Posting this on the RPi forum doesn't make much sense..
You won't find many 'haters' here..
and I'm quite confident that most of them will visit the RPi forum due to various reasons.. But if you ask for some counter argument.. Okay, I can do it.
gtechn wrote: ↑
Wed Jul 03, 2019 9:20 pm
Why no eMMC option? The CM3 has this!
: According to Mouser.com, cheap 32GB eMMC starts at about $22. More expensive, high quality eMMC is $44 for 32GB. That's more than the Pi 4 costs by itself. Even if it is offered as an option, running multiple SKUs of any product is a logistics nightmare. You'd probably have the Pi 4 1GB-RAM/No-eMMC, Pi 4 2GB-RAM/No-eMMC, Pi4 4GB-RAM/No-eMMC, Pi4 1GB-RAM/16GB-eMMC, Pi4 2GB-RAM/16GB-eMMC, Pi4 4GB-RAM/16GB-eMMC, Pi4 1GB-RAM/32GB-eMMC, Pi4 2GB-RAM/32GB-eMMC, and Pi4 4GB-RAM/32GB-eMMC. Ouch and expensive to produce, and if RPT overproduced one SKU, that could cause significant financial damages.
: There is no space on the PCB for eMMC even if RPT decided to add it. The Pi is actually a six-layer circuit, and spent over 6 months in design being ~5mm too wide because of the incredible difficulty in getting the current circuit to fit inside the Model B form factor. A brand new form factor would be necessary, but according to a recent interview, the RPT has no intention to launch a hypothetical "Model C" anytime soon. There is no way to fit an eMMC chip onto the Model B form factor without adding more layers to the PCB's 6 - which is currently very expensive.
: One of the beauties of SD Cards is that you can develop your project on a top-of-the-line Pi 4, take the SD Card out, and plug the SD card into a power-efficient Pi Zero. You also can mass-copy an SD card onto, say, 30 cards for classrooms. Not with eMMC.
: If someone damages an SD card due to heavy wear, it is replaceable. With eMMC (which is not necessarily more reliable than SD cards), you have a Pi 4 with bad storage. Can you imagine a Raspberry Pi wearing out over time? That's something that current Pi's don't really have a problem with. Industrial customers might not care - but students and teachers in schools will not be pleased with boards that can permanently wear out if overused. This could also lead to schools preventing students from trying certain projects for fear they would wear out the board.
: If an eMMC is corrupted, repairing it is difficult. Bricking a Raspberry Pi permanently is currently very difficult - just download a new SD card image, good as new, no matter what stupid thing you did (unless you overclocked to oblivion). With eMMC, you could, in fact, potentially put a Pi in a situation which might need professional Linux expertise to repair.
IMO the whole paragraph is a non argument at all.
And I'm quite sure @jamesh will happily disagree with my opinion cause we had this discussion already..
1. first mouser isn't the reference if you look for eMMC prices in bulk, and since the RPi would probably sell x (and compared to the competitors x is likely a huge number) boards.. bulk prices are of importance here. If the SoC allows it (means it has at least 3 SDIO interfaces), this can be solved as every recent design does it. Means you've a socket for eMMC, an SD card interface and the last one for your SDIO wifi chip. You sell different eMMC modules and everyone can pick whatever he needs/prefers. From personal experience.. every eMMC I ever had on a SBC outperformed SD-cards (even the 'good' SanDisk A1 rated cards). So you add a few cents for a connector which might not be always used (similar to the DSI and CSI).
2. Somehow one competitor manages it on the same formfactor to have a SD card interface, place for an eMMC module and an m.2 slot. How dare they are... and the board has dual channel ram and a bigger SoC.. these wizards... (and yes CSI and DSI is also routed)
3. swap the eMMC module, right it wouldn't work cause the VC4 based boards have for sure not 3 SDIO interfaces..
4. see 1. replace the module.. and software can be optimized to lower permanent writes.. but mostly.. see 1.
5. actually the same as 4 isn't it? then.. see 1. Most SoCs have a bootorder e.g. the old Allwinner chips (H3) goes SDIO (first the one on which the SD card hangs, then the eMMC one) --> SPI. So the whole thing is unbrickable.. If you mess up eMMC insert a SD card and rewrite it (actually the reason why all those cheap H3 tv boxes can be hacked so easy).. Others like rockchip allow maskroom mode http://rockchip.wikidot.com/how-to-ente ... skrom-mode
also not brickable.. No idea what the BCM chip will offer here.. but I assume things are similar otherwise the 'recovery.bin' (https://www.raspberrypi.org/documentati ... teeprom.md
) wouldn't exist..
And no, non of them needs professional Linux expertise...
The whole eMMC thing stands and falls with the SoC. If it has 3x SDIO, eMMC would be possible in a flexible way.. If it has only two.. we'll likely never see eMMC on RPi except the CM modules.. Cool thing for the boardmaker, on eMMC modules he can be quite sure that the user don't buys cheap unreliable crap (on SD cards a bunch of users do it), cause the market is not big enough for knock offs (hopefully if RPi ever opts for a eMMC socket, use a different connector as the competitors, otherwise the market might be big enough for knock offs..
gtechn wrote: ↑
Wed Jul 03, 2019 9:20 pm
Why no eSATA? This would be great for NAS!
: The SoC has no built-in support for eSATA, and Broadcom (as far as we are aware) has no licensable eSATA blocks for mobile chips. This means either spending easily
over a million dollars (this is not hyperbole - more like several million) to add SATA support to the SoC. The other option is to add a SATA to USB 3.0 adapter on the board, but that costs money and has the exact same performance and chip as a HDD enclosure you can buy on Amazon for $8. So, with little demand from the community at large from eSATA and the fact that you can get equivalent performance on the cheap using an enclosure, why pay to build it in?
: There is no space on the PCB for an eSATA to USB 3.0 conversion chip even if RPT decided to add it. The Pi is actually a six-layer circuit, and spent over 6 months in design being ~5mm too wide because of the incredible difficulty in getting the current circuit to fit inside the Model B form factor. A brand new form factor would be necessary, but according to a recent interview, RPT has no intention to launch a hypothetical "Model C" anytime soon. There is no way to fit an eSATA to USB 3.0 chip onto the Model B form factor without adding more layers to the PCB's 6 - which is currently very expensive.
Further on this tangent, let's say that Broadcom/RPT spent the cash to build an eSATA connection into the SoC. That eSATA Connector has to go somewhere on the board, and considering that there wasn't room for a full size HDMI port or even room for two mini HDMI ports, there probably isn't room for an eSATA port either, without (again) a new form factor.
Some good news: Shutting down your Pi correctly greatly reduces SD Card damage, Premium SD cards (+$5) last much longer than cheap ones, and the SD card bus is now TWICE as fast in the Pi 4. You will also get PXE and USB boot soon.
1. you mean PCI/USB/SATA bridge on board? sounds like a cocking device..
(all those PHYs tend to add heat..
). What about exposing PCI on some sort of a pinheader (as friendlyARM does) and selling a USB3 and a SATA hat?
IIRC the VC5 (never used in RPis I know) had PCI and USB3 so I assume VC6 should offer this too. Why was the 'On-SoC' USB3 not exposed? PinMuxed with PCI? As crippled as USB on VC4 was in the beginnings?
2. See 2 on the part above.. Somehow they managed to expose more interfaces on their board e.g. a NVMe capable m.2 slot..
I don't think SATA should be part of the RPi, cause you're right there's no native support on the bcm chip for it (the only cheap arm SoC with native SATA I've in mind is AllWinner A20). And SATA won't be use-case Nr.1,. PCI exposed to some sort of a pinheader would actually be a nice idea. Like FriendlyARM with their 'own' connector or as more or less all the others (mPCI/m.2). Offering a 'USB3 hat' and a 'SATA hat' making the Pi a bit more modular.
For the whole RTC stuff, I really don't care.. If you want one.. The cheapest i2c based ones go for 1-2$ 'RPi compatible' on aliexpress.. But if this would add 10-15$ you should hire a new engineer..
The microHDMI was actually a smart pick.. Exposing both HDMIs is something that to my knowledge none of the competitors can offer at the moment.. Let's see how performance works out on those once it's properly supported..
gtechn wrote: ↑
Wed Jul 03, 2019 9:20 pm
Why not 4 USB 3.0 ports instead of 2?
thanks god they didn't put 4 USB3 on the board. The nightmare of insane settings with adding 4 HDD/SSD powered over the RPi all suffering from underpower issues would be a nightmare. Even on 2 SSDs without additional powering we see first hints that it might not work reliable at the moment: https://www.cnx-software.com/2019/06/24 ... ent-564128
I assume it's mostly due to not enough juice to power two additional drives and the PHYs needed for USB3-SATA but honestly not tested on my own. I'm not interested in RPi4 as a NAS at all (I've a dedicated SBC for that job which just works fine).
The whole barrel jack discussion is somehow... well.. we'll never see a RPi with it so it doesn't make sense to discuss this really.. I've multiple SBCs with barrel jack (5 and 12V) and I can somehow manage it to don't connect 12V to the 5V ones..
I think most of your user base should be able too..
For the cellphone chargers.. well that's then the same story as RPi on microUSB cellphone chargers.. Some work, some won't work reliable. And for those it doesn't work, they'll allways blame the software, not their hardware setup..
IMO the combination of USB-C with a PMIC supporting USB PD would have been a great idea.
Towards UART, I've roughly 10 USB UART dongles laying around to debug my SBCs in case something goes wrong and I love SBCs with a dedicated UART pinheader only for debug purposes.. It's simply a way more infos once your board doesn't behave the way it should.. Especially during the time you try do get a new one up and running...
Why no model C.. For me, the Pi4 is a model C cause not the same formfactor anymore (swapping ethernet and USB) well.. it should actually be model D cause between RPi1 and RPi1b+ the formfactor also changed.. But well.. I don't care about names..
Overall the RPi4 seems to be well crafted, mine arrived today and I probably start tomorrow with some trials with the 64 bit kernel just out of curiosity but I don't think the argumentation is rock solid.. Other boardmakers showed that on the same (RockPi) or smaller (NanoPi) formfactor you can expose a bunch of interfaces. eMMC can be designed in an unbrickable way, support for other storage solutions (SATA, NVMe - I know the SoC doesn't support it) can be implemented in a modular way and powering over barrelplug is also not really a problem (at least I don't spot many topics that people fry their SBCs by using the wrong one