> Luckily we already know that we can reconfigure
> for [...] three X4 PCIe configurations.

I just discovered that the only configuration providing x4 x4 x4 ('0x14') doesn't provide any lane for SATA, so that would leave us with either x8 ('0x10') or x4 x4 x1 x1 ('0x2'), which both have all two SATA controllers connected (and use only 16 lanes overall out of 18 lanes).

Thanks, with those configuration I still don't understand why X8 and X1 aren't an option, but the X4, X4, X1 configuration is not a big loss (having one X1 in place of the last x4).
Are you telling me both SATA ports are run with one PCIe lane?

> with those configuration I still don't understand why X8 and X1 aren't an option

I don't understand either the reason there's no full configurability but 29 different fixed configurations you have to select one from. But Freescale told you so, and we have to accept that as fact for now.

> the X4, X4, X1 configuration is not a big loss (having one X1 in place of the last x4).

It's one x1 more, so all four PCIe controllers used in this configuration (and moreover, it's the *only* configuration with all four controllers used).

> Are you telling me both SATA ports are run with one PCIe lane?

No, I meant to tell that both configurations '0x10' and '0x2' have two (i.e. all existing) SATA controllers connected (using 1 SerDes lane per 1 SATA controller). Just look at the configuration file you thanked me so much for linking it. It's all in there.

Thanked you so much for? Until you referenced this, I hadn't been able to get anything like it out of Freescale. I couldn't even get a count of the number of available PCIe lanes available to the controllers.
Now, between what you've dug up and what they've recently divulged we have a better idea of how the P5 processor handle SerDes configuration and what available options there are.
Considering the Freescale has not lifted the NDA on these products I think your contribution to this was pretty significant.
You have a way of ferreting out useful information like this that is remarkable.

From what Freescale has told me, the evaluation board may be ready before the NDA is lifted. they've got a conference scheduled (I think for June) during which one of the topics to be covered will be porting Linux to the P5 (so my guess is that they haven't completed this yet).

Until we hear more from APM, the e5500 looks to be the most promising PPC to be recently announced. You brought this up before many were even aware of it.
Come to think of it, both you and Varisys' staff, started pointing to the QorIQ platform at about the same time.

> Thanked you so much for?

Yes, there:
https://morph.zone/modules/newbb_plus/viewtopic.php?topic_id=7183&forum=3&post_id=82274#82274

> Until you referenced this, I hadn't been able to get anything like it out of Freescale.

You mean they started giving you more information because they became aware that you had become aware of that link? Or was it just coincidence?

> between what you've dug up and what they've recently divulged we have
> a better idea of how the P5 processor handle SerDes configuration and what
> available options there are.

To be precise, we did know the options from the link I dug up already. The new information from Freescale was that those 29 ready-made configurations are really the only configurations available and that there's no way to create an own configuration.

> Considering the Freescale has not lifted the NDA on these products
> I think your contribution to this was pretty significant.

I think we should thank Freescale's Kumar Gala for this as he posted those information in January to a public mailing list for the whole world to read ;-)

> they've got a conference scheduled (I think for June) during which one of the
> topics to be covered will be porting Linux to the P5

Yes, the annual Freescale Technology Forum. The titles of this year's technical sessions have already been published:

http://www.freescale.com/webapp/sps/site/overview.jsp?code=FTF_2011_AMERICA_SESSIONS_CAT

From the 'Networking' tab:

"QorIQ P5020DS Processor: Overcome the Challenges in Porting U-Boot and Linux Operating System"

And also interesting:

"Using Freescale's AltiVec SIMD Engine for Graphics, Image Processing, Radar, Video Surveillance and Other Applications"

They didn't have sessions on AltiVec for some years. I guess this is going to be a foretaste of AltiVec-enabled QorIQ T.

> so my guess is that they haven't completed this yet

I think they may have, but they surely wouldn't antedate the FTF just because of this ;-)

> You brought this up before many were even aware of it.

Actually, I became aware of the e5500 only some hours before it was officially announced in June 2010 by Freescale, while public sources had been (unnoticed by me) mentioning it as 'e500mc64' since November 2009 at least. The P5020 was even explicitely mentioned in August 2009 already (but not revealed as being 64-bit back then).

https://morph.zone/modules/newbb_plus/viewtopic.php?topic_id=6313&forum=11&post_id=72666#72666
https://morph.zone/modules/newbb_plus/viewtopic.php?topic_id=6313&forum=11&post_id=74067#74067
https://morph.zone/modules/newbb_plus/viewtopic.php?topic_id=7001&forum=3&post_id=74074#74074

So you see, I was rather late to that party ;-)

> both you and Varisys' staff, started pointing to the QorIQ platform
> at about the same time.

I don't know when Varisys started pointing to the QorIQ, but I started in January 2009:

https://morph.zone/modules/newbb_plus/viewtopic.php?topic_id=6196&forum=11&post_id=61505#61505

And as you can see, not in a too favourable way back then ;-) February 2009 was when I began to see the suitability of QorIQ somewhat:

https://morph.zone/modules/newbb_plus/viewtopic.php?topic_id=6196&forum=11&post_id=62074#62074

>So you see, I was rather late to that party

I missed the party completely. At least you brought it up and drew everyone's attention to it.

>Actually, I became aware of the e5500 only some hours before it was officially announced in June 2010

At the time I was focused on Powermacs, as there seemed t6o be no point to continuing to explore the MPC86XX processors. Since I wasn't paying the least bit of attention to Freescale I completely missed it.

>And as you can see, not in a too favourable way back then February 2009 was when I began to see the suitability of QorIQ somewhat:

OK, before I had contact with Varisys, and the initial negative view? Understandable considering that the e500 is a significant step down from the e600.

So, whether later than some others or borrowing references from sources you found, it was all useful news to me. Right now I've got more information on this processor than I've ever had without an NDA. And since I don't feel like going to the trouble of setting up a corporate website, I'm not asking Freescale for an NDA (at least not yet).

So, yes, THANKS.

[ Edited by Jim on 2011/3/13 23:40 ]

> the initial negative view? Understandable considering that the e500
> is a significant step down from the e600.

Problem was that it took me some time to become aware of the e500mc. As you can read in the postings I linked to I originally mistook the faster and more suited e500mc based QorIQ chips to be e500v2 based like the slower and less suited QorIQ chips.

> since I don't feel like going to the trouble of setting up a corporate website,
> I'm not asking Freescale for an NDA (at least not yet).

So it's not just Applied Micro expecting its NDA partners to have a website, but Freescale as well? You once said you had an NDA with Freescale at some point in time (which somehow got lost because your contact at Freescale had vanished). Did you have a website back then? Or did Freescale's requirements change in that regard?

Btw, I don't know if you missed it, but could you please answer my question? (It was: You mean they started giving you more information because they became aware that you had become aware of that link? Or was it just coincidence?)

>Problem was that it took me some time to become aware of the e500mc.

I didn't look into that till you pointed it out. You'd assume that e500s would be more or less identical except for clock speed. And that assumption makes perfect sense. You actually have to go over the specs to realize its not another e500v2.

>So it's not just Applied Micro expecting its NDA partners to have a website, but Freescale as well? You once said you had an NDA with Freescale at some point in time (which somehow got lost because your contact at Freescale had vanished). Did you have a website back then? Or did Freescale's requirements change in that regard?

Yes, Freescale has adopted this policy as have a lot of other companies. Also, I get the impression that a fair number of people left the company at about the time plans for the e700 were dropped and they started to invest more in ARM technology.
The thing that surprised me is that I had a contact in the technical support department and in sales (the MPC8640Ds were actually supplied by technical support which is a little unusual). In the time between when they sent me the MPC8640D samples and when you brought up the e5500 cored processors both individuals had left the company.

I'm not sure a website is the greatest idea since I'm still discussing using a currently unused corporate name with the wife of one of my former employers (and paying for a S corp. instead of the current LLC).
To be honest with you, in the last few years the only reason I paid to maintain the last companies business license is so I could write off certain expenses and lower my tax burden.

Its easier to work for someone else than to continue to hustle looking for contracts yourself.


[ Edited by Jim on 2011/3/14 1:03 ]

> You'd assume that e500s would be more or less identical except for clock speed.

Actually, both e500v2 and e500mc reach up to 1.5 GHz. It's only the QorIQ chips where e500v2 maxes out at 1.2 GHz. In the PowerQUICC III the e500v2 core reaches up to 1.5 GHz.

> You actually have to go over the specs to realize its not another e500v2.

Unfortunately, I slightly misremembered what my misconception prior to February 2009 actually was. I said that I had thought all QorIQ chips were to be based on e500v2. But that is not true. I actually knew from the start about the e500mc and which QorIQ chips were to be based on that core instead of the e500v2. What I wasn't aware of back then is that the e500mc got a proper FPU which the e500v2 misses. Instead, I believed the only advantage of the e500mc over the e500v2 to be that more than just two of them could be connected to form multicore chips (e500v2 can be used only for single or dual-core).

> at about the time plans for the e700 were dropped

You actually know when that was? I ask because I don't have a clue, and public information on that are conflicting.

https://morph.zone/modules/newbb_plus/viewtopic.php?topic_id=6196&forum=11&post_id=72301#72301
https://morph.zone/modules/newbb_plus/viewtopic.php?topic_id=7001&forum=3&post_id=74210#74210

>You actually know when that was? I ask because I don't have a clue, and public information on that are conflicting.

No you didn't get a straight answer on this from the company at all.
All mention of the e700 just disappeared.
I guess Freescale didn't see a market for it.
And like I mentioned earlier, suddenly a lot of people left the company.
The P5 surprised me because I, like a lot of people, assumed they were re-focusing on ARM.

> All mention of the e700 just disappeared.

All the more strange that a Freescale presentation as recent as June 2010 mentions the e700 (see my links).

> I guess Freescale didn't see a market for it.

Considering that the e5500 actually is what the e700 was supposed to be after Apple's Intel switch announcement (which made Freescale change the e700's base from e600 to e500) they apparently see a market for it again.

> The P5 surprised me

The information that there will be a P5 outdoing the P4 in performance was published by Freescale when they introduced the QorIQ family and announced its first (P1, P2 and P4) members in June 2008. The P5's exact nature was not revealed back then, though. What surprised me in particular with the P5 was that it's a 64-bit chip. I didn't see that coming until few hours before the official announcement.

>they apparently see a market for it again.

Yep, a different one.

An e600 offshoot might have had a better expansion slot scheme.

> Yep, a different one.

What did Freescale claim as the MPC87xx's target market after Apple's switch to Intel? I can't find any information on this.

Quote:

I'm not sure what they thought the target for the MPC86XX markets was (especially after Apple's departure). The processors were too high priced compared to the (non PPC) alternatives.

Was there ever an outline of the basic features for the e700 cored products?

> I'm not sure what they thought the target for the MPC86XX markets
> was (especially after Apple's departure).

But how can you know that now with QorIQ P5 it's a different one then? (Btw, it's 87xx, not 86xx.)

> The processors were too high priced compared to the (non PPC) alternatives.

I don't think there were ever any prices mentioned for the MPC87xx anywhere.

> Was there ever an outline of the basic features for the e700 cored products?

Yes, numerous, and some of them kept changing over time. The MPC87xx was supposed to feature (selection from various points in time):

- single-core (87xx) and dual-core (87xxD) variant
- 65 nm
- <20 Watts single-core (don't know about dual-core)
- 3.0+ GHz (later decreased to 2.5+ GHz, then further decreased to 2.4 GHz)
- 1 MiB L2 cache per core
- 1 GHz DDR2/3 memory controller per core
- two PCIe controllers
- SRIO
- four GbE controllers
- pattern matching
- TCP offloading
- L4-7 content processing

>Btw, it's 87xx, not 86xx.

I actually did mean 86XX (as they are still available after Apple's departure, but are now somewhat pointless with the introduction of the P5).

I've rarely seen the X700 cored products (what ever they would have been like) refereed to as MPC87xxs (although that is what they would have likely been called).

Were there any real details (let alone prices)?

> I actually did mean 86XX

Okay. To relate to the point what Freescale thinks the target market for the e600 based MPC86xx has been, we shouldn't forget that the MPC74xx has always been used also in the embedded area, not only as CPU for Apple. So we can assume that the target market for the MPC86xx, which is the MPC74xx's successor, stayed the same. But let's not just assume, but rather see what Freescale says:

MPC8610:
"Robotic vision and navigation; Aerospace/defense display, control and image processing; Kiosks with image processing; Multi-function printers and scanners; Single-Board Computers"

MPC8640(D):
"embedded networking, telecom, aerospace and defense, storage, industrial and pervasive computing applications"

> as they are still available after Apple's departure

Actually, they started being available only after that. The first one of the series, the MPC8641(D), was announced in 2004, i.e. before Apple's switch announcement, but began sampling only in 2006.

> but are now somewhat pointless with the introduction of the P5

I wouldn't say so. The markets the MPC74xx and later the MPC86xx have been aimed at heavily rely on the capability of AltiVec code execution (or SIMD processing in general, with AltiVec being preferred for code reuse reasons). That's why the current QorIQ P series is not considered an adequate MPC86xx replacement by customers from those industries, albeit Freescale wishing it would. To compensate for that and not lose too many customers to Intel and its SSE and AVX capable processors, Freescale did what you know they did: announce the availability of AltiVec for future QorIQ processors. They wouldn't have done that if they had been able to avoid it. As soon as QorIQ with AltiVec will be there, MPC86xx will be rather pointless, yes. But so far, if you want AltiVec then MPC86xx is a viable option. More there:

https://morph.zone/modules/newbb_plus/viewtopic.php?forum=3&topic_id=7001&start=455

> I've rarely seen the X700 cored products

It's e700 ;-)

> refereed to as MPC87xxs (although that is what they would
> have likely been called).

Likely? 'MPC87xx' is the moniker Freescale used to refer to the e700 core based processors in their presentations and roadmap depictions. Just explore the documents linked from those Google results:

http://www.google.com/search?q=%22mpc87xx%22+freescale
http://www.google.com/search?q=%22mpc87xx%22+%22e700%22

There's even still one PDF file online on the Freescale website (page 6):
http://www.freescale.com/files/community_files/MCUCOMM/1033_e300_e500_e600_comp.pdf

Unfortunately, most of the old Freescale documents mentioning the MPC87xx are not online anymore and thus not indexed by Google anymore. (But don't be confused by some of the results as it seems there was an 'MPC87xx' already on Motorola's 2001/2002 roadmaps, but that was another thing altogether.)

> Were there any real details (let alone prices)?

I'm only aware of the details I listed in my previous posting. And as I said, I don't think there were any prices mentioned.

Thanks for getting me what you could.
And you're right, until the T5 replaces the P5
Then the product will have AltiVec.


Plus, the MPC8641 has more PCIe lanes (although only 1.0) and PCI slot capability.

But the speed advantage of the P5 and the future introduction of AltiVec makes this the processor to look out for.

The e600 is dead.

> the MPC8641 has more PCIe lanes (although only 1.0)

Yes, it can dedicate up to 16 SerDes lanes (x8 x8) to PCIe. But as you say, that only equals 8 lanes (x4 x4) when compared to the QorIQ P5's PCIe 2.0. So overall, the QorIQ P5 can dedicate up to 50% more bandwidth to PCIe than MPC8641(D), or up to 25% more bandwidth in a configuration suited for desktop computing (i.e. with on-chip SATA controllers enabled).

> and PCI slot capability.

Neither MPC8641(D) nor MPC8640(D) incorporate a PCI controller. Do you mean the MPC8610 here maybe?

> The e600 is dead.

...as soon as the AltiVec enabled QorIQ T is there, yes. Regardless, Freescale will offer the existing MPC86xx chips for quite some years to come.

https://morph.zone/modules/newbb_plus/viewtopic.php?topic_id=7001&forum=3&post_id=80669#80669

>> and PCI slot capability.

When used with a Southbidge a MPC8640/8641 design can incorporate PCI slots.

> When used with a Southbidge a MPC8640/8641 design can incorporate PCI slots.

Yes, but so can a QorIQ P5 design (and any other design based on a chip with an interface a southbridge can be attached to), so that's no difference between the two. And of course it doesn't need a southbridge chip for this. A PCIe-to-PCI bridge chip would do well. With the QorIQ P5 in x4 x4 x1 x1 PCIe configuration you could use one x1 to bridge to a PCI slot.

Interesting idea though. An SB800 Southbridge would cost us four PCIe 2.0 lanes, but would also supply four PCie lanes.
Which configuration was it that you said we lost SATA controllers?

> An SB800 Southbridge would cost us four PCIe 2.0 lanes,
> but would also supply four PCie lanes.

According to the 'AMD SB810/850 Southbridge Databook' there're only two PCIe 2.0 lanes supplied, not four:

http://support.amd.com/us/Embedded_TechDocs/44758.pdf (page 12)

> Which configuration was it that you said we lost SATA controllers?

The one you liked most ;-)

https://morph.zone/modules/newbb_plus/viewtopic.php?topic_id=7183&forum=3&post_id=82381#82381

So let me summarize the two options:

1. without southbridge:
- SATA controllers on P5 must be enabled, thus x4 x4 x1 x1 PCIe config is best
- use one x1 with bridge chip to get legacy PCI slot
- for audio there're those options to select from:
-- sound card in legacy PCI slot
-- sound card in PCIe x1 slot
-- on-board audio chip on PCIe x1
-- nerdy: use P5's SPI bus with a CPLD acting as SPI-to-I²S bridge and connect an audio chip via this I²S ;-)
- one x4 for graphics card
- leaves one of the following PCI(e) connection configs for use at will: x4+x1 or x4+PCI or x4+x1+PCI

2. with SB800 southbridge:
- don't need P5's SATA controllers, thus x4 x4 x4 PCIe config can be used
- use one x4 to attach SB800
- legacy PCI provided by SB800
- audio provided by SB800
- additional two PCIe lanes provided by SB800, either two x1 or one x2
- one x4 for graphics card
- leaves one of the following PCI(e) connection configs for use at will: x4+x1+x1+PCI or x4+x2+PCI

Great summary Andreas. I hadn't even had a chance to go over this yet (and now I don't have to).
The Sb800 SATA controllers are 3.0. What are the P5s?

With an SB800 series Southbridge we also gain additional SATA controllers (six total).

Very interesting comparison.
W/O SB - X4+X4+X1 PCIe 2 SATA
W SB X4+X4+X1+X1 PCIe 6 SATA and PCI

[ Edited by Jim on 2011/3/21 4:46 ]

[ Edited by Jim on 2011/3/21 11:04 ]

> The Sb800 SATA controllers are 3.0.

On SB850, yes. SB810 has only 2.0.

> What are the P5s?

2.0.

> Very interesting comparison.
> W/O SB - X4+X4+X1 PCIe 2 SATA
> W SB X4+X4+X1+X1 PCIe 6 SATA and PCI

The question is whether those benefits of having the SB800 would really be worth the additional costs.

[ Edited by Andreas_Wolf on 2011/3/21 17:04 ]