The Future of Pyra's CPU


ToastBucket

Very Active Member
Joined
Oct 12, 2010
Messages
330
Location
Seattle
What about the RAM that would be attached on top? We need to make sure to account for that.

-God Ginrai

RAM attached on top? You mean just the RAM on the SOC board? The height and width dimensions would affect that.

Memory isn't necessarily attached on top.

RAM is pretty much never attached on top? I've seen one weird custom spin of the i.MX6 which did this, but it was hugely problematic and eventually dropped due to yield issues.
 
Last edited:

EvilDragon

Administrator
Staff member
Joined
Mar 4, 2003
Messages
29,670
Age
43
Location
Ingolstadt
Pandora had, and OMAP5430 would've also had RAM popped on top.
But that's rare these days.
 

ToastBucket

Very Active Member
Joined
Oct 12, 2010
Messages
330
Location
Seattle
Pandora had, and OMAP5430 would've also had RAM popped on top.
But that's rare these days.

Ah interesting. Probably rare because it was really difficult and expensive. What seems a bit more successful is SIPs like those made by Octavo. The packages are definitely a bit larger, but its very convenient having RAM and PMIC and everything all buttoned up for you so long as you're happy with whats in the silicon.

 

EvilDragon

Administrator
Staff member
Joined
Mar 4, 2003
Messages
29,670
Age
43
Location
Ingolstadt
Well, the good thing about PoP was: You didn't need to worry about the connections from the RAM chips to the SoC.
That was quite a deal with the 4GB RAM on the Pyra.
 

ToastBucket

Very Active Member
Joined
Oct 12, 2010
Messages
330
Location
Seattle
Ya it’s usually common practice to directly replicate the RAM part selection and layout from the EVK, but still a small mistake can cause lots of problems. At least we have OCRAM to help us debug
 

Swordfish II

Advanced Member
Joined
May 20, 2015
Messages
1,163
Yes, but you don't need that to boot. You can use the system without them, you would just be missing these features.
Wifi / BT could be done with a USB stick, for example, and not everyone needs the SGX.

Yea I just wanted people to be aware that it is very difficult to be blob free
 

Djoga'Ro

moonstruck
Joined
Apr 3, 2016
Messages
1,762
Just a hypothetical question: Let's say an ideal SoC came along and it was a bit to high. Would it be feasible to "just" modify the CPU boards connector to get more room between it and the mainboard and modify the mold of the upper or lower half of the case to allow for that by making its wall taller? (As I understand it, adding to a molded part is the easier direction.)
 

levi

Still fresh, damnit!
Joined
Oct 6, 2008
Messages
13,749
Location
Somewhere off the coast of the EU
I think the battery connector is on the main board, not the CPU board. If you allow space for the CPU board to be thicker, you need to move the battery away from the main board, which might cause problems connecting power to the system. I guess the simplest fix if you can't find a taller battery connector, would be to put the battery connector on a little mezzanine board made out of 1mm PCB.
 

pyrat

Very Active Member
Joined
May 20, 2016
Messages
431
Btw, the 1.51mm height includes the BGA connections. See pages 28-29 of the datasheet for package measurements.
Yes, I thought we were all measuring from the PCB up ?
Yea I just wanted people to be aware that it is very difficult to be blob free
You're right, in fact I half remember the RK3399 needs a blob for the laptop screen and the IMX8 one blob for the HDMI ? I no longer remember the source, so I might be wrong. [EDIT: one source]
Also, the ARM Trusted Firmware sources include a blob for HDCP, but you can remove it and recompile and the RK3399 seems to be happy (not tried with protected content).
It's hard to live blob free but for me it's important.
[EDIT: I've just read (2021-02-05) someone that seems to know more than I claiming that IMX8M needs blobs for the DDR controller, so it can't use RAM without blobs. That's quite a show stopper for me. I didn't know it.
I don't know about other IMX8 models. The mention is in the last footnote, the rest of the article is interesting but unrelated ]
Probably the most notable SoC containing a Synopsys DDR4 memory controller and PHY is the NXP i.MX8M series, which was for example selected by Purism to form the heart of their Librem 5 phone. This was a strange choice for a product intending to have fully open source firmware, due to the aforementioned Synopsys firmware blob. Purism's response to this problem was an astonishing blog post in which it openly boasted about how it was going to game the FSF's RYF criteria to be able to ship this blob.

Someone did inform me informally that the RYF program intends to have none of this and will not certify the Librem 5, but this is just a rumour. Make of it what you will
 
Last edited:

EvilDragon

Administrator
Staff member
Joined
Mar 4, 2003
Messages
29,670
Age
43
Location
Ingolstadt
I haven't checked this myself, but Nikolaus thinks the connectors are available in different heights. So we should have at least 2.2mm available for an SoC (or 2mm including a heat spreader).
 

matzesu

Forum Addict!
Staff member
Joined
Oct 24, 2008
Messages
10,844
Age
36
Location
Germany,, Saarland, at home
I also thought that it may be an Issue whit the hight of a future CPU, and its cooling, unless whe want to make the Bottom Part of the Case a bit Higher so its more room for the CPU Board..
If the Battery connector have Springs in it this may work even if the room between Batterie and Main PCB gets farer..
So lets hope the Omap 5 is still available and whe dont need a new CPU Board in the future as everywhone is happy whit the Omap 5 ^^
 

ToastBucket

Very Active Member
Joined
Oct 12, 2010
Messages
330
Location
Seattle
I haven't checked this myself, but Nikolaus thinks the connectors are available in different heights. So we should have at least 2.2mm available for an SoC (or 2mm including a heat spreader).
Ahh I see what you mean. The SOC board connectors themselves have height differences. Itd be interesting to see if we can create any space by using a longer connector that is compatible
 

EvilDragon

Administrator
Staff member
Joined
Mar 4, 2003
Messages
29,670
Age
43
Location
Ingolstadt
Ahh I see what you mean. The SOC board connectors themselves have height differences. Itd be interesting to see if we can create any space by using a longer connector that is compatible

Hm? No, that's not what I mean :)

All SoCs have a variation in height. The OMAP5 is between 1.06 and 1.29mm thick, according to the datasheet.
That's the tolerance.

I grabbed piece of plastic with 0.8mm height and put it between the OMAP5 and the mainboard. It was a tight fit, but it did fit in there.

So the total size for an SoC we can put in there with our current board would be between 1.86 and 2.09mm (as I don't know how thick the OMAP5 is on the PCB I tried).

Additionally, the connectors are available with different heights. So we could add a bit more as well (as we have space of about 0.5mm between the CPU board and the battery compartment).

Taking everything into account and depending on what different heights the connectors are available for, we have at least 2.36mm height available (for the SoC as well as anything else like a heatspreader)
 

oskda

Very Active Member
Joined
Mar 29, 2015
Messages
313
I haven't checked this myself, but Nikolaus thinks the connectors are available in different heights. So we should have at least 2.2mm available for an SoC (or 2mm including a heat spreader).

As you stated in https://pyra-handheld.com/boards/threads/it-reminds-me-of-the-pandora.99197/post-1681984 OMAP5 board needed expensive 12 layer PCB:

The costs of the full CPU board (12 layer PCB....)...

*CPU board was a lot more complex than planned due to the OMAP5430 not being available. This added more layers and made the design 10x more complex

It could be possible with other SOC (not so complex like industrial/car oriented OMAP5 used in Pyra, as the other OMAP5 for mobile was not available) you could use a PCB with less layers so it could a bit thinner and cheaper.

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.

The RK3399 would fit then - at least the height...

Chinese RK3399 is a powerful SOC (I liked it a lot of when it was launched in 2017), but is its availability guaranteed for long time as some¹ NXP SOCs do?

A lot of NXP i.MX SOC are supported on NXP’s 10 and 15-year Longevity Program¹.

And more important: are support, drivers, open soft and DOCUMENTATION on Chinese RK3399 as good as on European² NXP?

Of course big i.MX 8 may be out for size, and even for generated heat. And that heat problem could be the same for RK3399.

We get crazy thinking on having 2 big Cortex-A72 cores (plus 4 efficient Cortex-A53 cores), BUT are we on real world? How many time, at what room temperature, can we maintain 2 28nm Cortex-A72 cores at full or good speed on a small case Pyra without heatsink on SOC and without a fan ???? That is the question.

Raspberry Pi 4 (quad-core Cortex-A72) is having a lot of problems and throttle down even with heatsinks impossible to fit on Pyra.

Think that in modern smartphones we have big cores, but they are not in 28nm. Even my high-end smartphones from 2016 had a SOC manufactured in 14nm, and today high-end smartphones go to 5nm. We are not on that race with available SOCs for Pyra.

If fabrication has not changed, RK3399 is manufactured on 28nm HKMG, while Quad-core Cortex-A53 i.MX 8M Mini is built using advanced 14LPC FinFE and can go up to 2 GHz. Going from 28nm to 14nm is a very big difference.

So it could be we end with Cortex-A72 on RK3399 going in little time too hot and throttling down a lot of, and consuming a lot of power, generating a lot of heat in Pyra and in your hands, while a much more efficient Cortex-A53 could consume much less power, generate less heat and maintain its normal speed all time or at least much more time.

On same uses you have no problem with big cores, because you use it a moment and they nearly stop (for example creating a text document), but if someone thinks "with that big core I can run xxx emulator" he may be wrong, because you will need high power all time for that emulator and that could be impossible.

It may be a 2 GHz quadcore Cortex-A53 i.MX 8M Mini (14nm) would be good compared to our actual dual core Cortex-A15 OMAP5 at 1.5Ghz, because this OMAP5 would throttle down earlier.

When a lot of ordered Pyra be sent, and summer/hot arrives, we will know how much time 28nm Cortex-A15 can run at full (or even turbo) speed (and how much battery drained that way). 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.

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 would reconsider NXP SOCs for all the aforementioned points. And more important, I would check power consumption and temperature/throttling of RK3399 (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.

Please consider that for example Pine uses RK3399 on Pinebook Pro SBC 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...

It may be so simply RK3399 is designed more for SBC and similar projects, but no for mobile/handheld devices like for example pocket computer Pyra.

On other side in-order Cortex-A15 is immune to Spectre bugs, while out-of-order Cortex-A15 and Cortex-A72 are affected by Spectre bugs.

¹: For example i.MX 8 and i.MX 8M family are on «10 and 15-year Longevity Program».

²: i.MX line was originally from Freescale Semiconductor (now part of NXP).

PD: NXP i.MX 8M Mini doesn't have HDMI interface (ED mentioned it earlier), so it isn't an option and we would need standard 8M (not Mini) or upcoming 8M Plus. 8M Plus will be made on 14nm process according to NXP site. And NPU (Neural Processing Unit) in 8M Plus could be an interesting point for us and for Pyra: first pocket computer with Neural/AI hard unit? :D (even if not being first, it would be very nice, because how much pocket computers with neural/AI hard do exist?)
 
Last edited:

Askarus

Forum Addict!
Joined
Sep 28, 2011
Messages
4,569
Location
Germany
Is till wonder why there is no Rockchip with a more recent Cortex.
A72 is pretty old now and I'm waiting for an A75 SoC.
 
Top