with the design, but it required an ATI Southbridge (an SB600) anf a FPGA for glue circuitry. The expense of building the prototypes alone was daunting.

Then the MorphOS team announced support for G4 Macs, and it seemed pointless to consider building an expensive 1.25 Mhz e600 motherboard.
Especially since I would have had to pay the MorphOS development team for porting (or at least provided several machines and convinced them that my project was worth supporting).

Anyway, yes Freescale did provide me with two MPC8641Ds worth at least a couple hundred a piece. But, that's because they took me seriously.

I've talked to the head of IBM's division that makes Cell processors (and the Xenon), Freescales reps (of course), and the designer of the X1000 (months before anyone else was discussing Varisys, Andreas and I were speculating about that company's owner). I tried to establish a relationship with Applied Micro (which I still have one contact with), but they're more secretive than IBM.

I (and I believe Andreas, although I wouldn't claim to speak for him) believe strongly that our future still lies in PPC (or PPC related) processors. While Freescale has canceled the e700, they appear to have incorporated some of the e700's features in their latest evolution of the e500 core. I've already sent in an inquiry about the P5020 and 5010 (the first 5500 cored processors).

While financing a new motherboard is a massive undertaking (and I don't doubt Treavor at A-eon's estimates of his current expenses), the design tools don't require a large team of people (just a lot of time - designs have gotten much more complicated since I was selling 68K based systems).

So, right now, yes I'm sitting on the MPC8641D samples (and a lot of preliminary design work tat has not seen recent revisions in months). IBM wouldn't agree to sell me Cell processors. Applied Micro is even harder to deal with and I have no idea how A-eon managed to secure PA6Ts (although Andreas has consistently insisted this would be the X1000's processor - and he was right).

Lets see if the 5500 core has potential. That lack of Altivec instructions is discouraging, but the first two processors are supposed to clock up to 2.2 Ghz (with 2.5 processors apparently coming soon and a design goal of 3.0 Ghz).

I've always believed that, with the proper financing, a motherboard with much better features than the boards that Acube sells could be built (and that it could be sold for about the same price as the SAM).

But, its never good to announce speculative projects. Only once something has reached at least a prototype stage, would I normally feel comfortable releasing any info on it.

The main reason I posted this, is that I want you all to realize that these things are worked on and discussed amongst a few of us. PPC is not dead, hell its not even resting.

> months before anyone else was discussing Varisys, Andreas and I were
> speculating about that company's owner

You and me started speculating about Varisys here on MorphZone beginning of April, in this very thread. But discussion of Varisys as the potential Nemo designers had started beginning of January on amigaworld.net (by BigBentheAussie):
http://amigaworld.net/modules/newbb/viewtopic.php?topic_id=30452&forum=2#532576

I have then merely joined this educated assumption:
http://www.amiga-news.de/de/news/comments/244232.html (German)

> While Freescale has canceled the e700, they appear to have incorporated
> some of the e700's features in their latest evolution of the e500 core.

As discussed here in this thread, some (see Wikipedia) even think the e5500 to actually be what was announced as e700 in 2004. But taking into account how Freescale's presentations suddenly changed from the e700 being based on e600 to being based on e500 shortly after Apple's mid 2005 switch announcement we can assume that the e5500 actually is what has been on the roadmap as "e700" since mid 2005, but not what had been on there as "e700" before, i.e. from April 2004 to mid 2005.

> I have no idea how A-eon managed to secure PA6Ts

Let's hope A-Eon will tell this "quite a story" as soon as they officially confirm the PA6T :-)

> a design goal of 3.0 Ghz

Is this insider information? Freescale officially announced the e5500 to be able to scale up to 2.5 GHz.

> I (and I believe Andreas, although I wouldn't claim to speak for him) believe
> strongly that our future still lies in PPC (or PPC related) processors. [...] PPC
> is not dead, hell its not even resting.

My stance on this:
Power Architecture seems dead regarding CPUs specifically designed for desktop use. But it's far from dead regarding CPUs that while being designed for other domains could be used for proper desktop designs as well. It's "just" that someone has to develop, build and market such designs for this small market ;-)
But to make it clear, "PPC or nothing" is not my credo, rather as I told before: As a MorphOS user, my interest lies where MorphOS is. For now that's Power Architecture :-)

Thanks for the clarifications on your part, Andreas (as I said, I would never presume to speak for you). And you had a clearer picture of the PA6T and about Varisys' owner than I ever did. After contacting Varisys, I found your input on these points to be very informative (and in the long run dead on).
I just got my first response back from Freescale. They want me to dig up the name of the contact I made when I first received my NDA.
I've looked for it in my e-mail, but when I found it the e-mail server crashed. So, now I'll have to wait till that's back up.
But then, if the P5010 isn't due to be released till Q4, I've got some time to make sure my NDA is up to date.

BTW - Thanks for bringing up this processor. The last of the e600 cored SoCs were becoming a bit dated and their development had stopped. I can't remember where I saw the 3.0 Ghz quote, I've been all over (the internet) since you first brought this up. But you're correct first 2.0 and 2.2 Ghz, then 2.5 Ghz. Twice the speed of the MPC8641! This looks very promising.

Once I have the basic specifications on the P5010 and can get an idea of how complicated a basic design utilizing this processor would be, I can reconnect with the manufacturer I have a relationship with in PA and see if its worth pursuing.

The 5500 cored processors certainly look more hopeful than the MPC86XX or the low end AM processors that Acube uses.

Thanks again for being such a well informed member of the community.

> I can't remember where I saw the 3.0 Ghz quote

It was e700 that was said to go up "to 3 GHz and beyond" when announced in 2004. Maybe you got confused by all these numbers? ;-)

> then 2.5 Ghz. Twice the speed of the MPC8641!

2.5 GHz is twice the clock of the MPC8640. Twice the clock of the MPC8641 would be 3.0 GHz :-)

Right on both points. And 2.0 and 2.2 Ghz are only .5 and .7 Ghz above the top MPC8641 respectively.

Still, if the DMIPS/MHz per core figures you've listed are even close to correct this sounds like a very powerful series of processors.

> if the DMIPS/MHz per core figures you've listed are even close to correct

I hope they (all) are. I didn't run the Dhrystone benchmark on any of these myself but found the (sometimes contradictory, like with PPC470) figures on product websites, fact sheets, product briefs, presentation PDF files and such. Still, the PA6T vs. PPC970 issue is puzzling me.

Yes, it would imply that MorphOS developers would be much better off porting to a well tested system like a G5 Mac and not to the X1000.

|||

[ Edited by Velcro_SP 25.04.2011 - 07:09 ]

|||

[ Edited by Velcro_SP 20.04.2011 - 07:59 ]

Quote:

I still don't, whatever...

> Show everyone where I said a second, redundant PowerVR should be
> added to a 5121e computer.

I didn't say you said "should", but that you said that a second PowerVR could be a potential source of performance improvement in an MPC5121e based system (which I disagreed with). There you go:

"As well, I'd like to know how far performance can be pushed on slow, low-wattage processor by way of more RAM and dedicated graphics chip, like PowerVR or Theora decoder. I just don't think the world has moved on from that. I think the world should move on *to* that.
I don't mean to zero in on the 5121e as the only thing, but really even given the uncertainty it seems much more established to me than this other low-wattage, small stuff."

And while you're in input mode let me take the opportunity to ask you again:

Did "other things being talked about" and "this other low-wattage, small stuff" include PowerQUICC II Pro, which was talked about? Yes or no?

|||


[ Edited by Velcro_SP 20.04.2011 - 07:54 ]

There's no point in being obsessively argumentative here.

And I'm not impressed with the performance of the 5121e (and PowerVR is ancient).

As to PowerQUICC II Pro based processors, since they use the e300 core (based on the PowerPC 603e that runs in some Amiga accelerators) these processors are not limited to the applications you've mentioned (they could easily run MorphOS).

Further, PowerQUICC is being supplanted by the software-compatible QorIQ platform on e500 cores. As we were discussing the latest development to the e500 core (the e5500) and the P50xx processors based on that core use less than 30 watts, that seems like reasonably low wattage. And there are lower wattage PPC processors available that don't give up as much performance as the 5121e.

I don't understand why are so many posters fascinated by low cost, low performance systems built for third world markets. Do you really want to put an OS on a PC that costs more than the computer?

Recently when I discussed PPC based systems with the owner of Varisys (the designers of the X1000 motherboard) I got this response:

"In terms of PPC we are thinking QorIQ here for higher end designs."

We can continue to discussing relatively dead end processors like the 5121e and the PA6T, or resume discussing processors that seem to have a future.


[ Edited by Jim on 2010/7/3 18:40 ]

> Even there, you can't resist removing the paragraph break. That didn't happen with the
> copy and paste. You went in and removed a blank line. Am I wrong? Yes or no?

No, you are right. I like to have things condensed. And I don't think that removing the blank line changed anything in meaning. After all, there's still the paragraph break, isn't it?

> At least you included the Theora part. You didn't include that in your first quote
> of this text, or in your second expanded one.

Yes, as I prefer to not quote things I'm *not* referring to. Wouldn't make any sense. I didn't refer to "more RAM" but to "dedicated graphics chip". And in that I didn't refer to "Theora decoder" but to "PowerVR". That's why I omitted "more RAM" as well as "Theora decoder". I can't see anything dishonest in that.

> You're mashing the paragraphs together

I removed a *blank* line. The paragraph break ist still there.

> they still don't say to add a second PowerVR to a 5121e computer.
> [...]
> None of that refers to adding a second PowerVR to a 5121e system.

Let's see. There are 4 paragraphs. In the 2nd one you talk about Linkbook, LimePC and your Cherrypal C114, i.e. MPC5121e based computers only. In the 4th paragraph you explain that the other "low-wattage" processors than MPC5121e are not established enough for your liking. And now you say that inbetween, i.e. in the 3rd paragraph, you didn't mean to include the MPC5121e when mentioning "slow, low-wattage processor"? Yes or no?

> I was not thinking of PowerQiCC.

Thanks for eventually answering that question. So it seems you don't necessarily think the MPC5121e to be much more established than the PowerQUICC II Pro. Good to have that clarified.

> I don't think I've ever discussed network or routers or automotive or
> point-of-sale etc. computers here.

Huh? PowerQUICC is not a whole system but a *processor* family. You referred to "other things being talked about" and "this other low-wattage, small stuff". And PowerQUICC II Pro actually *had been* talked about right before here in this thread and *is* low-wattage, small stuff. So there's no ill will involved in assuming you may have included PowerQUICC II Pro when mentioning "other things being talked about" and "this other low-wattage, small stuff".

Quote:

Hey, they're Amiga people- what do you expect?

Quote:
This could be a great way to go for penetration of developing market though: with MorphOS available to OEMs at a low-cost (or charitable donation) OEM price, if proven to perform well on eg the "$100 noteboook" compared to eg Linux, we could have a realistic way to increase the userbase by an order (or several) of magnitude. Not to mention people's brand loyalty to the first OS with which they learned to use a computer.

Dreaming perhaps, but an example of how the low-cost low-performance systems could be very interesting for MorphOS.

> PowerVR is ancient

The MBX Lite that is incorporated into the MPC5121e, yes. But there're newer and better PowerVR GPUs available than that: http://en.wikipedia.org/wiki/PowerVR#Series5_.28SGX.29

Hard to believe they're still developing PowerVR, Andreas, but they seems to have re-targeted their products so they don't have to compete with ATI and Nvidia.

I still don't see the point in focusing on the MPC5121e, its aging fast.

And there are a lot of other low wattage PPC processors.

Sorry, boot_wb, sometimes I forget that at times we all take things too seriously.

A cheap netbook would be nice, but right now I can buy a G4 Powerbook cheap so I hope we see that port soon. That would blank any 5121e based device.

> Hard to believe they're still developing PowerVR

They're quite successful with it, with Apple being one of their most famous customers (iPad, iPhone, iPod Touch).

> they seems to have re-targeted their products so they don't have to compete
> with ATI and Nvidia.

Yes, when Kyro III was ready for release in 2002 their partner STMicro pulled back, leaving them no other option than to enter portable graphics market and optimize Kyro III for that, resulting in the PowerVR MBX in the same year.

http://en.wikipedia.org/wiki/PowerVR#MBX

> I still don't see the point in focusing on the MPC5121e

Me neither. I'm a desktop user, and I consider a 400 MHz e300 too weak even for a netbook that could be of use to me.

> there are a lot of other low wattage PPC processors.

Exactly. That's why when talk was about the e300 in this thread I said that my vote would be on PowerQUICC II Pro (800 MHz MPC8377E), if it must be e300 by all means that is.

SO Velcro_SP was offended because you suggested a more powerful processor and somehow came to think that said processor was only suited to "network or routers or automotive or point-of-sale etc. computers".

Since I've been part of a company that marketed Point of Sale computers I'm vaguely offended. Unless you're talking about relatively dumb terminals, the computers behind a POS systems need to be fairly robust (you can't do business if your systems down).

And why can't we leave the low performing PPCs to the AmigaOS crowd?

> why can't we leave the low performing PPCs to the AmigaOS crowd?

Mind you, they're going to have the PA6T. There's currently no better performing CPU a PPC based AmigaOS like OS can run on ;-)

Quote:

From the DMIPS/MHz per core figures you've list earlier, the e500, e600, G5, and e5500 if run at similar speeds would all outperform the PA6T.

Of course, not all those cores clock to 1.8 Ghz (or above), but several do. And since the continued availability of PA6Ts is questionable I hope something else comes along.

And, of course, if you can't afford a X1000, the you better hope you can find a good used Pegasos, because otherwise you're stuck running on one of Acube's anemic boards.

> From the DMIPS/MHz per core figures you've list earlier

Dhrystone is a very artificial benchmark.

> the e500, e600, G5, and e5500 if run at similar speeds would all outperform the PA6T.
> Of course, not all those cores clock to 1.8 Ghz (or above), but several do.

From my list only e5500, PPC970 and PPC470 would be able to beat the 1.8 GHz PA6T in DMIPS figures at their respective maximum clock rates.

> you better hope you can find a good used Pegasos, because otherwise
> you're stuck running on one of Acube's anemic boards.

I'm eager to know how the PPC460EX performs in real life compared to the equally clocked MPC7447 in the Pegasos II G4. I expect the G4 to perform somewhat better though, even without AltiVec.

I'd expect the 7447 to outperform a similarly clocked PPC460EX as well, and the dhrystone measurement you gave for the Titan don't look too promising either.

A 7448 would outperform a 7447, and apparently the three processors you've mentioned would outperform that (of course, all based on - as you pointed out - a very artificial test).

And Acube's pricing, based on performance, is worse than A-eon's.

Now, as to the max speeds of the processors you've mentioned, they appear to be higher than the PA6T (unless that runs at 2.0 and so far its only been seen running at 1.8).

Wouldn't we still be better off with G5 Powermacs. I can get a fairly well equipped high speed G5 for hundreds less than an Acube based system.

And, when (if), we get Powermac support there are the 7447 and 7448 upgrades. A 1.8Ghz might or might not outperform an X1000, but its going to walk all over anything from Acube.

As HD video decoding with current MorphOS supported hardware can be problematic at really high res, why worry about even slower hardware than the Pegasos?

> apparently the three processors you've mentioned would outperform that
> (of course, all based on - as you pointed out - a very artificial test).

Based on that benchmark, a 1.8 GHz PA6T core would marginally (by 1.3%) outperform a 1.7 GHz 7448 (that's the maximum stock clock rate according to Freescale) as well.

> Acube's pricing, based on performance, is worse than A-eon's.

Taking only one core of the PA6T into account, Dhrystone performance of the 1.8 GHz PA6T is twice that of the 1.0 GHz PPC460EX. I expect a full Sam460ex system to be priced somewhat less than half of the X1000. On the other hand I expect real life performance of one 1.8 GHz PA6T core to be much better than only twice that of the 1.0 GHz PPC460EX, add to that AltiVec capability, faster bus, larger caches and the option to use the 2nd core whenever OS4 will support that. So yes, you may be right in your estimation.

> as to the max speeds of the processors you've mentioned, they appear to be higher
> than the PA6T (unless that runs at 2.0 and so far its only been seen running at 1.8).

Not quite. The PPC470 is clocked at up to 1.8 GHz as well (in the LSI Axxia, while being specified at only 1.6 GHz max by IBM btw).

> Wouldn't we still be better off with G5 Powermacs.

Performance wise, probably yes. But not everyone would be happy with the high wattage of the PPC970 I guess. That's where the PA6T shines in comparison. In real life performance (i.e. not Dhrystone) I expect the PA6T to have significant higher performance/wattage figure than the PPC970 by providing only slightly less performance/clock.

> there are the 7447 and 7448 upgrades. A 1.8Ghz might or might not
> outperform an X1000

In terms of Dhrystone an e600 core above 1.72 GHz would outperform a 1.8 GHz PA6T core. So the overclocked 1.8+ GHz G4 upgrades deliver more DMIPS than a 1.8 GHz PA6T core. In real life performance though I think a 1.8 GHz PA6T core would outperform an equally clocked 7448.

> but its going to walk all over anything from Acube.

Yes, that's for sure.

> As HD video decoding with current MorphOS supported hardware can be problematic
> at really high res, why worry about even slower hardware than the Pegasos?

I'm just curious to the whole thing. Be assured that I definitely won't downgrade from my 1.5 GHz G4 ever :-)