8GB Raspberry Pi 4 Launched For $75 USD

[wizard69]

Another SKU? Seems to me like it would be about time for the rPi foundation to drop the 1GB model in order to avoid having too many SKUs with very minor differences for people to keep proper track of.



Kind of obvious isn't it? Besides, the SoC, and SoCs of this type in general, don't tend to have native SATA support so it's going to be another chip that sits behind the USB controller like the Ethernet chip used to before they moved to an SoC with an Ethernet controller on the silicon. Let's also not forget that SATA is data only so anyone using a drive trough a native SATA port would have to figure out some external SATA power supply arrangement when a cheap USB-to-SATA adapter does all that in a single plug with no external power setup.

Designing in SATA on hardware of this type is just foolish at this stage. It makes about as much sense as an IDE or SCSI port at this point in time. Beyond that I suspect it would be far easier to find a chip with PCI-Express support than SATA. PCI has a future, SATA is dead in the water.

[wizard69]

People who wish the RPi with SATA ports should really ask themselves if their needs are not better satisfied with a X86 board. Just look at the other RPi competitors that have all sorts of connectivity. They all cost a lot more than the RPi and even surpass low cost PC boards.

There are low cost, mostly ATOM based, boards that are small and powered by 12V laptop PSUs that have most connectors people here want. The RPi will never be a desktop replacement. Is not their market.

It hasn't been their market but the PI is extremely useful for many markets where some sort of GUI must be used. Further one more SoC upgrade will put it into laptop competitive performance. Frankly for education PI with a GUI running is certainly good enough to teach programming or computer science on.

The biggest problem with PI is the I/O hardware and board outline which makes it hard to advance the platform as a more general purpose solution. I'd really like to see them change PI in the Pi 5 revision to make it a little more agreeable to modern device outlines. Specifically support for M.2 slot variants. However other adjustments to the board lay out to increase I/O would be welcomed.

[wizard69]

I have to agree with a lot of what yo posted here. I'm really hoping that when it comes time to do PI 5 they will honestly consider a board outline revision.

These SBC board manufacturers need to start standardizing more connectors. The Pi hat is a standard, and it's a very simplistic standard, but there's three problems with it:
1. Sspace on an SBC is very limited, and the connector is pretty big.

True but at this point you really can't get rid of it. It would become a legacy port in my mind. More importnaly we can't get hung up on board outline, the PI Foundation needs to reconsider the outline to support future capability.

2. It's designed for stacking boards, but there's a lot of situations where you don't want extra vertical height.

So? I mean really there is no zero space expansion system.

3. It's GPIO only.

That is an issue but the best way to deal with that is to support additional expansion through well defined ports.

Most SoCs used in SBCs have support for a lot of extra features that don't get used because:
1. It increases the base cost to mount connectors for these and design new board layouts.
2. Not every user *wants* the same connectors.
3. Manufacturing multiple versions of the same SBC just drives up costs.

There is no doubt that the original design goal for PI was extreme low cost. There is nothing wrong with that. What I think the community has discovered is that charging even a few tens of dollars more makes a huge difference in capability. $75 dollars for this new variant is still bringing a great deal of capability for the price.

So the trick with the next PI is to offer changes that make sense but don't blow that pricing budget all to hell.

I'd like to see some versions with smaller connectors and some multi-purpose "feature connectors" - a storage connector, a PCI-E connector, a USB connector, internal/external video connectors, etc. They can even be combined into high-density slots or connectors that expose multiple features, and they can ship little squid adapters to use them. Use of the connector requires you to route every feature you support. Any board that doesn't implement a feature available on the connector still exposes the full connector, but documents what's not supported and disconnects the pins.

The cheapest way to do this is to make use of card edge connectors. That way the cost is put onto any expansion card purchased. This would work for all non standard I/O. For standardized ports like USB, or HDMI you really need to make use of standard connectors on the base card.

Standardize supported distances for the data traveling over the connectors, build small PCBs that use the connectors that be mounted to the side or above, if you get the right set of standoffs for it. The small component PCBs would have standard sizes, and you could get mounting surfaces for stackers at cheap rates. Maybe we could even start shunting a lot of special purpose hardware to 1x PCIe lanes. Perhaps the module-style SBCs could get boards that run all of pins.

If you want to use PCI-Express lanes then you need to use industry standard I/O. A compliant card edge, most likely 4X, would be a good start after an M.2 slot. The last thing we need is PCI-Express going to a new card format nobody uses. Note I'm thinking along the lines of the PCI-E port being a card edge connector that would plug into a PCI express back plane. Again Zero cost to the main board with those needing expansion flipping the bill.

With this, people could start building special-purpose hardware with commodity components. Maker clubs and conventions could group-purchase the squids and addon boards. You could swap out the main SBC and the kernel to get newer features. People could buy build chassis adapter kits for servers to convert them to house several low-power ARM boards.

We need a way for SBCs to offer standardized features without having to mount much beyond power and internal connectors. This could reduce the implementation cost and the time to manufacturing. SBCs should be leading the way towards componentizing everything. And they need to design a means to expose new chip functionality without requiring a new SBC design.

Unless I misunderstand you the answer is No; componentizing everything is not the right move. You want to leverage the SoC as much as possible, offering up that capability in a way that doesn't drastically increase the cost of the main board. Again the goal should be keeping an eye on the bigger reasons the board exists in the first place - that is education.

Other things needed:
1. Power switch support.
2. Soft-off support so you can actually shut down the hardware without physical interaction.
3. Standardized quality board power connectors for power supplies.

This is actually huge!!! Power is one of PI's shortcomings. I'd actually like to see them adopt Intels new 12VDC standard adapted for external power supplies. Again though this impacts costs.

As-is, I don't know that you really save much money building a NAS out of these things. Yes, power costs are lower, but if you're only doing storage on your NAS and it's not super old, it should be using < 35 watts without drives anyway.

Watts are watts, if you manage a 16 watt NAS you have cut your power usage in half. With careful design I'm pretty sure a respectable 6 watt NAS could happen. Even then a good portion of the power budget would go to the SSD.

[oiaohm]

Designing in SATA on hardware of this type is just foolish at this stage. It makes about as much sense as an IDE or SCSI port at this point in time. Beyond that I suspect it would be far easier to find a chip with PCI-Express support than SATA. PCI has a future, SATA is dead in the water.

Not really Sata and SAS will stay around for quite some time for spinning storage as in HDD also for optical based drives.

There is a interesting issue with PCI-Express vs Sata/SAS.
Sata supports a cable length of 3.3 foot/1metre. SAS is 33 foot/10 Metre. Remember that 10 metre end of a SAS can be all SATA drives.

PCIe max cable length by specifications following is PCI version followed by length of run.
1.0 15inches.
2.0 12 inches.
3.0 8 inches.
4.0 you are not getting off the motherboard without repeaters. Yes some people have been able to push PCIe out to 3 meters are other stupidity but that is not by specification so is depending that you have the right combination of motherboard and card and extender.

Next the SAS/Sata cables are able to bend more than PCIe extenders can. For your mass storage Sata and SAS is most likely here to stay for a long time its not like spinning media is going to need more transfer speed than Sata and SAS can do any time soon.

Now of course I do kind of agree that SATA/SAS ports on something like a raspberry pi is getting close to pointless unless it is targeted as NAS market with spinning drives.

M.2 NGFF pcie would be more all round useful as you can get M.2 NGFF pcie to Sata/SAS controllers for the cases where you want SATA or SAS or put pcie based m.2 based storage in them. This is not that SATA is dead in the water because sata will have a usage case for quite some time to come. Just there is a more flex-able option. You find this on the back of the rock pi where they have a M.2 PCIe on the back so you can use SATA/SAS with that device by populating that slot with a controller.

Something to remember here to be able to boot from the sata controller the core card system firmware would have to support it. This is why I would be looking at the right out M.2 pcie on back and supported SATA and SAS controllers to go into it. This way you cover all users fairly well.

[tuxd3v]

RaspberryPi.. its almost closed source, its not mainline kernel..
So its competitors are way ahead, the 8GB are well received, but the rest is falling apart..

[Raka555]

The thing I would like to see the most, is a fully open graphics core.

[hotaru]

RaspberryPi.. its almost closed source, its not mainline kernel..

my Pi 3 is running a 64-bit mainline kernel just fine. the only closed source software on it is the bootloader and firmware for the VideoCore.

[lowflyer]

...
The current Raspberry Pi 4 only have 4 identical cores.

That's why they named it "Raspberry Pi 4"

[Terrablit]

2. It's designed for stacking boards, but there's a lot of situations where you don't want extra vertical height.

So? I mean really there is no zero space expansion system.

3. It's GPIO only.

That is an issue but the best way to deal with that is to support additional expansion through well defined ports.

Agreed and agreed. But what I'm talking about is designing expansion that's not fixed in one axis. Expanding up on a flat board is great, except:
1. it blocks cooling solutions (another point I had in mind that I forgot to mention. we need space on the board for mounting real cooling)
2. limits expandability via multiple hats unless they're really good about passthrough and have compatible pin usage.
3. sometimes we don't have vertical space, but we have lots of horizontal space. Like in a laptop. Mounted behind a monitor, or inside a desk.

A lot of the low-bandwidth HAT solutions could be a separate board connected via I2C or SPI buses instead of a gigantic connector with a large pitch. High bandwidth solutions could multiplex on PCI-E lanes.

That GPIO connector was designed with physical durability constraints in mind, and that's fine. But a general purpose solution needs to also consider available board space and signal quality.

What I'm thinking of is sort of like a riser card/daughterboard style thing that lets you use all the standard connectors, but doesn't require them to all be on the board with the SBC. If you look at a lot of hobbyist attempts to miniaturize consoles (like all the old Ben Heck projects), sometimes they'd cut the boards into pieces and use wires to keep them connected to get them into smaller cases. Similarly, ports in a PC case can sometimes be quite a distance away from the board.

We've got two considerations:
1. How do we route all the chip functionality without making the boards grow? We're out of edge space and tight vertical space isn't that great when we start putting in faster processors.
2. How do we let hobbyists/makers use all the chip functionality they need without gigantic boards or non-standard board sizes and layouts? We've still got to mount it somehow.

That's why I was thinking cards, ribbon cables and squid adapters for the things that can still work well at some distance.

Ports that are intended to go outside a case (serial, HDMI, USB, RJ45) and connectors for things that can have longer cables (SATA) or run at lower speeds (GPIO, serial) can probably be removed from the main SBC and routed through another connector. While things like PCIe and M.2 slots can't be moved too far away and are probably best suited for being mounted on the board itself or a companion solution that explicitly controls the distance.

The cheapest way to do this is to make use of card edge connectors. That way the cost is put onto any expansion card purchased. This would work for all non standard I/O. For standardized ports like USB, or HDMI you really need to make use of standard connectors on the base card.

Yeah, card edge connectors are in line with what I was thinking. I've seen some of them in weird places, but I've not seen many of them in cross-vendor computing standards. Though I do think standard ports might go out there as well. Like, if you can get the connection stable, you could maybe put storage and networking on one, audio/video on another. and peripherals and GPIO on a 3rd. Maybe you could even get PCIe going through a connector. Right-angle connectors could still let you expand upward if you wanted. The HAT style could be some spacers in the screw slot and a cable connecting the edge connector to the board on top. Flexible cables are nice to give you freedom to reposition, but there's always the risk someone gets a bad cable, or uses one that's too long, or abuses it until the wires break.

If you want to use PCI-Express lanes then you need to use industry standard I/O. A compliant card edge, most likely 4X, would be a good start after an M.2 slot. The last thing we need is PCI-Express going to a new card format nobody uses. Note I'm thinking along the lines of the PCI-E port being a card edge connector that would plug into a PCI express back plane. Again Zero cost to the main board with those needing expansion flipping the bill.

Yeah, like that. A 4x connector can often be split into 4 separate 1x lanes. The card edge connector would be a new standard because no one makes PCIe devices like this, but in the end it's likely to lead to a backplane of some sort that exposes the actual standard.

This is actually huge!!! Power is one of PI's shortcomings. I'd actually like to see them adopt Intels new 12VDC standard adapted for external power supplies. Again though this impacts costs.

Yeah, higher voltage would drop the current a lot. And USB-C can supply up to 20V. Rather than USB-C power on the main board, we'd have a dedicated connector for it that can be placed elsewhere. Which could in the end be a USB-C connector, but it's not on the board so there's more freedom in what's used. The separate power supply could offer 5V or 12V to the board depending on what it needs. There are some pretty good solid state power supplies at affordable prices that can handle up to 48V input and output 3v, 5v and 12v. Finding good quality 5V USB adapters can be a lot of trouble, so I wouldn't mind disrupting that. I'd rather buy a sleek 120w PSU than take my chances with rando counterfeit shit that I'm not supposed to open.

In the end, we'd have SBC manufacturers making boards with almost nothing on the board itself at much lower prices, and the hobbyists would have to spend the saved money on connecting what they needed. A minimal storage connector could just offer an SD card or a SATA port. Minimal A/V could be an HDMI port and stereo jack, and minimal peripherals gives you 4 USB 3.1 ports on a board with a separate power connector if you want to add it in. And a power switch somewhere. The total cost for everything might be higher, but the connector boards could be reused between models, or case solutions could be sold with mounted connectors leaving hobbyists to find the brain that works. I mean, you can get ATX aluminum cases with switches, USB and audio ports on them shipped for $30. There ought to be a way to get something useful in 1/4 the size.

And it's not like the RPi doesn't get massively expensive when you buy the special camera and screen and put a HAT on top to get the extra functionality you want bodged in through USB or GPIO. I'd also like to think that splitting this all out reduces the chance of having to scrap the board (or resolder) when the USB connector gets loose.

I think if you mandated that edge connectors, when used, need to be on a certain side of the board and in a specified order when it's a certain form factor, it would work.

It's not going to end the current style of SBCs. Some people are going to want everything on a small, immediately-usable board. But I think it'll get newer, more powerful SBCs out of the design stage a lot sooner, and raise the utility limit on these devices. It could make offering a development board for an SOC *significantly* cheaper. It'll open up a new market for expansion and case manufacturers. The biggest difficulty will be getting everything to mount properly when you hodgepodge a bunch of expansion components. But chances are if you're willing to do that, you either don't mind it looking shameful or are willing to build something yourself.

Unless I misunderstand you the answer is No; componentizing everything is not the right move. You want to leverage the SoC as much as possible, offering up that capability in a way that doesn't drastically increase the cost of the main board. Again the goal should be keeping an eye on the bigger reasons the board exists in the first place - that is education.

Kind of. I want SBCs to have the expectation to wire things up, and to standardize around having internal connectors rather than expecting to fill one edge and stop trying. I want the connectors to be componentized. But you may be right in that the board itself needs to have some minimal connectors for interacting. Or at least it could ship with an optional right-angle edge connector that has them on it. But honestly, I think it's two different markets. Some want to tinker with what's there. Others want to buy a low-power SoC and wring the last drop of potential out of it. Those people are just forced to buy what's available. I think the RPi is going to keep to the current market. But I'm hoping all the vendors trying to jump on the Pi wagon could consider a new market like this with endless flexibility. Many of them are using far more powerful chips, but trying to compete with the Pi and maintain the form factor and HAT compatibility means they lose a lot of the advantages they could have.

Watts are watts, if you manage a 16 watt NAS you have cut your power usage in half. With careful design I'm pretty sure a respectable 6 watt NAS could happen. Even then a good portion of the power budget would go to the SSD.

Yes, I just think it's a scale with diminishing returns. If all you want is always-on storage with no concern for speed or anything else, yes. But if you're hoarding large amounts of important data, you're going to want redundancy, backups, applications, etc, and you could either split that into tons of devices, or get a more powerful one that does a lot more and uses more power. It's a trade-off, for sure. I just think it'll be hard to compete in low-power design against the commodity ARM/Atom NASes by building your own with commodity hardware.

I am a bit biased, though, as I'm getting ready to repurpose an old Supermicro Super Storage server case for my growing data needs.

[oiaohm]

Yeah, like that. A 4x connector can often be split into 4 separate 1x lanes. The card edge connector would be a new standard because no one makes PCIe devices like this, but in the end it's likely to lead to a backplane of some sort that exposes the actual standard.

The reality is give me a M2 PCIe and I can do a lot.

https://www.amazon.com.au/ADT-Link-Riser-Gen3-0-PCI-Express-Extension/dp/B07Z4S18RB

This to get it out to a standard pcie 4x connector from a M2 PCIe.

https://www.amfeltec.com/gpu-oriented-x4-pcie-4-way-splitter/

Then something like in the pice 4x connector giving you 4 pcie 16x slots that are basically 1x slots. Normally you would not go this far as you are getting more expensive than a AM4 motherboard with one of the cheapest cpus with less functionality at this point.

So Terrablit you were wrong that exactly needs a backplane. The pcie slots holding full pcie cards will need more power than the raspberry pi and most small boards can provided. Splitter cards for pcie are not really cost effective.

The advantage of M2 PCIe is that is does not have the requirement to provide a hell load of wattage.

Of course if you don't go the M2 route the other option is the thunderbolt port route.

A item I would love to see is M2 to Thunderbolt3 /usb4 that way thunderbolt/usb4 enclosures could be used.

For a compact board like raspberry pi I see the USB4 or M.2 routes as most productive as you get most functionality with the least surface area usage and without blowing the power budget.

USBC is fun So USBC could be powering something like raspberry pi from the enclosure that is holding the drives and sata cards that the like Raspberry Pi is accessing for data from. Of course this also means you could shutdown your nas done by this board and directly connect you laptop/other device up for full speed access as well.

I don't see pci express slots directly or sata or sas on something like a raspberry pi as right its all due to the power you need to drive full standard pci express slots or SAS fully. Items like M.2 pcie and USB4/Thunderbolt 3 that can provide PCIe functionality external get around this power limitation . This is also why I don't exactly see sata ports in a raspberry pi like device as useful either particularly if you are wanting to use full 1 metre sata cables. Its really simple to run out of power budget on these small boards and crash due to out of power.

Generally NAS boards are a little larger than raspberry pi boards to allow for bigger power traces. Compact comes with its fair share of issues of course that does not mean there are not options to-do a lot in compact design sometimes it admit at X size something is just not suitable done that way as you will run out of power budge so you have to go another way that avoids running out of power. More than one way to skin the cat.

PCIe Sata or SAS card in the "ADT-Link Riser PCIe X4 3.0 PCI-E 4X to M.2 NGFF NVMe M Key 2280 Riser Card" would have the power to fully function. PCIe Sata or SAS card in a Thunderbolt3/USB4 enclosure would also have the power to fully function.

Remember one of the use cases for the raspberry pi is running on battery in fairly compact way. So you really don't want to be constantly spending power fully power Sata, SAS or PCIe slots.


M2 pcie to Sata controllers normally cannot do the as per standard sata 1 meter cables due to lack of power budget.

Its really simple to wish for things and not think about what the power price to-do them is. I don't see Sata/SAS as dead yet but I also don't see them as a good fit to be built on something like a Raspberry pi due to the power you need to do them properly.