Which 3 things do you want to see in the Raspberry Pi 4?

[CobraX]

Which 3 things would you like to see in the Raspberry Pi 4? Here's my list:

1) Much faster I/O system (I believe the current limit is 25 MB/s for sequential reads/writes, I'd like to see 75 or 100 MB/s and also faster 4K random read/write speeds).
2) More RAM (2 GB maybe?)
3) Faster CPU that runs cooler under max load (max 65°C with small heatsink, if that's not possible definitely not past 70°C)

[W. H. Heydt]

First...this thread won't last long. The whole topic has been done to death in most peoples eyes and tends to devolve into either recrimination or silliness.

I am hoping for one single improvement: Faster I/O, probably in the form of USB 3.0. The current limit is the USB 2.0 port at 480Mb/s, which can give mass storage I/O up to 35MB/s when tested. USB 3.0 should run that up to around 125MB/s and also permit (for those who want it) GbE.

Everything else falls into "nice to have" but it doesn't matter if it isn't added. I would be quite content with the same RAM and the same CPU clock speed.

[CobraX]

Well, I'm hoping for a mature discussion/wish list cause I'm curious about what other people would like to see in the Raspberry Pi 4.

I can live with the current ethernet speeds even though my cable internet is 200 Mbps down and 20 Mbps up. I also hardly use the USB ports, so I don't really need USB3 speeds.

[Heater]

CobraX,

I'm hoping for a mature discussion/wish list...

That is a reasonable thing to hope for. Problem is that such a mature thread gets started here about once a week. They inevitably devolve into all kind of random requests for features and speculation about what will come next.

Then the thread get's locked as it has become pointless.

...cause I'm curious about what other people would like to see in the Raspberry Pi 4.

Have a scroll back through all the recent threads and you will find many containing people's pet wish lists.

Bottom line is that the Pi Foundation does not disclose what it has under development. We just have to wait and see what comes.

[CarlRJ]

W.H. nails it pretty accurately. The Pi's Achilles heel is the bottleneck of having one USB2 port for all high-speed IO. So, I want to see the USB2 replaced with USB3 (it doesn't need to venture into type-C connectors yet - given that the Pi is intended to work with hardware you already have laying about, type-C might cause more problems than it fixes at this point). Swapping in USB3 will require an overhaul of the silicon, a major undertaking. While they are in there, putting in the plumbing for more RAM would be a good idea though I don't think the Pi really needs that RAM yet. And a die shrink would probably improve the thermal situation enough to pave the way for higher clock speeds. Although, again, I don't see the current speed as a huge impediment yet. And, finally - and quite seriously - while they're rejiggering the silicon, I'd love to have them put a microscopic picture of a unicorn in a spare corner of the chip mask (this is not without precedent - other chips have tiny "Easter egg" pictures - and it would appease everyone who asks for unicorns and pink ponies).

I'm pretty thrilled with Pi hardware as it is, really - the Zero W makes a whole range of cheap wifi projects practical, and the 3B is quite capable for larger things. Faster IO would make a range of fileserver-type applications more practical. Lots of people get caught up into improving the Pi into a full-blown desktop PC, bit that seems a bit silly when the world is littered with desktop PC's that can kick its ass any day of the week, in terms of raw horsepower. Better for the Pi to play to its strengths - low cost, low power, "good enough" hardware for a huge range of tasks where you don't want the overhead, expense and power/thermal needs of a traditional PC.

[stderr]

Swapping in USB3 will require an overhaul of the silicon, a major undertaking. While they are in there,

Rather than inventing the SoC that people want, shouldn't people be looking around at what is available that can be used in the pi4? People keep pretending that the reason the pi is the way it is is because that's the way that those who designed it wanted it to be. Well, it's the way it is because of some of that, of course, but mostly it's because of what was possible given what was available and available from their suppliers.

For example, consider the following which would answer a lot of the complaining pretty well, it's even from Broadcom, the BCM58712: Quad-Core 2.0GHz 64-bit ARM®v8 Cortex®-A57 based SoC, https://www.broadcom.com/products/embed ... /#overview . It's not for mobile phones, it is for networking and that sort of thing, so it would not make those looking for video machines happy. But the router and NAS folks would be a lot happier than now.

So what is available? And how much effort would it take to support for the next whatever time they plan to support these devices? They can't change everything every time they turn around or there will be a support nightmare. These sorts of threads that get shut down that I've seen rarely delve into these sorts of issues very much.

[W. H. Heydt]

While they are in there, putting in the plumbing for more RAM would be a good idea though I don't think the Pi really needs that RAM yet.

Very much agreed, but that requires not reworking the CPU, but a rework on the VC4, probably a tougher proposition.

And a die shrink would probably improve the thermal situation enough to pave the way for higher clock speeds.

I strongly suspect that a switch to 28nm (from the current 40nm) is in the works. There are two ways to go with that. One is improved thermals with the current clock speed (say goodby to almost all the angst about cooling solutions) or boosting the clock. It's also possible to split the difference and get a modest clock boost (1.4GHz?) without as much risk of thermal issues. One way to do that would give a Pi4 a default 1.2GHz clock with "authorized" overclock settings.

Faster IO would make a range of fileserver-type applications more practical.

This. Think classroom servers to link back to the educational purpose, but it would permit any number of other server types of uses. It would also permit USB3 GbE Ethernet adapters to run at full speed, even if the board itself only has 10/100 Ethernet.

[W. H. Heydt]

Rather than inventing the SoC that people want, shouldn't people be looking around at what is available that can be used in the pi4? People keep pretending that the reason the pi is the way it is is because that's the way that those who designed it wanted it to be. Well, it's the way it is because of some of that, of course, but mostly it's because of what was possible given what was available and available from their suppliers.

I'm an old fan of James Burke's "Connections" series. I've always assumed the original chip was used because that was what was available to the people putting it all together, and then did a couple of hacks to make the '2836 and '2837 SoCs.

[fruitoftheloom]

Which 3 things would you like to see in the Raspberry Pi 4? Here's my list:

1) Much faster I/O system (I believe the current limit is 25 MB/s for sequential reads/writes, I'd like to see 75 or 100 MB/s and also faster 4K random read/write speeds).
2) More RAM (2 GB maybe?)
3) Faster CPU that runs cooler under max load (max 65°C with small heatsink, if that's not possible definitely not past 70°C)

Do you really believe that RPT are not aware of current limitations ?

This has been discussed on numerous occassions !

Why have you not designed and manufactured a better SBC for $35 ?

There are other SBC's which fit your requirements and run ArmbianOS !

[bensimmo]

Why, but leave that to the mods to decide what people can talk about, once they read the Pi4 sticky.

What I don't understand is the obsession that it has to have VC4 and not a more generally available processor.
In two/three years time when this may come to market they can start differently, it doesn't have to fall into the Microsoft trap of being backwards compatible under the same image.

It's not as if the user will notice much difference in use, it still be debian probably and Pixel in some form but that depends how they work their software over the years.

USB C in 2/3 years time may also make sense as things will have moved on and more and more device now are moving over as it's only an adaptor to use older stuff and people seem quite happy using adaptors for Zero's and vice versa depending on what your cables are.

They have some good brains in there and It would be great to see the old white board scribbles (or are they trendy scribble on glass window people or old fashioned scrap of paper over a beer.. I assume nobody actually uses computers and phones for idea taking)
Like the pink pony idea, perhaps just on the processor casing as the emblem?

[gregeric]

What I don't understand is the obsession that it has to have VC4 and not a more generally available processor.

A large number of the RPi engineers, software & hardware, are the same team who designed & refined the VC4. Why wouldn't they want to stick to what they know, when they can perhaps tweak it (another slice of qpus etc) to further improve it? They know it inside out & have invested much time already in software for it.

[fruitoftheloom]

What I don't understand is the obsession that it has to have VC4 and not a more generally available processor.

A large number of the RPi engineers, software & hardware, are the same team who designed & refined the VC4. Why wouldn't they want to stick to what they know, when they can perhaps tweak it (another slice of qpus etc) to further improve it? They know it inside out & have invested much time already in software for it.

That is about the best response so far

Should of posted that also in the Risc-V thread

[RaTTuS]

e-ponies that's what I want - dual e-ponies if possible

[fruitoftheloom]

I demand that the next Raspberry Pi has this for $35.00:

https://www.servethehome.com/exclusive- ... enchmarks/

[jamesh]

Ah Ha! The weekly next model wish list thread. I wondered where it was.

Lots of people have good ideas, lots of people have bad ideas, lots of people have no ideas at all.

But to answer one question above. Making an ARM CPU based SoC is easy. You just buy the plans from ARM and make the chip. Making a GPU is really really NOT easy. You could use MALI, just buy plans from ARM, but that is missing a lot of features of the VC4 (camera ISP is a prime example, and what about H264 encode/decode?), so where does that come from?

Truth be told, the VC4 is a pretty good bit of kit, even after 10 years. It does a lot, and to replace it would not be very easy at all, notwithstanding the amount of work that has gone in to its software development just for the Pi. Of course that is a sunk cost, so should not really influence future plans, but I really wouldn't fancy rewriting the huge amount of code we have to make things backwards compatible, which is of vital importance.

And finally, the RPF is entirely aware of the limitation of the Pi. Do people think that we are just going to sit back, relax, chill, drink beer and just watch the money flow in?

[runboy93]

SoC: Higher clock on processor (1.5GHz - 2GHz quad core), higher clock on VideoCore IV (400MHz - 500MHz), better energy efficiency (reduced nm)

RAM: 2GB

Storage: There must be faster card readers out there, I don't mind paying a little more.

And I also hope they will keep similar layout to RPi 3.

[bensimmo]

What I don't understand is the obsession that it has to have VC4 and not a more generally available processor.

A large number of the RPi engineers, software & hardware, are the same team who designed & refined the VC4. Why wouldn't they want to stick to what they know, when they can perhaps tweak it (another slice of qpus etc) to further improve it? They know it inside out & have invested much time already in software for it.

I understand that, but it's not a limitation to the future goals if it holds ideas and solutions back.
I would have thought for a lot of the team, what it runs on is by and by. It's if they can make software and devices people can learn with.

Why waste time and money pushing hardware with a small team, when the larger manufacturers can do it for you, unless it is cost effective (I don't know and not going to guess if it is.)
The best part of the Pi is not the hardware itself* it is the software an support and contacts with companies they have created.


*What they physically created for the cost we buy it for is wonderful, but the broadcom vc4, what extra does it do that other processors cannot do.

[bensimmo]

....

And finally, the RPF is entirely aware of the limitation of the Pi. Do people think that we are just going to sit back, relax, chill, drink beer and just watch the money flow in?

I doubt they do, but with recent talk on website mentioning Pi4 (be it ideas, from Ebon chats). Enthusiastic people who like the products are always going to want to talk about it, say what's good, bad and could be improved.

[gregeric]

Why waste time and money pushing hardware with a small team, when the larger manufacturers can do it for you, unless it is cost effective (I don't know and not going to guess if it is.)

Use an off the shelf GPU/SoC & the cloners will be very happy. Meanwhile, RPiF's revenue stream will suffer. So you either persuade a 3rd party to make a SoC just for you (unlikely), or use your in-house & existing supplier's skill sets to refine/roll your own.

[jamesh]

*What they physically created for the cost we buy it for is wonderful, but the broadcom vc4, what extra does it do that other processors cannot do.

Well, depends on the device. Rockchip and Allwinner do devices with very similar or better (or sometimes worse) features.

So lets have a thought experiment. Lets says the next version of the Pi uses one of these chips. The RPF is now reliant on these suppliers for technical support. These suppliers (and indeed suppliers of all SoC's) are notoriously bad at support, so the platform is unstable, and takes weeks or months to get fixes because the very experienced with VC4 engineers at RPF have no idea how these new chips work. Not only that, but all userland software aimed at the Pi that used to run on all previous models of Pi now no longer works. So for example Raspistill and all the camera stuff no longer works. The GPIO system has subtle differences that means some things work, but others don't. Our market fragments, we still need to support the existing 13M installed base, so our engineers are still working on the VC4, but we are getting endless support calls for the new chips Raspberry Pi4, which we are handing off to the SoC supplier because we have no in house expertise. The supplier doesn't give a damn, because the volume involved simply cannot pay for extensive support, so fixes are few and far between. The Pi name turns to mud because our product doesn't work properly, all the 3rd party HAT's don't work, and people are having to rewrite their software for the new platform.

People underestimate the amount of pain to move to a new CPU architecture, and how important backwards compatibility is for this market, and how support can make or break a product. At the RPF, right now, we have lots of engineers who worked on the VC4, which means we don't actually need any support from the SoC supplier. But we still have a great relationship with that supplier which is worth its weight in gold for future plans.

Which do YOU think is the best approach?

[fruitoftheloom]

Truth be told, the VC4 is a pretty good bit of kit, even after 10 years. It does a lot, and to replace it would not be very easy at all, notwithstanding the amount of work that has gone in to its software development just for the Pi. Of course that is a sunk cost, so should not really influence future plans, but I really wouldn't fancy rewriting the huge amount of code we have to make things backwards compatible, which is of vital importance.

Yes the videoCore has been a long time in development, I was under the impression it dates backs over 15 years ??

https://en.wikipedia.org/wiki/VideoCore

[jamesh]

Truth be told, the VC4 is a pretty good bit of kit, even after 10 years. It does a lot, and to replace it would not be very easy at all, notwithstanding the amount of work that has gone in to its software development just for the Pi. Of course that is a sunk cost, so should not really influence future plans, but I really wouldn't fancy rewriting the huge amount of code we have to make things backwards compatible, which is of vital importance.

Yes the videoCore has been a long time in development, I was under the impression it dates backs over 15 years ??

https://en.wikipedia.org/wiki/VideoCore

Probably, I was 6 years at Brcm, then another 2 elsewhere, now done a few months at RPF, and when I joined Brcm, the VC3 was just being tested out. So that's 8 years at least, plus the extra few years for the earlier versions. Total dev must be getting on for 15 years at least.

[i486]

I can support only point 2 for 2GB RAM. For the rest - buy ultra slim desktop PC.

[fruitoftheloom]

Which do YOU think is the best approach?

Personally speaking I would like to see a VC5 which can address up to 2GB Memory and somewhere beyond
that Aarm64 Raspbian and beyond that a ARM OctaCore CPU -- that is the next five years or so covered.

[hippy]

CobraX,

I'm hoping for a mature discussion/wish list...

That is a reasonable thing to hope for. Problem is that such a mature thread gets started here about once a week. They inevitably devolve into all kind of random requests for features and speculation about what will come next.

Or simply get derailed by people who should know better asking for unicorns, ponies and the like. It's almost as if they want to shut down discussion, that mature discussion won't be allowed.

I don't really understand why these threads cause so much trouble, no matter how fantastical they get. Just ignore them.It seem it is mostly people saying 'you can't have that, won't get that', turning it into an 'I'm right, you're wrong' battle which seems to create antagonism which drives it into the gutter.