Pi 4 questions

[pi4b]

I find the Pi3B+ to be quite good, especially when running a 64-bit OS like the one from sakaki.

That said any piece of technology has room for improvement.

Question 1:
What about the SoC prevents the use of external RAM? Even slow external RAM would be faster than paging.

Question 2:
Would it be feasible to have two SoC's on the next Pi, so that a complex task could be split across the two, and the other chips could be shared between the two?

Thanks.

[drgeoff]

The SoCs used in all RPis to date cannot address more than 1 Gbyte RAM.

[rpdom]

Question 1:
What about the SoC prevents the use of external RAM? Even slow external RAM would be faster than paging.

The only interface for external memory is via the GPIO running in SMI mode. That can address a small amount of memory at a slow speed. Paging is faster.

Question 2:
Would it be feasible to have two SoC's on the next Pi, so that a complex task could be split across the two, and the other chips could be shared between the two?

Don't ask.

Seriously.

The next model is already in either the final design stages or in prototyping. Nothing we can say will affect the design now. All we know is that it will have some cool stuff on it.

The model after that is well into the planning stages. Again, no point in speculating or asking for stuff because it has mostly been decided already.

Also, asking for features on the "next" Pi often means people start adding wish lists and the thread gets locked.

I'm not having a go at you, but this is generally what has happened many times when someone says "I think the next Pi should have...".
Just to let you know, that's all.

[DougieLawson]

The next model is already in either the final design stages or in prototyping. Nothing we can say will affect the design now. All we know is that it will have some cool stuff on it.

I just hope my request for a pink pony and a unicorn has been built in to the next Raspberry. If not then I'll consider giving up with them.

Got to have unicorns and the ponies.

[echmain]

Thread will be locked in 3...2...1...

[bensimmo]

To be fair, ones a question a reasonable one at that, two the answer is probably not, it would almost certainly cost to much.

[jamesh]

To answer, the databus to the Ram and the controller itself are limited to 1Gb. It would require a new controller to increase memory. There's another reason that escapes me.

As for two chips, the associated circuitry to deal with two chips and plug them together is prohibitive. Notwithstanding the cost of 2Socs on the board. Much easier to add cores inside the SoC where all that is handled. Although adding more cores has it own problems. 4 is a sweet spot.

[LTolledo]

Mine is simple question

How much longer (or shorter) is the wait?

...me hands are ichin'....

[W. H. Heydt]

Mine is simple question

How much longer (or shorter) is the wait?

...me hands are ichin'....

The wait shortens at a rate of 1 day per day.

[W. H. Heydt]

Question 1:
What about the SoC prevents the use of external RAM? Even slow external RAM would be faster than paging.

Pis use external RAM, in that the main memory is not integrated into the SoC. For Pis that use the BCM2835, the RAM package is on top of the SoC ("Package-on-Pakage, aka "PoP"). For all other Pis, it's mounted separately on the bottom of the PCB.

[W. H. Heydt]

The next model is already in either the final design stages or in prototyping. Nothing we can say will affect the design now. All we know is that it will have some cool stuff on it.

We know one other thing about the SoC for the Pi4. It won't use a 40nm process. It will use 28nm...or less. I'd bet on 28nm.

[LTolledo]

Question 1:
What about the SoC prevents the use of external RAM? Even slow external RAM would be faster than paging.

Though I dont have any use for it now... I like the idea of being able to add external RAM (through an external circuit/board)...

[W. H. Heydt]

Question 1:
What about the SoC prevents the use of external RAM? Even slow external RAM would be faster than paging.

Though I dont have any use for it now... I like the idea of being able to add external RAM (through an external circuit/board)...

Given the form factor of the B-series Pis (the physically largest models), and the form factor of typical DIMMs, I can't see that ever happening even if the cost structure would permit it. Can you imagine finding a place to put an SODIMM connector so that something the size of a CM could be attached?

[Gavinmc42]

Old timers here know about the unicorn and pink pony code for locking a post when entering fantasy land.
But noobies need to learn too

The price for Pi4 is $35, that is the part we do know and it is pretty much set in concrete.
It will be "cool" as expressed by someone in the know who has seen the roadmap.
That it will be 28nm or less is also known as a high probability.
The 1GB dram memory limitation is also a known issue.

What solutions are needed for these issues to meet that $35 price point?
Someone probably knows by now, we will know in 2020 probably

In the mean time practice writing tight aarch64 code.
"640K ought to be enough for anybody." 1GB is much more than that

[W. H. Heydt]

The 1GB dram memory limitation is also a known issue.

To be clear about it, the 1GB limit is for the VC4. It is *not* known if the SoC for the Pi4 will have that limit, and--indeed--it would seem to be a very poor decision to design a new SoC with that limit. That does not, however, mean that the Pi4B, at least when it arrives, will have more than 1GB RAM even if it can address more.

What solutions are needed for these issues to meet that $35 price point?
Someone probably knows by now, we will know in 2020 probably

Or 2021....

[ejolson]

To answer, the databus to the Ram and the controller itself are limited to 1Gb. It would require a new controller to increase memory. There's another reason that escapes me.

As for two chips, the associated circuitry to deal with two chips and plug them together is prohibitive. Notwithstanding the cost of 2Socs on the board. Much easier to add cores inside the SoC where all that is handled. Although adding more cores has it own problems. 4 is a sweet spot.

Thanks for an informative answer to an honest question. It is sometimes said the only dumb question is the one not asked. It is nice to see a Pi-enthusiast enjoying the freedom to ask a Pi-related question without being scolded for asking the wrong question.

[Heater]

ejolson,

It is sometimes said the only dumb question is the one not asked.

I thought that was just something we said when people asked a dumb question. In an attempt to alleviate their embarrassment

I've also often heard it said: "Judge a student by the questions they ask, not the answers they give."

Our OP said: "...any piece of technology has room for improvement."

That is no doubt true. However the suggestions are:

1) Add external RAM. - Even if this were possible it would add expense and board complexity. This is not an improvement.

2) Use 2 SoC. - Again an increase in cost, a massive increase in complexity. This is not an improvement.

My feeling is that if you really need a 64 bit OS and bigger RAM it's time to think about a different SBC. There are plenty around now a days.

The Pi is what it is. It will no doubt see improvements in the future, withing the constraints of cost and practicality.

[Gavinmc42]

To be clear about it, the 1GB limit is for the VC4. It is *not* known if the SoC for the Pi4 will have that limit, and--indeed--it would seem to be a very poor decision to design a new SoC with that limit. That does not, however, mean that the Pi4B, at least when it arrives, will have more than 1GB RAM even if it can address more.

I did know that 1GB limit was due to the VC4
How old is the VC4 design now, 10years?
Speculation a VC5/6 will address more?
But the current VC5 used in those Set Top boxes does not have everything the VC4 has.

The amount or RAM it will come with depends on RAM pricing at the time in 2020 or 2021 or whenever it shows up.
But the RAM limitation shows up due to bloaty Browser code and Video apps etc.
Why not leave it at 1GB until this current/next generation of coders learns how to write tighter code

I guess it depends if RPF wants to grab part of the Desktop market
The Pi3B+ is getting very close now to being usable for most things if you use the latest 64bit kernel/mesa etc.
USB3 will give the Pi4 the speed for faster SSD etc.

Still plenty to learn on Pi Zeros and 3B+, I'm in no hurry for a Pi4.

The limitations of the current Pi's is known.
We have been making wishlists since it has come out.
It has takien 7? years to get most of the software working pretty good and the software can still be improved some more too.

[spock]

https://www.tomshardware.com/news/raspb ... 38585.html

"I don’t have a route to do something this year," he said. "I think we kind of understand what featureset we want and what would be involved in getting that featureset. I don't think we have a defined plan for turning that into a product yet."

that statement by eben upton is kind of weird though when here on the forum everyone always is like:

The next model is already in either the final design stages or in prototyping. Nothing we can say will affect the design now. All we know is that it will have some cool stuff on it.

[Imperf3kt]

I cant help but think a lot of the replies here very rude.
Certainly, these types of threads are a dime a dozen and certainly all the folk who've been here a few months can see the trend.
But a new user doesn't know this and shouldn't be belittled when asking such questions. Surely a polite "this gets asked frequently, here is why it's not going to work as you think", is all that is needed?

[Heater]

"very rude"?

Dismissive perhaps. Stern perhaps. But "very"?

I used to get far worse from my school teachers. They really knew how to put you down, politely.

[LTolledo]

Nice numbers you have there Heater...

nicenumbers.jpg (6.82 KiB) Viewed 3555 times

Sorry for the off topic

[jamesh]

I cant help but think a lot of the replies here very rude.
Certainly, these types of threads are a dime a dozen and certainly all the folk who've been here a few months can see the trend.
But a new user doesn't know this and shouldn't be belittled when asking such questions. Surely a polite "this gets asked frequently, here is why it's not going to work as you think", is all that is needed?

If you feel a post is inappropriate, please report it. Mods are mostly reactive, not proactive, so although I have read this thread, and commented on it, I won't delete stuff unless a) Its reported and I agree or b) is so bad I delete it immediately on reading

[jamesh]

https://www.tomshardware.com/news/raspb ... 38585.html

"I don’t have a route to do something this year," he said. "I think we kind of understand what featureset we want and what would be involved in getting that featureset. I don't think we have a defined plan for turning that into a product yet."

that statement by eben upton is kind of weird though when here on the forum everyone always is like:

The next model is already in either the final design stages or in prototyping. Nothing we can say will affect the design now. All we know is that it will have some cool stuff on it.

Depends when Eben gave that TH interview. Pi4 feature set IS set in stone. Of course, I'm not going to say what that feature set is, but it's pretty cool.

[Gavinmc42]

Of course, I'm not going to say what that feature set is, but it's pretty cool.

Stop saying that, it drives me nuts trying to imagine what "cool" means to you.
If it really is cool it will come in blue, perhaps the same blue as The Orville colour scheme

Seriously I hope production can keep up because this next one could sell 100Million the first year
Make 10Million a month? ok now a realistic 1Million/mth? er that is still scary big numbers.
Such a problem to have

Clip clop - little unicorn feet time?
Nope just reruns of Monty Python.