The Future of Pyra's CPU


Looks like they are about to release an A76, late this year:


It could be Rockchip RK3399 at 28nm is not an option for Pyra. It is not a SOC designed for mobile/handheld devices.

Future Rockchip RK3588, if done at 8nm, could be an option. But for the moment it is too much beautiful :)

The RK3588 succeeds the RK3399Pro as flagship SoC. It's expected to be available in Q3/Q4 2020.[60]
  • CPU – 4x Cortex-A76 and 4x Cortex-A55 cores in dynamIQ configuration
  • GPU – Arm "Natt" GPU
  • NPU 2.0 (Neural Processing Unit)
  • Multimedia – 8K video decoding support, 4K encoding support
  • Display – 4K video output, dual-display support
  • Process – 8 nm LP

At 8nm it would be really interesting. Too much beautiful to be real soon :)

And of course, we don't know if it would fit on Pyra CPU PCB space.
 
Last edited:
From https://www.nanotech-elektronik.pl/index.php/en/info/stackup I can see a 12 layer PCB can be 1,6mm in best case, while for example a 8 layer PCB can be 1,0mm thick (best case), so it could be 0,6 mm difference between them. And that could add up space for a bit thicker SOC chip. A 10 layer PCB can be 1,3mm thick, so it would be only 0,3mm difference, but even in this case it adds up space.

Since we are already splitting hairs (almost)... I think the PCB would end up centered on the socket, so if you have a board 0,6mm thinner, you'd only get a 0,3mm extra margin (on each side, but your SOC will sit on only one side, so you can only replace it with one that is 0,3mm thicker, a soc 0,6mm thicker won't fit even if the PCB was 0,6mm thinner).
I'm not so sure how it works, though. It's very small numbers and I guess you need some margin to be able to insert and extract it (maybe you're counting it's the same existing margin, whatever it is).
But I doubt you can so easily reduce number of layers just by choosing a different (and more powerful) SOC.
The way to reduce layers would be to reduce components in the board. Maybe if you could get fewer RAM chips for the same RAM, or you choose to leave out eMMC. I don't know.
If you go for narrower buses you should get worse performance (and if you go fewer RAM chips possibly too, it may depend).

Btw. What was the comment about ? OMAP5430 had POP RAM (so no need to add SOC-RAM traces to the board, which must be a lot of them) and OMAP5432 needed the RAM chips on the PCB, so more traces ?
O wasn't that the difference?
I've heard current SOCs don't use POP anymore, but anyway, wouldn't the stack SOC+RAM be even taller than the current OMAP5432 ? We'd also have more problems with heat, I suppose.
POP should also be harder to solder, I have heard, so possibly lower yield?.

And more important: are support, drivers, open soft and DOCUMENTATION on Chinese RK3399 as good as on European² NXP?
Having PineBook Pro and RockPro64 inter alia should help there.
I think it needs a blob for eDP, but I think Pyra uses MIPI DSI, so maybe we're lucky ?
I know a RockPro64 can boot and use a desktop on HDMI, with browser, display videos, play sound, etc. with a uBoot that should be now mainline, an ATF without the hdcp blob, linux-libre and debian 10.
I even tried a while without a fan (but in open air, not enclosed). Once I mounted a fan on the SOC (it covered the SOC and most part of the RAM chips) I had to make it play 3 internet videos at a time in overlapped windows
to get it to turn on (I don't remember the threshold temp now). I'm crappy like that at benchmarks. But I don't have it anymore.
I think I remember documentation is available, but maybe not so complete as NXP.

When a lot of ordered Pyra be sent, and summer/hot arrives, we will know how much time Cortex-A15 can run at full (or even turbo) speed (and how much battery drained that way).

Or if some of the first preorderers are in the boreal hemisphere ?

By other way we could get a RK3399 SBC without fan nor heatsink and we could encapsulate it in a small box to simulate in some way Pyra environment, and we could see power consumption and sustained speed, how much it would throttled down, etc.

Err... I wouldn't know how to begin to try that. I mean how do I know the environment is similar enough ? Or what to test ?
I know Pine64 recommends using a fan or a big dissipator on the RockPro64. My intuition is that it would get hot on the Pyra.
How much or how usable it'd be temp-constrained, I have no idea.

About NXP SOCs I have seen more open aware hardware on ARM is selecting NXP:
I think using so NXP hard Pyra could attract not only retrogammers and some Linux users, but also free minded people looking for less closed hard/soft. And it would be another selling point for Pyra.

I don't know iMX8M but I used to like i.MX6Q, so yes, I have good vibes for NXP.

That MNT Reform page you linked warns that
HDMI port (on i.MX8M, requires an optional HDMI TX firmware blob to work)

But otherwise seems very nice.

I would reconsider NXP SOCs for all the aforementioned points. And more important, I would check power consumption and temperature/throttling of RK3388 (28nm?) in an environment similar to Pyra (no heatsink on SOC, no fan, small and near closed box) in front of 14nm Cortex-A53 NXP SOC. It would be bad we end with a burning hands while holding a Pyra and with very slow Cortex-A72 after severe throttling down in hot summer.
Is RK3388 a typo or some chip I don't know ? Did you mean RK3399, RK3288, or RK3588 ?


Please consider that for example Pine uses RK3399 on Pinebook Pro but on Pinephone and Pinetab they use an Allwinner Quad Core Cortex-A53 SOC. So they think that RK3399 isn't optimal for handheld devices: heat, power consumption, no heatsink, no fan...
Yes, but their Pinephone is thinner than a Pyra, isn't it?
And how does the enclosure of a PineBook Pro compare with a Pyra ?
That's hard for me to understand.

I don't know. I'd like RK3399 or i.MX8M better than OMAP5432, but I don't know how feasible that would be, and what I know is that we now have an OMAP5432.
First we need to use and develop for this.
In a future one can see.
But in the future the available SOCs will be other.
If we start choosing SOC too early, then we can block future decisions before we need to.
Once the Pyras are sold, investment recovered, and EvilDragon, or someone decides to take the risk of a new CPU Board, it is then the right time to see what SOCs are available and which are feasible and which one we like better.
 
The OMAP3 had PoP RAM, the OMAP5430 did not. But the OMAP5430 used LPDDR2 RAM, which permits a memory bus of 32 or even 16 bit instead of the full 64 bit, which has to be used for OMAP5432's DDR3 RAM.

No, the OMAP5430 had PoP memory. The OMAP5432 is the one that does not.

Texas Instruments said:
The OMAP5430 offers a smaller 14x14 mm2 package with PoP LPDDR2 memory, while the OMAP5432 offers a 17x17 mm2 package with non-PoP DDR3 memory.
Source: https://www.ti.com/pdfs/wtbu/SWCT010.pdf

-God Ginrai
 
I think the PCB would end up centered on the socket, so if you have a board 0,6mm thinner, you'd only get a 0,3mm extra margin.

I am afraid you are wrong: CPU board sits on main board "vertical" connectors, so all CPU board thickness you can reduce you gain it on top.
Mainboard - Pyra Wiki
CPU-Board - Pyra Wiki

You would be right if secondary board (CPU board) were not directly connected/sat on main board (vertical connector) but sitting on fixed grooves on case.

The way to reduce layers would be to reduce components...

Btw. What was the comment about ? OMAP5430 had POP RAM (so no need to add SOC-RAM traces to the board, which must be a lot of them) and OMAP5432 needed the RAM chips on the PCB, so more traces ?
O wasn't that the difference?

If you read my message you can see I cite ED comment. He said they had to use expensive 12-layer board because having to use complex OMAP5 (designed mainly for car use) instead of simpler OMAP5 for mobile/cellphones (it was cancelled). So with other "normal"/mobile SOCs less layers are needed, and production plus development is cheaper.

Or if some of the first preorderers are in the boreal hemisphere ?

In that hemisphere they will reach summer solstice next month, and I think most Pyra sales are on northern hemisphere. It may be we are lucky if someone living on hot place on southern hemisphere get Pyra before its summer fade out.

Is RK3388 a typo or some chip I don't know ? Did you mean RK3399, RK3288, or RK3588 ?

It was a typo, now corrected. I wanted to say 3399.

Yes, but their Pinephone is thinner than a Pyra, isn't it?
And how does the enclosure of a PineBook Pro compare with a Pyra ?
That's hard for me to understand.

A lot of Pyra thickness came from keyboard+gaming controls, very big battery (6000 mAh vs 2800mAh on Pinephone), two boards one on top of other (and other board under display), shell design, lots of connectors including a standard USB, 2 standard size SD slots (in smartphones you usually have 1 microUSB/USB-C, 1 mini jack and 1 microSD), etc.

Pinebook Pro is a 14 inches display notebook and it have a magnesium alloy case (I don't know thermal conductivity and there are different Mg alloy composition; I have had Thinkpad with Mg alloy cast and I think it transfer heat better than plastic, but worse than pure aluminum). It has a lot of space to spread heat. And in this picture you can see RK3399 on Pinebook Pro surrounded by big metal plate, and in this review you can see a thick thermal pad over that chip, so I suppose it will be thermally connected to a very big Mg alloy case (I suppose that thermal pad is thick so it touch case).

Pyra has a small plastic case, and plastic is more an insulator than a thermal conductor :)

First we need to use and develop for this [OMAP5].
In a future one can see.
But in the future the available SOCs will be other.
If we start choosing SOC too early..

This thread is about Future of Pyra's CPU :) So it is about future and speculative on itself. On other side I suppose ED team will be looking for future CPU much earlier before we knew that final decision so they don't suffer an Osborne effect, and they will be working on that CPU upgrade design a lot of time before we know, and a lot of time before it reaches market.

too early: For Pyra? Are you joking? :D I think next CPU board will be much easier than the nightmare they lived with plastic parts and original OMAP5 board and other parts.
 
I am afraid you are wrong: CPU board sits on main board "vertical" connectors, so all CPU board thickness you can reduce you gain it on top.
Mainboard - Pyra Wiki
CPU-Board - Pyra Wiki

You would be right if secondary board (CPU board) were not directly connected/sat on main board (vertical connector) but sitting on fixed grooves on case.

AFAIK the die of the OMAP5 goes on the side of the connectors, that image on the wiki is not populated. So you have <CPU board> - <OMAP5 die> - <heat spreader with glue & insulation> - <gap> - <Mainboard board>. If you reduce the board thickness I suppose it changes nothing, as the die and connectors sit on the same plane.

Here is an image populated:

 
I am afraid you are wrong: CPU board sits on main board "vertical" connectors, so all CPU board thickness you can reduce you gain it on top.
Mainboard - Pyra Wiki
CPU-Board - Pyra Wiki

You would be right if secondary board (CPU board) were not directly connected/sat on main board (vertical connector) but sitting on fixed grooves on case.

Ok. I didn't see it.
I was thinking of a different connector than the ones used there. Sorry.
If @RZR is right then cpuboard PCB width doesn't matter anyway. I can't see that twitter image (I should disable too many adblockers) but I understand what he means.

If you read my message you can see I cite ED comment. He said they had to use expensive 12-layer board because having to use complex OMAP5 (designed mainly for car use) instead of simpler OMAP5 for mobile/cellphones (it was cancelled). So with other "normal"/mobile SOCs less layers are needed, and production plus development is cheaper.

Sorry I wasn't clear. I meant to ask about why ED said that. @Letalis Sonus answered me perfectly. Thanks.
I still think a new CPU board would be likely 12 layers if not more. At least it won't be fewer tracks.
We won't go back to 32 bit buses, I think. That would kill a lot of performance.
I don't know if there's some way to produce thinner tracks or something and then maybe fewer layers, but the complexity will remain.
It's a lot of signals in a small surface.
But I'm no expert.

In that hemisphere they will reach summer solstice next month, and I think most Pyra sales are on northern hemisphere. It may be we are lucky if someone living on hot place on southern hemisphere get Pyra before its summer fade out.
I feel if I know speak of the tropics I'd be only prolonging a pointless discussion. I'd better shut up...:p

Pinebook Pro is a 14 inches display notebook and it have a magnesium alloy case (I don't know thermal conductivity and there are different Mg alloy composition; I have had Thinkpad with Mg alloy cast and I think it transfer heat better than plastic, but worse than pure aluminum). It has a lot of space to spread heat. And in this picture you can see RK3399 on Pinebook Pro surrounded by big metal plate, and in this review you can see a thick thermal pad over that chip, so I suppose it will be thermally connected to a very big Mg alloy case (I suppose that thermal pad is thick so it touch case).

Pyra has a small plastic case, and plastic is more an insulator than a thermal conductor :)
Ok, I didn't know all that. Thanks. Yes, the Pyra should be more problematic with heat than the Pinebook Pro.

This thread is about Future of Pyra's CPU :) So it is about future and speculative on itself. On other side I suppose ED team will be looking for future CPU much earlier before we knew that final decision so they don't suffer an Osborne effect, and they will be working on that CPU upgrade design a lot of time before we know, and a lot of time before it reaches market.

You have a point. Maybe I should just leave this thread to be coherent.
In any case, if ED is already considering starting another CPU board and not saying it yet, for me it's fine, but I'll act just the same as he wasn't starting that anyway.
If/when he wants any opinion he'll ask.

too early: For Pyra? Are you joking? :D I think next CPU board will be much easier than the nightmare they lived with plastic parts and original OMAP5 board and other parts.

I don't think a new CPU board will be easy or fast, but it always depends on the budget. Of course the plastics won't have to be revisited just for a new CPU board.

But it doesn't matter, I'll see myself out. Thanks for your corrections.
 
Reportedly, he's not starting even thinking about it really until he's got the original pyra including it's OMAP5 CPU board out to preorderers. If we can come up with a happy solution for him here before then, then he might well read that and investigate supplies and costs and prototyping and stuff like that.
 
Where's the RAM socket on that board then? It's either that chip below the SPI port or on the opposite side of the board, or it's POP on top of the CPU. It might be that chip on the opposide side of the board since I see a lot of VIAs across from the CPU (under the wifi module) and lines going across in the vague direction of that chip on the back, but I'm not sure why they have to go around all of those passive modules if that's the case. I think it's more likely a EMMC chip. That chip under the SPI port is a candidate but if so I'm not sure why it's labelled FTDI which as far as I'm aware are mostly creating usb access chips.
 
The OMAP5430 may not have been available for purchase, but prototype boards did exist - and none of the ones of which I can find pictures of has any kind of PoP. Take this one as an example: http://www.embedded.rs/sites/default/files/EPP-Pico-OMAP5430_Datasheet.pdf

Every reputable source I can find that discusses the OMAP5430 and the type of its memory says it is PoP. Also, the data sheet you linked doesn't specify whether its memory is PoP or non-PoP. However, I did find a board using the OMAP5430 that does specifically mention that its memory is PoP: https://web.archive.org/web/20130607043405/https://www.seco.com/en/item/uq7-omap5/

-God Ginrai
 
Also, the data sheet you linked doesn't specify whether its memory is PoP or non-PoP.
There are pictures of the board, without any kind of heatsink hiding details. PoP is very easy to spot, it would require a single RAM chip of the exact same dimensions and a close to zero sized layer of solder between them for being that hard to recognize in these pictures. It's possible that TI initially planned to create a PoP design (or at least offer it as an option in a different chip package), but ended up not doing, with many old documents remaining unchanged. These kind of things happen all the time and are often blindly copied by the guys creating other documents about things using it.

Edit: Okay, according to some old documents the PoP design actually includes holes on the top side of the chip for the top contacts, which would allow the RAM chip to sit on top without leaving any gap for the solder inbetween, that might explain why it's not that obvious. For reference: the same document claims a max heigth of 0.81mm for everything: OMAP, RAM, and the solder between OMAP and PCB.
 
Last edited:
AFAIK the die of the OMAP5 goes on the side of the connectors...

The board can be populated on both sides, and I don't know if there is a limitation the CPU can go on the other side.

Other possibility could be to use elevated connectors for this board, so in this way you get gained space between both boards (where you suppose CPU is placed) instead of on upper side.

Raised connectors are available in different heights.

PD: I don't know if we have too much imagination to be feasible :D
 
Last edited:
In any case, if ED is already considering starting another CPU board and not saying it yet, for me it's fine, but I'll act just the same as he wasn't starting that anyway.

I think your and hers is correct acting way. If you have ordered Pyra there is no reason to change it: you will get an enjoy a Pyra with OMPA5, and the day CPU board be on sale you can upgrade it. I see a win-win: you enjoy much earlier and upgrade later (possible before other people get later Pyra with new board inside), no problem.

At the same time the best for Pyra is this way so that sales can go fluid as needed.
Post automatically merged:

You are so wrong.
This thread is about writing big letters.

Sometimes it is needed to write big letters to show important or core question. In normal/low letter it can get lost :)
Other times I use different colors, underlying, etc.

But thanks for the joke :D Laughting is very important :)
 
Back
Top