Addendum:

> I doubt Varisys simply forgot to use more than two PCIe controllers, but believe
> there is a good reason for using only two of them. It might be connected to the
> fact that the board has been designed for three different SoCs (P3041, P5020,
> P5040) which are not 100% pin-compatible, so the actual board design
> represents the lowest common denominator between the three SoCs.

I just remembered that 6½ years ago, Jim and I discussed the SerDes assignment options for the P5020's PCIe controllers:

http://morph.zone/modules/newbb_plus/viewtopic.php?forum=3&topic_id=7183&start=154
http://morph.zone/modules/newbb_plus/viewtopic.php?forum=3&topic_id=7183&start=177

As can be seen, the most sane PCIe config for the P5020 used in a desktop computer would have been x4 x4 x1 x1. Now as the P5040 lacks the fourth PCIe controller, this option can't be used on the board, so Varisys had to use the next best option instead, which is x4 x4 (with bridging the 2nd x4 to x8). And even if there was an x4 x4 x1 config for the P5020 (which there isn't), this would have resulted in only 1 lane more (or 3 lanes less without the bridge for the 2nd x4).
I think this should explain why the Cyrus board only uses two PCIe controllers.

but the most intelligent think will be for me if they use the P5021 and not the P5020 for have less problem?
or dont made the 5040 version and only 5020 and P3041 because are Pin compatible. P3041 plus is pin compatible with 4040 and 4080


Quote:

> for them was best made two different board one for 5020 and one for 5040
> put less slots and dont use pcie-pcie bridges and use the 3 pcie controller

If they had made two different boards for X5000/20 and X5000/40, they could have used all four PCIe controllers of the P5020 in x4 x4 x1 x1 config. For the P5040, the only SerDes config that includes SATA and all three PCIe controllers is x4 x1 x1, which is even less lanes than the x4 x4 config used now.
This means having two different boards would have benefited the X5000/20, but not the X5000/40.

29 SerDes assignment configs for P5020
14 SerDes assignment configs for P5040

As you can see, the only matching configs between the SoCs are 0x11, 0x15, 0x2a, 0x34, 0x35 and 0x36. Cyrus uses 0x11.

andreas
but 4x 1x 1x can be good if will gave the best and lest issue compared to put bridge over bridge.
if an user was need a pci for example there are many adaptor on amazon and this will made the mobo probably prizeless for the less components on it.

right now on my x5000 im using only 4x 1x 1x and im ok.

[ Edited by tlosmx 13.09.2017 - 15:28 ]

Well, since the 5040 is essentially pointless (from an "Amiga" perspective), it would have been better to only do *one* board that fully make use of the *5020 specs*. Now they have obviously made compromises to be able to use the same board with a CPU that no Amigan wants or needs.

Jim wanted one.
I did too but opted to get the X5000/20 as my Sam460ex was acting up.

Quote:

*5010 have same specs but can be more amigan

Quote:

There are several "Amigans" who want an X5000 5020 or 5040, who either want it to also run Linux PPC, or still believe that Hyperion has a viable plan for making some kind of "hacky" (wacky) way of eventually getting AmigaOS4.x to utilize more than a single core, without breaking all backward compatibility.

Since I no longer have any faith that Hyperion can accomplish any serious improvements to AmigaOS4, I will ignore them completely, until they have something to show running on my X1000 (like the free upgrade to the imaginary AmigaOS4.2. Doesn't this smell familiarly fishy, like when Hyperion delayed the release of AmigaOS4, just to steal it from Amiga Inc. control via the shady legal agreement/contract, they had gotten Bill Mc fricken Stupid Ewen to sign? Oh wait, I guess it this a different and completely incompetent programming and impossible claims for the OS kind of thing that is preventing Hyperion from completing AmigaOS4.2!)

Quote:

Hardly as an Amigan. Hardly as a Linux user as well (ARM and x86/x64 offers plenty of better options, whatever you need). He is a PPC fetishist, and it still remains to be seen if he will actually put his money where his mouth is in this matter.

Personally, I doubt it.

It's a lot of money.

Quote:

The problem is the "or" in your sentence above, since this option of using the 5040 with the same board design seems to have cripled the board for all 5020 users (and 5020 is less of a *waste* for Amigans than the 5040, hence this is what most Amigans will choose). No doubt the 5040 will be more expensive, it will have 3 unused cores instead of 1, it may be clocked faster but still without being fast at all, not measured against modern hardware, not even measured against 2005 level Mac PPC hardware. Obviously not even in Linux using all 4 cores, where an ancient Quad Powermac easily beats it at the fraction of the price tag.

Quote:

Yes, the problems the 5040 (and even 5020) are having make it less attractive to rational people, but if the 5040 runs AmigaOS4.1FE faster than my X1000, or even just marginally faster than the 5020, there will be a small number of AmigaOS4 fanatics that will pay the price to have the best experience running their beloved AmigaOS4.

None of the Amiga and Amiga inspired options are rational choices today for anything other than nostalgic fun, so it is all just a matter of levels of lunacy in trying to use any of our systems for modern tasks. I see it as similar to many other hobbies, including collecting and/or restoring and reselling antique cars. The money spent is often several times more expensive than the price such antique cars can be sold for, after they have been restored, if the car is old and needs major repairs and/or is missing expensive replacement parts. I think most antique car owners buy them to enjoy for a short while, then sell them at a loss to the next collector, who gets some enjoyment from driving, or just looking at them in their garage. On rare occasions, some of the antique cars will increase in value, depending on the rarity of the model, and the general state of the world economy. Buying the already restored antique cars at auctions can sometimes yield lucky results, and lead to a potential profit, but those opportunities in any hobby I think are fairly rare.

We did discuss this quite a while ago, and now that you go back over it, I guess this is about the optimal configuration.

As these are Gen2 PCI-E lanes, they have twice the bandwidth, so it's really not that bad.

> I actually preferred the X1000's expansion slot setup as the P50xx doesn't offer
> as many PCI-E lanes

Indeed, 23 of the 24 SerDes lanes of the PA6T on the Nemo/X1000 are assigned to PCIe, while only max. 12 of 18 lanes are assignable to PCIe on the P5020 and even only max. 8 of 20 on the P5040. Freescale really didn't cover themselves in glory there.

> Seems like the expansion busses and PCI-E routing on these boards is less than ideal.

That's the price to be paid when having to settle for the lowest common denominator for a board used with two SoCs (three originally) that are not 100% pin-compatible. Given that task and the number and type of expansion slots A-Eon probably requested, Varisys found the ideal solution in selecting x4 x4 config and using a bridge on the 2nd x4 to gain 4 more PCIe lanes for a total of 12.
Varisys can only work with what they get, and they can't overcome Freescale's bad decisions (or A-Eon's, for that matter).

> This is one area the X1000 seems to have a leg up on.

Absolutely, as the PA6T seems to allow an arbitrary SerDes lane assignment, whereas the QorIQ series is really limited in this regard by only allowing to select from a number of fixed assignments (none of which defined with desktop use in mind, of course).

>Indeed, 23 of the 24 SerDes lanes of the PA6T on the Nemo/X1000 are assigned to PCIe, while only max. 12 of 18 lanes are assignable to PCIe on the P5020 and even only max. 8 of 20 on the P5040. Freescale really didn't cover themselves in glory there.

It's probably related to the intended use of the CPUs. The PA6T was intended to be more of a general purpose processor, whereas the the Qorlq line is aimmed at communications devices.

> the most intelligent think will be for me if they use the P5021 and not the P5020
> for have less problem?

The Cyrus Plus board with x4 x4 PCIe config as it is now is already configured optimally for the P5040 (and P5021). Using the P5021 instead of the P5020 would have brought no benefit except of a 2.2 GHz X5000/21 instead of the 2.0 GHz X5000/20. PCIe config would have been the same.

> or dont made the 5040 version and only 5020 and P3041 because are Pin compatible.

Indeed, the P3041 and the P5010/P5020 are both 100% pin-compatible and 100% SerDes config compatible. A board made for them (or just for P5010/P5020) would have meant no compromise in board design and SerDes lane assignment, allowing for x4 x4 x1 x1 PCIe config.

> P3041 plus is pin compatible with 4040 and 4080

P3041/P5010/P5020 are not more compatible with P4040/P4080 than with P5021/P5040 (which doesn't mean the last two groups are more compatible with one another, just in case someone might hit on this idea).

Quote:

Because of the lousy situation the ppc hardware market is in, I advocate the x64 shift since years. ppc is nice, ppc processors may be okay, but boards.... Not to question the comptence of the varisys guys, but I guess ASUS or Dell are just more professional on designing hardware - they are just more experienced. And if one board performes poor, there are plenty that don't.

To make it short: Avoiding ppc - avoidig problems.

> Not to question the comptence of the varisys guys, but I guess ASUS or Dell are just
> more professional on designing hardware - they are just more experienced.

As said, Varisys are certainly not to blame for the design compromises made for the Cyrus/X5000. If anyone is to blame, then it's Freescale/NXP for the limited and incompatible SerDes configs provided in their SoCs, and maybe also A-Eon for commissioning a common board for two sub-100% compatible SoCs and requesting the given number and type of expansion slots at the same time.
I doubt that, given these premises, ASUS or Dell would have provided a better solution to this specific task.

> if one board performes poor [...]

AFAIK, the reason for the poor memory performance of Cyrus/X5000 (clearly contradicting the specs of the memory controller) has not been determined yet. We just know it's not the OS as it's equally poor on Linux, OS4 and MorphOS. This leaves the SoC (memory controller), the board design or the U-Boot firmware as contemplable culprits.

>AFAIK, the reason for the poor memory performance of Cyrus/X5000 (clearly contradicting the specs of the memory controller) has not been determined yet. We just know it's not the OS as it's equally poor on Linux, OS4 and MorphOS. This leaves the SoC (memory controller), the board design or the U-Boot firmware as contemplable culprits.

Hopefully its a configuration or firmware issue that can be rectified in the future.
Obviously it can't be the memory, which is the same sort of DDR3 used in other machines.

If this is something that can be addressed, then its a little early to start throwing up red flags about it.
And we knew about the PCI-E limitations quite awhile ago.

> 4x 1x 1x can be good if will gave the best and lest issue compared to put
> bridge over bridge. if an user was need a pci for example there are many adaptor

While there're x4 x1 x1 PCIe configs for both P5020 and P5040, they are not compatible with each other for the rest of the SerDes assignments, so unfortunately none of these configs can be used for a common P5020/P5040 board.
And for a P5021/P5040 board, I think x4 x1 x1 without bridge would be really scarce: x4 for graphics card and x1 for Xorro leaves just one x1, which may have to be occupied by a PCI adapter in case of need. And if a bridge is needed anyway, it makes sense to use the max. number of native PCIe lanes (which is 8 with x4 x4 config, as opposed to just 6 with x4 x1 x1 config).

> right now on my x5000 im using only 4x 1x 1x

From SoC's point of view that's still x4 x4 :-)

Quote:

Could be. However, I think it has a lot to do with volume and economy of scale. A board made for volumes in the hundreds could only carry a rather limited cost for design, testing and post-launch support, despite a price tag of €2000-€3000 or so. Someone will always have to draw the line somewhere on how much time (read: money) that could be spent on optimizing, polishing the details, etc, as well as post-sale support such as firmware updates. Someone will sooner or later have to say "we could do more but this is good enough, spendings stops here". ASUS and Apple etc can set that "good enough" limit a lot further away for their products sold in hundreds of thousands(?), than Varisys could do for a product projected to be built in the hundreds. Mainstream and mass volume is good. That's why "Apple" and "ASUS" means a whole different dimension of quality compared to what "AmigaOne" could ever hope of...

> If this is something that can be addressed, then its a little early to start throwing up
> red flags about it.

The issue was known und mentioned in public already last year by me (in comparison to PowerMac G5) and others (in comparison to X1000):

"The X5000 [...] gets oblitrated by the X1000 when it comes reading and writing to memory"
http://morph.zone/modules/newbb_plus/viewtopic.php?forum=11&topic_id=11137&start=623

Has anybody from A-Eon or Varisys been investigating this since?

> we knew about the PCI-E limitations quite awhile ago.

What I didn't know until I looked it up today is that the P5040 is even way more limited than the P5020 in this regard. I knew that the P5040 had one controller less, but it adds 2 SerDes lanes ...just not for PCIe, apparently, as the P5040 can use only two thirds of the lanes for this compared to P5020 (8 vs. 12), or four fifths for SerDes configs that include SATA controllers (8 vs. 10). That's a shame and very unfortunate for a common board design, severely crippling the more capable (in terms of PCIe) SoC.

Quote:

reported by me in june 2016.
and reply was "all is all right"

[ Edited by tlosmx 14.09.2017 - 00:54 ]

If this is as good as it gets, there's obviously a significant underlying issue, as the memory itself should provide much higher bandwidth.

And the Qorlq line's focus on network capability, with it deemphasis of serdes connectivity is problematic.
If you were to use the e5500 or e6500 cores in a CPU designed with end user computing devices, you'd probably want to rework some of this to provide better support for other devices.

Still, from the recent benches I have seen Spectre660 post, the overall performance of the X5000 still measures up fairly well.

Quote:

So:

1. A Cyrus motherboard designed exclusively for the P5010/P5020 specs, equipped with the cheaper P5010 CPU, would have been the best option for OS4 in regards to price/specs.


2. Instead they used the more expensive P5020, making OS4 users pay extra for a CPU core they have no need or use for.


3. But not only that, the Cyrus board design itself is cripled (being less than what it could have been with 5020/5010) to be able to use the P5040, which is even more expensive and will have *three* unused cores in OS4, thus being even further away from what makes sense for OS4.

IMHO this rymes poorly with the statements that the "AmigaOne X5000 is exclusively built [being an "Amiga"] for OS4"