New Raspberry Pi 4 Model Splits RAM Across Dual Chips (omgubuntu.co.uk) 55
The blog OMG Ubuntu reports that a new version of the Raspberry Pi 4 Model B has been (quietly) introduced. "The key difference? It now uses a dual-RAM configuration."
The Raspberry Pi 4 Model B (PCB 13a) adopts a dual-RAM configuration to 'improve supply chain flexibility' and manufacturing efficiency, per a company product change notice document. Earlier versions of the Raspberry Pi 4 use a single RAM chip on the top of the board. The new revision adds a second LPDDR4 chip to the underside, with a couple of passive components also moved over... In moving to a dual-chip layout, Raspberry Pi can combine two smaller — and marginally cheaper — modules to hit the same RAM totals amidst fluctuating component costs...
This change will not impact performance (for better or worse). The Broadcom BCM2711 SoC has a 32-bit wide memory interface so the bandwidth stays identical; this is not doubling the memory bus, it's just a physical split, not a logical one. Plus, the new board is fully compatible with existing official accessories, HATs and add-ons. All operating systems that support the Pi 4 will work, but as the memory setup is different a new version of the boot-loader will need to be flashed first.
This change will not impact performance (for better or worse). The Broadcom BCM2711 SoC has a 32-bit wide memory interface so the bandwidth stays identical; this is not doubling the memory bus, it's just a physical split, not a logical one. Plus, the new board is fully compatible with existing official accessories, HATs and add-ons. All operating systems that support the Pi 4 will work, but as the memory setup is different a new version of the boot-loader will need to be flashed first.
Performance impact (Score:2)
They say no performance impact, presumably meaning compute performance. But what about power efficiency performance?
Re: (Score:3)
It's LPDDR4 RAM. Having a second chip is insignificant, an entire SODIMM typically uses 300mW of power and usually has 8 of these things on them.
Re: (Score:2)
if the RPi4 had per-chip power gating, you could potentially have better power savings. But it doesn't, and the amount you'd save from power gating versus PASR on LPDDR is pretty small. I think pi4/5 still don't support self-refresh or suspend-to-ram, but the hardware should be capable when the firmware and kernel are ready for it.
Frankly. Raspberry Pi was never aimed at being ultra lower power. It's aimed at a price for a mostly open and programmable platform that is accessible to hobbyists.
Re: (Score:2)
Do you really imagine the power needs of two 2 Gig RAM chips is that much different from one 4 Gig RAM chip?
Re: (Score:2)
This is the reverse of what I remember was a HUGE issue with early Mac Book Air M1 systems - All the higher memory M1 laptops used two memory chips to speed up memory access when the lowest memory models used only one memory chip for performance that was half the speed of the dual-channel, higher-memory models.
Re: (Score:2)
I think the Pi SoC only has one memory channel.
Re:RAM recycling? (Score:5, Interesting)
Recycling was a lot easier when RAM was on SIMMs and DIMMS (and socketed DIPs for the old-timers).
Re: (Score:2)
Unlikely that it would be compatible. You would need a reflow oven and xray machine at the minimum.
Re: (Score:3)
Re: RAM recycling? (Score:2)
Absolutely it's possible to get chips off a phone and reuse it. Doing it manually isn't worth it even if RAM were to be many times more expensive. But if you can feed phone PCBs to some image recognition that automates matching to know phone models and removes valuable chips. Then sure, refurbished RAM from a 2-3 year old smartphone could be repurposed.
But you'd have to test and reball any chips you pull. And testing is like half the cost in chip manufacturing.
My ZX Spectrum had 4 memory banks (Score:2)
Way ahead of its time in 1984
Re: My ZX Spectrum had 4 memory banks (Score:5, Interesting)
I wonder if anyone made a magnetic core memory module for the Pi?
Core memory [core64.io] for the Raspberry Pi Pico.
Re: (Score:2)
They can't even make a web site that will display properly in a non-Chrome browser. Hard pass.
Re: (Score:2)
Re: (Score:2)
Everspin sells MRAM (Magnetoresistive RAM). I think in up to 8MB sizes. But I think it only offers I2C and SPI as the interfaces, so not very friendly for running your entire Pi on it. But could make for a rather handy non-volatile storage.
Re: (Score:2)
The ZX81 is an even better example. Officially it was a four chip solution - ULA, CPU, RAM, and ROM. This was announced at release and heavily advertised (part of the appeal was you could buy it in kit form and put it together with minimal soldering.) In practice multiple models, for the same reason, came with two half-kilobyte RAM chips instead of a single kilobyte chip, because Uncle Clive found it was easier to source them.
There is much better hardware anyways (Score:1)
Get an Orange Pi, for example. Designed by people with some actual understanding of hardware, not the clowns that designed the RP. Typically cheaper, more options, etc.
Re: (Score:3)
Can you elaborate?
Re: (Score:1)
The Pi folks fuck up bigly on approximately every design. I've been watching casually since I had the raspi model A with the shitfest USB that all the fanboy clowns defended to the death even though it made the device completely worthless for any USB traffic heavy applications. They especially half-ass the power supply section of the board almost every time, which is very very bad since if that doesn't work right, nothing else does either.
Re: (Score:3)
And yet the RPi still sell very well after all these years...
The purpose of the RPi was not to build pocketable super computers, it was to introduce a new generation of students to a computing experience akin to the late 70s and 80s, Apple, Tandy, Commodore, etc, as well as encourage harware mods like leds, relays, and sensors.
The original designs were based on cellphone chips, which oddly lacked things like Ethernet, so yes, early designs were not optimal, but they were functional.
This hardware change make
Re: (Score:2)
Bingo. It was a product designed to be made as cheaply as possible.
Re: (Score:2)
Bingo. It was a product designed to be made as cheaply as possible.
Obviously - the headlines were about the price, $35, not the MIPs of the CPU.
Lets travel back in time to the original release of the RPi, we were still pretending the $100 OLPC really was a $100 computer the average third-world user would find better than having no computer in front of them.
Once produced, the RPi foundation started making money, which they rolled into developing better systems, struggling to maintain the price-point they established with the first model. The RPi 5 is a marvel, light-years b
Re: (Score:2)
And yet the RPi still sell very well after all these years...
Consumer ignorance is always a good bet.
Re: (Score:2)
Indeed. That is why you can make tons of money with inferior products. You just need to find enough idiots that cannot tell the difference.
Re: (Score:2)
Re: (Score:2)
it was to introduce a new generation of students to a computing experience akin to the late 70s and 80s
Which is does not do at all. Just the boot loader is more complicated than the entire software stack on many of those early 80s vintage machines. Yes you can learn about doing some digital logic stuff with GIO pins and interfacing them to python and what not as a true beginner but that is about it.
They all got used by geeks wanting to make cheap media pcs/home servers/smart sensor projects.
Are the useful dang right they are, are they fun sure, but for mission of intro to computing, way to tall a stack. Th
Re: (Score:2)
An 8088? Seriously? A kludgy, ugly, CPU from an era where it was expensive to add pins to chips resulting in weird multiplexed buses, and an ISA only a motherboard designer could love*?
This would be better than a RISC CPU why?
And this is being proposed because... the bootloader on the Pi is complicated? Why not just propose simplifying the bootloader? Or a minor architectural change that'll result in a simplified bootloader for specific tasks, such as using an EEPROM for the firmware?
* Yes, that's a really
Re: (Score:2)
An 8088? Seriously? A kludgy, ugly, CPU from an era where it was expensive to add pins to chips resulting in weird multiplexed buses, and an ISA only a motherboard designer could love*?
Sure, why not?
I learned on a Z80.
Old CPUs were simple to bootstrap and directly program. Great for learning.
This would be better than a RISC CPU why?
RISC? RISC isn't even a relevant categorization, here.
The Pi could be VLIW for all you care. That SoC is so fucking convoluted that you quite imply cannot bootstrap it. You need BCM to do it.
And this is being proposed because... the bootloader on the Pi is complicated? Why not just propose simplifying the bootloader? Or a minor architectural change that'll result in a simplified bootloader for specific tasks, such as using an EEPROM for the firmware?
See above. Ask me how I know you don't know what you're talking about right here.
* Yes, that's a really bad pun. But honestly, the 8088 has an overcomplicated ISA for what we're talking about here. While I'm not as big a fan of RISC as some here, it's rather obvious that a simpler RISC system like ARM would be more appropriate for a learning computer.
What on Earth are you talking about? The 8088 has a bloody simple ISA.
Again, it doesn't matter what the Pi is- you'll never tou
Re: (Score:2)
> Sure, why not? I learned on a Z80.
Yes, but you didn't say a Z80, you said an 8088. Again, why a kludgy CPU with a kludgy ISA and multiplexed bus?
> Old CPUs were simple to bootstrap and directly program. Great for learning.
8088s were NOT simply to directly program outside of maybe limiting your computer to 64k of RAM. They were fugly. C programming was a nightmare of different word sizes.
> RISC? RISC isn't even a relevant categorization, here. The Pi could be VLIW for all you care. That SoC is so
Re: (Score:2)
Yes, but you didn't say a Z80, you said an 8088. Again, why a kludgy CPU with a kludgy ISA and multiplexed bus?
An 8088 isn't particularly worse than an 8080.
8088s were NOT simply to directly program outside of maybe limiting your computer to 64k of RAM. They were fugly. C programming was a nightmare of different word sizes.
You mean the half-width bus? That was hard for you?
*headdesk* If it happens that this is literally true of currently available ARM chips, and it might be, and there's literally no way to get an ARM with more simple requirements, then there are still dozens of architectures to choose from that are (1) ACTUALLY FUCKING SIMPLE AND INTUITIVE TO PROGRAM and (2) do not have this limitation. Way to miss the point. Does RISC-V have this problem? Or if you have an allergy to RISC (it sounds like you do!) what about one of the descendants of the 68000 architecture? The 68000 at least had a clean ISA and memory map, and no fucking hardware multiplexing.
An allergy to RISC? Of course not- an allergy to people who think that means something.
Also, no it's not true of all currently available Arm parts (stop calling it ARM)
At this point I have to consider the probability you're actually trolling. A segmented architecture chip with a far from uniform instruction set will never be a "simple" ISA. There's a reason everyone at the time preferred the 68000, and even Bill Gates said the choice of an 8088 was the one thing he disagreed with IBM on with the original PC.
Your handwavy "segmented architecture" is eye-rolling.
Built-in segment registers was far superior to how it was done on things like the 8080 with explicit paging.
You're an idiot.
Re: (Score:2)
Wow - Pedantic Much?
They should built little boards with a 8088, integrated chipset, with parallel and serial ports with freedos on a ROM.
No, we don't need them to have to grab solderless breadboards or a soldering iron for pete's sake! We want them to learn coding using tools like scratch or python, and the target audience was a group that lacked access to their own desktop/laptop in schools or at home... A $35 computer that hooks to their TV, uses an old cellphone adapter for power, and needs a USB keyboard/mouse that can usually be found for a few dollars was the goal.
But let's play out your vision - they take a 40 year
Re: (Score:2)
It used to be one had to learn to crawl before they walked. These days we teach them how to fly before they're potty trained.
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
There's a ton of value in broad availability of supply, huge support and peripheral ecosystem and so on.
If I need to squeeze performance on a budget, then sure. But for many cases the RasPi is fine, and the time spent not faffing around with niche hardware greatly outweighs the few tens of dollars of extra cost for the Pi.
Re: (Score:2)
There's a ton of value in broad availability of supply, huge support and peripheral ecosystem and so on.
I suppose it depends what you're going to do with it. If you're going to make and market an industrial controller, or a networked bathroom snitch puck, and require (moderately) guaranteed-stable-and-available hardware that you can vibe code in Python, Pi is a good solution. If you've got children and want to teach them about computers from YouTube videos but miss out all the fun parts of actually understanding how they work and how to fix them when they don't, Pi is the answer.
If you're doing DIY stuff an
Re: (Score:2)
I should follow this up with the addendum that the RPi folks hit it out of the park with the Pi Pico.
Custom silicon rather than taking a shitty SOC and drawing a smiley face on it, features no one else has (mainly the PIO blocks), and a price point so good that no one's really going to bother knocking it off (there are clone boards, but they all seem to use the RP chip).
Different market, but a full embrace of the DIY maker ethos that RPi started with.
Re: (Score:2)
The goal was Python/Scratch programming, not explaining to 4th graders how bootloaders work and how to program in assembler...
That early RPi had USB-based ethernet on a slow USB port was fine - the goal was to download webpages over a wired network connection, not serve as a network drive for a server farm.
Re:There is much better hardware anyways (Score:4)
"Better" requires you to define your requirements. All Orange Pi models come at a considerable price disadvantage compared to RPis. So already it is a failure for the requirements of many people, those who already consider the RPi 4 to be too expensive.
Please stop suggesting solutions to people without even asking what the problem is.
As to your complaints, there's been one major screwup and that was the power design on the original Pi. Everything else is a tradeoff. The original RPi had a perfectly find USB interface for many of the applications it was designed for, and certainly had a perfectly fine USB interface given the cost of the device. If it doesn't suit your needs and you bought it, then you are the person who has no clue. Next time spend more and buy a device commensurate with your project requirements.
There's a reason the RPi was a success despite your assertion that the people don't understand hardware (and it's easy to be a criticising arse of a keyboard warrior when you never tried designing such a system yourself)
Re: (Score:2)
The original RPi had a perfectly find USB interface for many of the applications it was designed for
Where "applications it was designed for" means "things I used it for without problem".
If it doesn't suit your needs and you bought it, then you are the person who has no clue.
No, that's not how this works.
Next time spend more and buy a device commensurate with your project requirements.
I do agree with you on this point.
With the Pi, you 100% get what you pay for.
There's a reason the RPi was a success despite your assertion that the people don't understand hardware (and it's easy to be a criticising arse of a keyboard warrior when you never tried designing such a system yourself)
Ah, yes. The good ol' appeal to popularity.
Re: (Score:2)
Where "applications it was designed for" means "things I used it for without problem".
No, it means applications it was designed for. If you shoehorn it into something beyond its capabilities then *you* are the problem, not the RPi. USB 2.0 is a perfectly fine interface for many applications.
No, that's not how this works.
This is exactly how it works. That's why we have things such as spec sheets. No one with a single braincell bought an original Raspberry Pi expecting USB 3 performance. It was clearly advertised what was on offer.
Ah, yes. The good ol' appeal to popularity.
It's stating the fact that the RPi has a proven track record of being excellent in a huge var
Re: (Score:2)
No, it means applications it was designed for.
No, it doesn't. It means exactly what I said.
If you shoehorn it into something beyond its capabilities then *you* are the problem, not the RPi. USB 2.0 is a perfectly fine interface for many applications.
USB 2.0 is a perfectly fine interface for HID applications, and that's it.
This is exactly how it works. That's why we have things such as spec sheets. No one with a single braincell bought an original Raspberry Pi expecting USB 3 performance. It was clearly advertised what was on offer.
lol. I'd say 99.9% of people who bought a Pi didn't even know the difference between USB2 and USB3. They just bought a cheap ultra-low performance Arm computer.
It's stating the fact that the RPi has a proven track record of being excellent in a huge variety of cases. It may not fit your case, but that just means it's not the right device for you. The OP made the claim there was some fundamentally flawed engineering, and yet all evidence of the projects that worked successfully demonstrate the contrary. The device is popular for a reason, it demonstrated its performance / price position and its suitability to many applications.
It was an argumentum ad populum, which is stupid. Don't do it.
Re: There is much better hardware anyways (Score:2)
I'm pretty ignorant when it comes to Linux but when I bought a Pi and played around with it drivers that enabled all the GPU hardware seemed to be a significant issue.
Re: (Score:2)
Get an Orange Pi, for example. Designed by people with some actual understanding of hardware, not the clowns that designed the RP. Typically cheaper, more options, etc.
And BetaMax had superior video quality. And yet, people ended up at VHS because technical quality is not always the sole determining factor of who wins.
Also: do you think OrangePi is immune to the current RAM sourcing problems?
SODIMM (Score:3)
Just give us a SODIMM slot already.
Re: (Score:3)
They can't change the form-factor, and there are no commercial SoDIMM sockets that would conveniently fit on the RPi 3/4/5 form-factor without removing other needed connectors and blowing the height of the current design.
A 8 Gig RPi 5 costs $120, a 4 Gig DDR4 SoDIMM costs $60
Ok then... (Score:1)
Re: (Score:2)
Dude, seriously? They didn't "improve older models", they are working around a parts-constraint issue, either one they are experiencing or expecting soon.
You think using two lower-capacity memory chips is an "improvement"? They literally said it provided no improvement.
As I mentioned previously, RPi has a legal obligation to produce/maintain inventory for each of their products for 10 years, if going from one memory chip to two helps them keep costs down/improve availability, then that's great.
Re: (Score:2)
Why make a Pi5 that is supposed to be 25% faster if youre just going to go back and improve older models?
Unlike a lot of electronics manufacturers, the RPi folks support their devices for a long time. They still manufacture and sell the Model 3B, which was introduced in 2016. The Model 4 came out in 2019, and they have committed to keeping it in production until at least 2034 [ref [raspberrypi.com], page 4]. This change with the RAM is one way to help make that happen.
Like most electronics manufacturers, RPi has different models at different price points, with performance specs to match. The Model 5 has the best capabili
Did they remove the audio jack? (Score:2)
Re: (Score:1)
Re: (Score:1)