The Case for the Pi3A+

[W. H. Heydt]

Good points.

Even if they used the older BCM2837 and the older WiFi package, it would be a huge step up from the A+. Or a downclocked BCM2837B0. It wins either way.

Yeah... Sigh... That would be a Pi3A...which was talked about by Eben Upton when the Pi3B was launched as something that would be launched 'later [that] year, after the CM3'. Didn't happen, obviously. Personally, I think that it would make sense to upgrade the CM3/CM3L and Pi2B to use the BCM2837B0 for the same reason that the Pi2B was moved from the BCM2836 to the BCM2837 in the first place. It's just that I think the Pi3A+ deserves to go first because it has been so very long (in terms of Pi development) since the A+ launch, that the Pi3A+ should be at the "head of the line" for this upgrade.

Tempering that with the need to keep up with Pi3B+ demand (which is probably eating up every BCM2837B0 being made), I could see it being late this year or early 2019 before a launch is even possible.

And taking this opportunity for an aside... I want to thank everyone who has commented in this thread--and especially those who don't think a Pi3A+ is needed--for keeping the thread mostly on topic, reasonable, and calm.

[Lob0426]

+1

[hippy]

I am wondering how the development of the Pi 4 being actively pursued changes things ?

A Pi 4 may not arrive for a year or so but it could make far more sense to produce a 4B and a 4A rather than create even further divergence, divert efforts.

A consolidated range has a lot going for it. There could be a full fat 4B+, a 4B without integrated WiFi/Bluetooth, a 4A+ with reduced footprint, with WiFi/Bluetooth, but no LAN, single OTG USB, and a 4A without WiFi/Bluetooth, without LAN, single OTG USB.

The "+" becoming a designator for having WiFi/Bluetooth, "B" with LAN and multiple USB, "A" no LAN, single OTG USB, smaller footprint.

Pi 4B+ : Replacement for Pi 3B+ ( and Pi 3B )
Pi 4B : Replacement for Pi 2B ( and Pi B+ )
Pi 4A+ : Replacement for proposed Pi 3A+, an alternative to using Pi Zero W
Pi 4A : Replacement for Pi A+, an alternative to using Pi Zero

The 4B+ and 4B could be the same boards just with different parts fitted ( much like the original A and B variants ), and likewise for 4A+ and 4A. In fact all variants could use the same single board if there were no concern about footprint, though I suspect the desire for reduced footprint boards would make the effort and cost of delivering those worth it.

I would actually favour using "W" as a suffix over "+", keep the "+" for some other potential upgrade, but doing that above would have clouded the replacement mapping more than necessary.

And there would probably be a CM4 and CM4L as well.

[W. H. Heydt]

I am wondering how the development of the Pi 4 being actively pursued changes things ?

I don't think it changes anything outside the walls of Pi Towers. Saying "wait for the next iteration" is endless and gets one into a "L'Arliesienne" or "Waiting for Godot" scenario.

A Pi 4 may not arrive for a year or so but it could make far more sense to produce a 4B and a 4A rather than create even further divergence, divert efforts.

Well... We've already been through a wait for a Pi2A and a Pi3A. How many more iterations should people wait? The A+, using the BCM2835 is already pretty "long in the tooth" and is, I think, past due for an upgrade. Had it happened, a Pi3A would have been a marked improvement by adding WiFi and BT, thus making the limitation of single USB port less of an issue. Unless there were to be some really important (to the A Series) feature on the Pi4 SoC, I can't see waiting for a Pi4A. And, FYI, I wouldn't consider more than 1GB RAM or USB 3 something to wait for in an A Series Pi. I'll spot you a "maybe" for a 28nm part, though.

And there would probably be a CM4 and CM4L as well.

That's not clearly a likelihood. Bear in mind that there never was a CM2. I think a more likely scenario would be CM3+/CM3L+ fitted with the BCM2837B0 package, and I'm not sure that would actually work since the CMs use a small board and it might not be enough heatsink.

The problem that is a distinct possibility with a CM4/CM4L is if USB 3 is part of the Pi4 SoC. I suspect that would require either modification or a new specification of the edge connector data, and that would mean redesigning every carrier board in existence to use it. That would be a major incompatability step.

On the more general topic...the big jump from BCM2835 to BCM2837 is the change from ARMv6 to ARMv8, making the newer SoCs able to run in 64-bit mode. Whatever is done for the Pi4 SoC, that is unlikely to change. The *real* trick is going to be to find a way to fit a 64-bit-capable SoC to the Pi0/Pi0W. I don't see that happening for quite a while, given various known constraints.

tl;dr No, I don't think that waiting for the Pi4 SoC to upgrade the A+ is worth waiting a minimum of two more years.

[hippy]

Well... We've already been through a wait for a Pi2A and a Pi3A. How many more iterations should people wait?

The bottom line is we'll have to wait as long as it takes. Think yourself lucky you have not been waiting for a Parallax Propeller 2 chip

No matter what anyone wants, no matter how desperate their pleas to have it now, it is RPT who will ultimately decide. Will they create a Pi 3A+ alongside developing the Pi 4 or rather create a Pi 4 range including something which fits that bill ? I don't know. But If I were doing it I would likely select the second path.

I suspect there's going to be some degree of break in compatibility in the move to the Pi 4 but I think it's best just to suck that up and get on with it. It likely won't be as bad as the initial shouting would have it, even if there is radical change.

The Zeroes present a problem. I can't see a full Pi 4 version arriving, and not at the $5 or $10 pricing, so they either have to keep that alive as it is, let it die a death, or maybe create a variant of a a Pi 4 chip which fits the bill.

Whatever they do we'll just have to wait and see.

[Lob0426]

A Pi3A would most likely be with a BCM2837 or a BCM2837B0. Reason? Extends the lifetime of that processor series.

To get inexpensive boards you cannot use the latest and newest package. You use the older packages. If they had hypothetically built the Zero on the Pi2 BCM2836 (hypothetical because TDP is too high for small board area) the Zero would have been a $20 board, not a $5 board. The A+ was a $25 board because it was using the same BCM2835 package as the B+. If the A+ came out today it would probably be $15 to $20 because the PCB size, full size components, component count and type.

I was surprised that the Zero/ZeroW actually have a more advanced processor package at their price. We can probably attribute that to newer manufacturing and the success of the Foundations design. Early on the foundation said it was very unlikely that Broadcom would be able to design a cost effective upgrade to the BCM2835. They were searching for the next processor. Then the BCM2836. Later 2B's were built on the down clocked BCM2837 also. Cheaper to use the same processor than to have two separate packages. Especially when they are at the same price point.

But sales reached 19 million in march this year.

I think you will see a Pi3A+, not a wait for a new iteration of processor. There is a rumored prototype. I don't see them wasting that work and waiting. But they did just release the upgraded Pi3B+. They will at least wait until sales of that have settled down before they release anything. They do have to decide whether to use the older WiFi and processor or bring it up to the newer parts. Some advantages to bringing it up if they foresee market sales to manufacturers.

I do not think it will have as limited of sales as the A+ had. The addition of WiFi/Bluetooth and a multicore processor package would make the 3A+ even more flexible than a ZeroW.

[alphanumeric]

I'm happy if with any delay to make use of the new 3B+ bits. A 3A+ is fine by me. Or just a 3A. Either way I'll be buying at least two.

[W. H. Heydt]

Well... We've already been through a wait for a Pi2A and a Pi3A. How many more iterations should people wait?

The bottom line is we'll have to wait as long as it takes. Think yourself lucky you have not been waiting for a Parallax Propeller 2 chip

No matter what anyone wants, no matter how desperate their pleas to have it now, it is RPT who will ultimately decide. Will they create a Pi 3A+ alongside developing the Pi 4 or rather create a Pi 4 range including something which fits that bill ? I don't know. But If I were doing it I would likely select the second path.

Even if the RPT/RPF *do* move foreward with a Pi3A+, I don't expect to see it until either late this year or early next year. Between fab work and board manufacturing capacity, I don't think there are enough BCM2837B0 chips or time on the production lines to make Pi3A+ Pis for some months yet. Plus, one of the perennial problems with Pis is anticipating actual sales. The demand always seems to be rather higher than expected--something I don't expect to change in the foreseeable future--and there just aren't enough SoC packages to go around.

I suspect there's going to be some degree of break in compatibility in the move to the Pi 4 but I think it's best just to suck that up and get on with it. It likely won't be as bad as the initial shouting would have it, even if there is radical change.

From the user perspective, in the Pi4B, I don't think so. There will be internal software issues, of course, but that shouldn't be apparent when using a Pi4B. A CM4/CM4L, though, is a different kettle of fish because you have to match the "pin" assignments on the board to the carrier.

The Zeroes present a problem. I can't see a full Pi 4 version arriving, and not at the $5 or $10 pricing, so they either have to keep that alive as it is, let it die a death, or maybe create a variant of a a Pi 4 chip which fits the bill.

There are two fundamental issues with the Pi0/Pi0W. The first is manufacturing cost, and by derivation, the SoC cost. R&D for the BCM2835 is all sunk cost now, but that's probably not the case for the newer SoCs. The other is the manufacturing cost constraint of having all soldering on one side of the board, which implies the need for PoP RAM. As a result, the "path forward" that could work, if backed by someone with very deep pockets, would be low power (because of the insulating effect of the PoP RAM) SoC that takes PoP RAM. If someone paid to have the chip developed without looking for a return on investment--that is, as a *gift*--a suitable SoC could probably be made. What might be reasonable would be a single core Cortex-A53 design. That might hold the TDP in check and still give the Pi0/Pi0W 64-bit capability. It's a case of "first catch your 'angel'".

Whatever they do we'll just have to wait and see.

This is true, but at least for the next gen, Eben did ask for posts of what people want, so I don't think other suggestions are out of line, so long as one sticks within the realm of actual possibility (like using existing SoCs).

[alphanumeric]

so long as one sticks within the realm of actual possibility.

This has been one of the longest such threads I can remember that didn't get locked. IMHO, mainly because its been constructive comments. Its a nice change from the usual unrealistic wants etc. "It would be nice" goes a long way over the "I want".

[Gavinmc42]

Yep most old grumps are only using Pi's until Prop2 shows up
Hmm Prop2 PCB on same footprint as a 3A+? Which will be first?

A break at Pi4 is no problem, there is a history of that. USB3/C/Sata will dictate it anyway
If the Pi4 comes out with the VC5 as the top of the line, cost amortization should then allow trickle down to something like the Zero+ A35/VC5.
A35 is a smaller die size than A53, but it is still 64bit, it will help lower the silicone cost, low enough ????

Production qty's will not decrease, so at some point 64 bit will out number 32 bit.
But that's all PFU (coloured horned pony)speculation, a 3A+ could get some market share.

How much share?
There is also considerable engineering to be done,the 3B+ was the biggest engineering effort so far.
The 3A+ will need nearly as much effort, much more than the A+.

To make computing and digital making accessible to all through providing low-cost,
high-performance single board computers and free software.

At what point is the engineering cost of doing a 3A+ too much or too soon?

The actual design layout could already be done and perhaps a few prototypes are floating around for software testing.
It is a big step then to pushing the button to mass produce them with unknown market demand and perhaps not even enough chips to go on them as they are all being used on the 3B+?

A Pi4 is going to need even more engineering effort and much more money, how to pay for that and still make low cost SBC's?
High Performance is the other key criteria. The 3B+ was probably an order of magnitude more difficult than the 3B.
A Pi4 is that again, compared to the original Pi B, three orders of magnitude? One thousand times harder to do?

Yet that is the Mission - low cost, high performance SBC's.
Probably some late nights at RPT

I have yet to use an A+ in any products, will 3A+'s just sit on the bench too?
If it comes with the RF combo, probably not
It should be good for non wired, even mobile apps like robotics which needs the cpu power 4 cores adds.

[Lob0426]

Yeah no one is asking for a new processor, or USB3.0, Lightning or a SATA. And no one has asked for 16GB RAM either! LOL!

I ran an A+ with a two port mini hub ("CUTE" mini USB 2.0 Hub) with Octoprint on a 3D printer. It worked fine as a web interface only. But mine had the 256MB memory not the 512MB. So when I tried CURA slicer with it, it was not up to it. I then went to a B+. That worked OK. Next up was a Pi2 when they came out. No comparison when slicing. I found the WiFi was smoother on the A+ and the external hub, rather than the LAN9512 on the earlier B or the LAN9514 on the B+.

I also ran it on a Motorola Lapdock. No problems with the onboard USB hub. Ran pretty well, but not as well as the Pi2B that is on the Lapdock now. A Pi3A+ would be perfect on a Lapdock.

[W. H. Heydt]

Production qty's will not decrease, so at some point 64 bit will out number 32 bit.

That is either already the case, or soon will be. The Pi3B has been out for 2 years and selling 5 million units per year, so half of the 20 million Pis sold so far are Pi3B boards. That doesn't count the Pi2Bv1.2 (also 64-bit) or the CM3/CM3L boards (likewise).

But, to a large extent it is moot for now. The ongoing sales of BCM2835 based Pis means that a migration to a 64-bit default OS is years away. Moving the A Series to a 64-bit SoC in the form of a Pi3A+ will be a step towards that. Then the only remaining boards going into the market in quantity will be the Pi0/Pi0W...and getting those to 64-bit capable is going to be a very hard nut to crack.

I have yet to use an A+ in any products, will 3A+'s just sit on the bench too?
If it comes with the RF combo, probably not
It should be good for non wired, even mobile apps like robotics which needs the cpu power 4 cores adds.

Do you remember when 100K+ A+ boards were snapped up and they became hard to find? Those boards went into *commercial* use, primarily turning flat panel monitors into "smart" displays. Just because you have not done a product using the A+ doesn't mean that no one else has, either.

[hippy]

The ongoing sales of BCM2835 based Pis means that a migration to a 64-bit default OS is years away. Moving the A Series to a 64-bit SoC in the form of a Pi3A+ will be a step towards that. Then the only remaining boards going into the market in quantity will be the Pi0/Pi0W...and getting those to 64-bit capable is going to be a very hard nut to crack.

If we assume that it's not going to be feasible to make a 64-bit Pi Zero in the immediate future; it could make sense to stockpile BCM2835's and ring fence them for the Zeroes, move everything else to whatever SoC a Pi 4 uses, replace the old product range with Pi 4 variants as previously suggested. That could move everything to 64-bit sooner rather than later.

Going forward with a consolidated range, all 64-bit, all based on the same SoC, except the Zero 'runt', has its advantages.

Stock of currently used SoC's - other than for the Zeroes and CM's - could be wound-down, held on the shelves for commercial or on-off bulk builds of older kit, or even written-off. That might seem costly, may be costly, but it might not be too costly to entertain.

That's not to suggest older kit won't continue to be supported but often there comes a time when a new era starts and the past has to become less relevant. One cannot always afford or allow oneself to be held ransom to history.

But it's all crystal ball stuff. We don't know what Eben or RPT are thinking and they aren't currently saying.

[W. H. Heydt]

Going forward with a consolidated range, all 64-bit, all based on the same SoC, except the Zero 'runt', has its advantages.

With the sole exception of the A+ (aka "A Series"), that is the case now. No need to wait for a Pi4 to appear. A Pi3A or Pi3A+ would mean that there is the same SoC (bcm2837) available for all lines: B Series (Pi2Bv1.2, Pi3B, Pi3B+), CM series (CM3, CM3L), and--if launched--the A Series (Pi3A or Pi3A+).

[ejolson]

Going forward with a consolidated range, all 64-bit, all based on the same SoC, except the Zero 'runt', has its advantages.

With the sole exception of the A+ (aka "A Series"), that is the case now. No need to wait for a Pi4 to appear. A Pi3A or Pi3A+ would mean that there is the same SoC (bcm ki2837) available for all lines: B Series (Pi2Bv1.2, Pi3B, Pi3B+), CM series (CM3, CM3L), and--if launched--the A Series (Pi3A or Pi3A+).

Whether called a 3A+ or something else, a quad-core 64-bit SBC that does not have a USB hub in front of the USB port is interesting, in my opinion, because of the possibility of using it in gadget mode. The Zero, for example, can connect just like a mobile phone by using the same USB cable for data and power when running in gadget mode. This works because the Zero runs fine with the 0.5 amps available from a standard USB port.

While a new lower-power quad-core SOC may appear sooner than people expect, if using the same SOC as the 3B+, the greater power requirements imply that it couldn't be back powered over the data cable from a standard USB port in the same way the Zero currently can. However, USB C cables and ports are designed to provide more power. Even though the data rates might still be limited to USB 2 speeds, fitting a Model 3A+ or whatever it is called with a USB C socket could allow using a single USB C cable for data and power while running in gadget mode in a way similar to the current Zero. Therefore, I would suggest a USB C socket is a natural choice for a quad-core SBC that does not have a built-in USB hub.

[Gavinmc42]

Therefore, I would suggest a USB C socket is a natural choice for a quad-core SBC that does not have a built-in USB hub.

Great point, but probably that means some sort of power ID interface?

Using a 3A+ as USB boot smart camera will need much more power than 500ma.
The 3A+ could be used for all sorts of USB C gadgets

Google time
https://en.wikipedia.org/wiki/USB-C
HDMI on the same connector?
Up to 3amps power?
Does the 3B+ power chip do USB-C?

Ouch, more engineering effort, well I guess those RPT guys need something to do now that 3B+ is out.
Perhaps the 3A+ could be used as USB-C tester?
After all on the A+ the single USB connector is the biggest connector, but it is not rated for enough current?

A single USB- C connector that does boot, power and HDMI?
Hmm that method could shrink the Zero too, leaving room for a DSI connector

[jamesh]

Much as I don't like weeing on peoples fireworks, any new A+ is quite a way away. Like, a REALLY long way. We do have a lot of development work to do in a number of different areas, so spending time on a new 3A+ is not considered particularly cost effective.

The Zero has to some extent taken over from the current A+ with commercial customers. Smaller, similar performance, cheaper.

[Gavinmc42]

Much as I don't like weeing on peoples fireworks, any new A+ is quite a way away. Like, a REALLY long way.

I hear the little coconut shells clapping getting closer.

So no holding breath and back to my Zero's and 3B+'s, message received

[piglet]

any new A+ is quite a way away. Like, a REALLY long way

Good. I'll embed zeroes into things that don't need ethernet.

However I'd (will) happily replace several Pi's with a more powerful 4B/5B/6B model as soon as it's available, and I suspect that many many others would immediately drop their 3's for 4's or something that gave better general purpose usability.

When I say "general purpose usability" I mean being able to browse at better speed. At the moment it's just a little too painful not to have a PC running at the same time to do the looking things up....

I really don't see the use for an A.

[DougieLawson]

A Zero+ with 1GB RAM and a 3B+ BCM2837 would be seriously cool and a fair alternative to an A+. But asking for all these new things is tending to make this thread into YAPP&UT*.















* Yet another pink ponies & unicorns thread.

[ejolson]

A Zero+ with 1GB RAM and a 3B+ BCM2837 would be seriously cool

I think it would be a bit cooler if 3 of the 4 cores on the BCM2837 were disabled and the other limited to 600 MHz clock speeds. Still, if such a Zero+ could run 64-bit code while meeting the 0.5 amp power budget of the original Zero, that would make it useful in my mind.

[W. H. Heydt]

A Zero+ with 1GB RAM and a 3B+ BCM2837 would be seriously cool

I think it would be a bit cooler if 3 of the 4 cores on the BCM2837 were disabled and the other limited to 600 MHz clock speeds. Still, if such a Zero+ could run 64-bit code while meeting the 0.5 amp power budget of the original Zero, that would make it useful in my mind.

That still crashes on the rock of either having to using PoP RAM or increasing manufacturing cost by having to solder to both sides of the board. So while I understand the idea that an upgrade to the A+ "is a long way away" (though with no indication of how that individual calculates time), and in spite of the fact that Eben actually showed a Pi3A prototype a couple of years ago (i.e. at least an initial board does exist for that version), I think such an upgrade remains a good idea.

As for the remark that the designers have time... No. Any design team that is sitting around not doing anything in particular has very poor management and I flatly refuse to believe that of the RPF/RPT. They're busy. Busy doing what, we don't know, though at least some of their efforts are undoubtedly directed to the next version of the Pi.

[DougieLawson]

The current price for the A+ (at RS Components) is £18 (£22.19 inc VAT). The ZeroW (at Pimoroni) is £7.63 (£9.16 inc VAT). So there's plenty of room to play with adding things to a "Zero+" without going beyond that A+ price.

[Lob0426]

Thankfully this thread has made efforts to stay away from "Pink Ponies and Unicorns". No USB 4.0, No 4GB RAM, no onboard SATA6,,,,,,,

There was discussion of Zero's and BCM2837 which was tossed due to TDP concerns in another discussion. And there has been realistic discussion of a Pi3A+ mission and usefulness.

I thank everyone for keeping on topic and NOT asking for new features and unrealistic additions. Hopefully RPF will, someday, decide the A+ format has a place and dust off the plan to upgrade to a better processor package.

[W. H. Heydt]

The current price for the A+ (at RS Components) is £18 (£22.19 inc VAT). The ZeroW (at Pimoroni) is £7.63 (£9.16 inc VAT). So there's plenty of room to play with adding things to a "Zero+" without going beyond that A+ price.

A better comparison would be to the Pi0WH, since the A+ has the headers install at the factory. At that point, the cost difference between the Pi0WH and the A+ is $5, but the A+ also brings power protection to the table, so the effective margin is even narrower.

Since pretty much the point of the Pi0/Pi0W is minimal price, anything that drives up cost (and, therefore, price) isn't going anywhere. The A+ doesn't have that particular constraint, though I'm sure that the intent is that any new board would sell for the same amount. If that can be achieved, then a Pi3A+ at $20 (or even $25, the price of the original Pi Model A) the Pi3A+ should then be a much more attractive board for any use that needs more CPU power than a Pi0W.

If an updated A Series Pi is "a long way away", then an updated Pi0/Pi0W is a much, much farther away than that.