The Raspberry Pi Foundation has just introduced a updated version of their popular Raspberry Pi 3 Model B board with a Model B+ that increases the processor clocked up to 1.4 GHz, adds dual-band 802.11ac WiFi, and Bluetooth 4.2, Gigabit Ethernet (via USB 2.0 to Ethernet bridge), as well as support for PoE (Power-over-Ethernet) via an external HAT add-on board.
Raspberry Pi 3B+ specifications:
- SoC – Broadcom BCM2837B0 64-bit ARMv8 quad core Cortex A53 processor @ 1.4GHz with dual core VideoCore IV GPU
- System Memory – 1GB LPDDR2
- Storage – micro SD slot
- Video & Audio Output – HDMI 1.4 and 4-pole stereo audio and composite video port
- Connectivity – Gigabit Ethernet (via Microchip LAN7515 USB 2.0 to GbE bridge), maximum throughput 300 Mbps), WiFi 802.11 b/g/n/ac and Bluetooth 4.2 LE (via Cypress CYW43455 based module)
- USB – 4x USB 2.0 host ports, 1x micro USB port for power
- 40-pin GPIO header
- MIPI DSI for Raspberry Pi touch screen display
- MIPI CSI for Raspberry Pi camera
- Power Supply
- 5V up to 2.4A via micro USB port
- PoE via HAT board
- Dimensions – 85 x 56 x 17 mm
The new Raspberry Pi board is said to improve PXE network and USB mass-storage booting, as well as thermal management thanks to a heat spreader placed on top of BCM2837B0 SoC.
Since they’ve only done a few tweaks, software support will remain the same with NOOBS, Raspbian, and the many other available distributions.
The Raspberry Pi 3 Model B+ is now sold for $35 on Element14 or RS Components, and will remain in production until at least January 2023. The PoE HAT is available on backorder for $20 with stock expected for April 9, at least on Newark.
Jean-Luc started CNX Software in 2010 as a part-time endeavor, before quitting his job as a software engineering manager, and starting to write daily news, and reviews full time later in 2011.
81 Replies to “$35 Raspberry Pi 3 Model B+ Launched with a Faster Processor, 802.11ac WiFi, Gigabit Ethernet, and Optional PoE”
Amazing. The RPI foundation keeps addressing the non-issues, while neglecting the actual issues.
So they added more high-speed interfaces to share the single USB 2.0 bus?
But you get shiny bits of metal now as well…
It’s quite impressive they manage to retain the pricing though.
That’s the cost of software compatibility…
But it’s not 100% compatible across the different models though, since they moved from 32-bit to 64-bit with the RPi 3.
I think this is more of a ecosystem issue, i.e. housings, hats etc.
Dang, fanless aluminum case not supported, different layout for components.
2GB RAM + improved Micro SD card (faster write and reads by default) are missing 🙁
Wish they instead upgraded LAN7515 to LAN800 to provide USB 3.1 Gen1
I’m not sure why they should upgrade to USB 3.1 when the SoC is stuck to one USB 2.0 for all the devices.
Exactly what I was thinking. I was really hoping for giga Ethernet, but instead it’s half-baked. I hope this is a stepping stone to the Pi4 with “proper (not shared)” giga Ethernet.
I don’t think RPF ever moved to 64bit — raspbian is the bastion of 32-bitness, and understandably so, given the amount of effort it would take the foundation to move to 64bit (VC-what?), and upgrade their boards to something above the gargantuan 1GB…
For hardware ecosystem compatibility, they could have keep the same layout with a better SoC like their competitors do.
So disappointing, and this thingy suppose to last to 2020 or later… for the future release of the Pi 4 *cough cough*
By the look of things they might not even getting 2 GB RAM. More like 1.5GB RAM LPDDR2.
By the time they release that, most of the other chip maker will be on 4GB standard. And if you can spend $50 on top of the Pi 4. You can get something the Pi is getting in 10 years time. Like can run Windows.
Yes, indeed. Hope they can better the SoC soon
Nice, so I were right back in January forecasting ‘most crappy Gigabit Ethernet’ ever implemented soon by RPi folks: https://forum.openmediavault.org/index.php/Thread/18991-New-approach-for-Raspberry-Pi-OMV-images/?postID=166467#post166467
Folks, please understand that all Raspberry Pi since the first one sold 6 years ago are all more or less the same. It’s an old an boring 32-bit processor called VideoCore IV that got one or more crappily integrated ARM cores to the die. All the limitations (only 1 x USB2, only 1 GB DDR2 and so on) are due to this and RPi Foundation can’t change that since they would loose their only asset: backwards compatibility.
Not easy. When they come up with a new SoC they loose backwards compatibility which none of their customers will be able to understand. They would either need new OS images or they can convince Broadcom to add the anachronistic VideoCore IV bootloader crap to a decent SoC design (which I doubt).
Software wise it’s 100% compatible since they let the Cortex-A53 cores remain in 32-bit mode. And while a lot of people continually ask for 64-bit RPi support that’s both challenging on this platform (since the main CPU is the old and boring 32-bit VideoCore IV and not the ARM cores! No idea why no one wants to understand/accept this) and also somewhat stupid when the SoC is limited to laughable 1 GB DRAM (since 64-bit software eats up more memory while only providing negligible performance improvements): https://www.raspberrypi.org/forums/viewtopic.php?f=91&t=192321&p=1281326#p1281326
The VideoCore IV has only one single USB2 OTG port and this will never change as long as RPi Foundation relies on this platform. What do you want to do with peripherals that are SuperSpeed capable if 1) the host is only capable of ultra slow ‘Hi-Speed’ and 2) all USB receptacles and the Ethernet chip have to share the horribly limited bandwidth of the one USB2 port?
The average RPi buyer still doesn’t realize this ‘shared bandwidth’ limitation, now reads ‘maximum throughput 300 Mbps’ and thinks a RPi would now make up for a nice NAS since +30 MB/s possible. No way, it’s less than 20 MB/s under best conditions since every single bit has to pass the ‘single USB2 port’ bottleneck twice. The horribly outdated SoC is the problem, not the peripherals 🙂
Even if they now again slightly exchanged stuff related to the ARM cores the BCM2837B0 is still the same old and boring VideoCore IV that was designed in 2011 or maybe even 2010.
Which product can you recommend that offers form example a PCI Express slot and USB3 ?
Hopefully this case atleast is supported (by dimensions atleast):
I looked the for oldest VideoCore IV press release I could find. It was first announced in BCM2763 in December 2009.
Any Raspberry Pi case will the supported. They did not change anything to the layout.
> The average RPi buyer still doesn’t realize this ‘shared bandwidth’ limitation,
The average RPi buyer is not *YOU* @tkaiser.
You’ve never understood or appreciated the market the RPi is aimed at, so why do you (and others of your ilk) routinely bore everyone to death with your SBC “insight” and demands that the RPi should be a frickin’ supercomputer for $35?
If the RPi (even this 3+) isn’t for you, why not just ignore it, eh? Your rants reveal you to be such a fanboi.
Anyway, heatspreader on top of SoC doesn’t fit with that fanless type.
And LAN component is moved.
Maybe in Raspberry Pi 4
The Flirc cases maybe not.
Depends always on the use case of course. If headless operation is an option you get both (and real Gigabit Ethernet + real SATA) on an EspressoBin for example. With ‘general purpose’ use cases in mind maybe one of the many (soon to be released) RK3399 boards fits your needs?
But the Raspbian OS is still 32 bits, so in terms of software compatibility, the change of word length does not change a bit!
I fear that the addition of the new wifi module and the GbE PHY in the USB hub will make the power problems of the RPi even worse… also while before you had more or less no contention issues with the Ethernet on USB due to it’s low bandwidth, the network interface can now saturate the USB bus and if you have some storage class device attached – it will be probably worse than before if there is no traffic shaping for mass storage/ethernet…
Why should they choose USB Ethernet in an upcoming incompatible RPi 4? If they will ever move away from a SoC designed in last decade (VideoCore IV) to one that has been designed in this decade then why should this new SoC be that limited as the VC4 is? If they switch to an incompatible modern SoC I would assume the feature list contains native Gigabit Ethernet, a memory controller capable to deal with more than 1 GB DRAM (DDR2) and everything else that’s more or less standard today on ARM SoCs.
But that’s exactly the problem they face: moving to another SoC means becoming incompatible to the existing one which also greatly affects software support situation (one of the results of VideoCore IV being an old design is of course mature software support today. Choose a new SoC and this might just look like it was back in 2012 with VC4)
I would suggest to use 3A atleast (even they say 2.5A should be okay)
Flirc case is okay, as only SoC is covered.
This is exactly what’s happening with external Gigabit Ethernet adapters connected to an RPi today. As soon as the access patterns affect disk and network at the same time performance drops drastically compared to Fast Ethernet. This affects tasks like opening/closing small files as well as overall sequential write performance dropping from ~9MB/s with Fast Ethernet to lower numbers with GbE. Though as long as the writes go to filesystem buffers/caches this is not noticable of course and up to 30MB/s are then possible:
(please not that the same effect happened in this benchmark. The ‘read performance’ is just pushing data from RAM to network since the test size was too small and the entire 300 MB fit into the 1GB DRAM)
But I would assume this problem gets addressed maybe even at chip level inside the LAN7515?
Why? You would need to increase input voltage maybe to even 6V to be able to feed that high amperage to an RPi: https://www.cnx-software.com/2017/12/05/libre-computer-renegade-sbc-features-rockchip-rk3328-processor-with-up-to-4gb-ddr4-ram-crowdfunding/#comment-549684
BTW: The Micro USB connector is still rated for 1.8A maximum (for a reason). I would fear to burn my fingers touching Micro USB jack or receptacle when trying to get 3A through those laughably tiny contacts 🙂
Gigabit over USB2 bridge… something tells me this will be slow especially when you are also accessing other USB drives/devices. Looking forward to seeing benchmarks.
Headless for OpenWRT / Linux. Im looking at using CJDNS (either with SoC encryption support or for example a Silicom PESC62-RoHS Security Protocol Processor adapter – will need adapter to convert to PCI-E x4)
So will maybe try a Gemeni Lake X86 baord (NUC or Motherboard) also
So will look at the ESPRESSObin (it offers OpenWRT support)
Thank you for your recommendation
Well, the older RPi 3 does not support ARMv8 crypto extensions (AFAIK the only two 64-bit ARMv8 SoCs that miss this feature are Broadcom BCM2837 and Amlogic S905 on ODROID-C2 and NanoPi K2) which can have a huge impact on AES performance if this matters for you. Without those crypto extensions AES performance is magnitudes lower than with, see for example https://forum.armbian.com/topic/4583-rock64/?do=findComment&comment=37829
The SoC on the aforementioned EspressoBin supports both ARMv8 crypto extensions and an own Marvell proprietary ‘5Gbps Security Engine’ — mainline kernel support has been added recently and you find some information here.
Wrt BCM2837B0 on the new RPi and ARMv8 crypto extensions I thought I simply ask those who could know already: https://www.raspberrypi.org/forums/viewtopic.php?f=63&t=207888
The other reason they do not change the SoC is that no other SBC builder can get the SoC. It ties, padlocks the buyers into RPI software, hardware and excludes competition. Other SBC maker have to use other SoC brands, designs.
If I recall correct, the original SoC which inspired RPI is a media player SoC ( think Roku box or bottom end Now TV box , white original. Low specs).
Hence the original was never designed with more than one USB in mind. Bits have been bolted on and parts tinkered with, hence it is a home brew monster now.
Only the exclusive use of the SoC give RPI their market, IMHO
@manuti The flirc case is a slightly tighter fit than before as the heat spreader adds about 0.5mm, but it fits. I’ve been using the Kodi case with a test sample for the last month.
Me too: https://www.raspberrypi.org/forums/viewtopic.php?f=63&t=207897
There’s also the Clearfog Base which is pretty decent and comes with an enclosure now. The price is of course much higher but might be OK depending on your use case.
Well, with this product not improving over time, the gap between its capabilities and all other boards is significantly increasing. People want to use it for everything so it needs to be generally decent. Nowadays, 1GB RAM makes it too small to run a browser. The outdated CPU cores make it too slow for computer vision. The shrinking I/O bandwidth makes it out of question for any server-like activity or a basement NAS. This new optional fan makes it unsuitable for any domestic use case. And the 2.4A over micro-USB (1.8A max) will cause lots of spurious freezes and reboots making it painful to troubleshoot for newbies. What remains ? Blinking an LED ? That’s easier done with an Arduino and it really teaches you much more about computer science. I’m sorry but @tkaiser is right. RPi is from the previous century and while it has attracted a huge amount of people to embedded products, it has also made many of them believe that ARM-based products were unsuitable for anything serious, which is sad considering the large choice of serious competing designs now.
Yep – though the RPi foundation’s core aims are still education – not flogging hardware for profit.
The Raspbian ecosystem with stuff like NOOBs etc. usually ‘just works’, and has a much better ‘novice experience’ than a lot of the other ARM boards which are less well developed in software terms.
The huge numbers of Pis out there compared to pretty much every other SBC, and the massive number of developers, means you end up with a much better experience in lots of areas.
I’ve got ODroid C1, C2, HC1 (and had a U2). I’ve got a Rock64. All these boards are more powerful – but they have a lot more sharp edges to cope with too… (The amount of time it took to sort an HDMI issue on the C2 was terrible. A similar issue on the Pi was solved in days – not months…)
Adding better (and 5GHz) WiFi makes a lot of sense (and that isn’t sharing the USB 2.0 bus), though the GigE stuff is a bit more contentious I guess (but in some ways better to have it than not have it). If the heat spreader removes the need for a heatsink in more cases that’s also a good move. Allowing PoE to be added simply is also a neat solution – particularly coupled with network PXE booting.
Well, the fan is really optional and based on what they showed should really not be needed. They closed the gap to cheap ‘competing’ devices by copying their design: the RPi 3 B+ is said to use the PCB’s ground plane as giant heatsink for the first time and they also use ‘fancy’ stuff like an I2C accessible PMIC (that’s what we’ve seen on an $15 Orange Pi PC already years ago). Quoting James Adams, Chief Operating Officer, Raspberry Pi: ‘The MxL7704 has allowed us to improve both the output currents and efficiency of the power rails on the new Raspberry Pi 3 Model B+ as well as reduce the overall cost of the power-chain by removing the need for external sequencing and monitoring components (as compared to the Raspberry Pi 3 Model B).’
So this new RPi implements sane DVFS for the first time and thus can reach higher clockspeeds for a longer time even without a heatsink (since the PCB design helps now too).
Yep, this makes me sad too, especially when thinking about OMV or the ‘NAS use case’. Approx. 2000 RPi users download each month my OMV image not realizing that they chose the worst board possible for this use case. And in the forum exactly the same people then argue that it’s x86 for great NAS performance since ‘ARM sucks’ (while ‘RPi sucks’ is the correct term).
Currently thinking about whether to ‘upgrade’ the OMV image for the new RPi 3 B+ (the new voltage regulation requires a new ‘firmware’ of course so that the VC4 can talk to the PMIC via I2C and then some proprietary stuff — safe_mode_gpio=8 — has to be added to /boot/config.txt to instruct the totally undocumented RTOS running on the primary VC4 CPU to do whatever magic). Maybe a better idea is to simply drop support for the entire platform…
Lucky guy having test units!!!
I see the April Fool hoaxes started early this year. This one is the best ever!
I’ll expect this to sell more units for pure NAS use than all the competing dev boards in total.
So the results are: ‘new’ BCM2837B0 lacks ARMv8 Crypto Extensions as before so everything AES related will perform really poorly. It’s also funny that the openssl benchmark numbers done with RPi 3 B+ are much lower than those from a RPi 3. But that’s (not so) surprisingly just the result of the RPi (under)powering sh*t show related to Micro USB. Thank you RPF for again ignoring the problem instead of fixing it. The recommendation for a Micro USB PSU with 2.4A is just insane given the countless RPi users running frequency capped due to crappy Micro USB and suffering from undervoltage under load.
Yes, me too. And that makes me really sad since I care about users and not RPi Trading’s sales…
Videocore IV is gonna be proclaimed as the successor of mali 400 lol
I wonder why they don’t just do the migration is a seamless way, releasing a board with the compatible stuff and an alternate one with a new shiny SoC to start creating a solid base. After a couple of years it will have almost everything fixed and they will be able to cease releasing old ones. Simple and clean.
Then probably that having a page showing the NAS performance of a few boards with their price will help users realize that the board they choose is the problem, not the architecture. Also maybe a warning at boot on OMV could be instructive, something like “Warning: Ethernet is shared with USB2, this board is not suitable for use as a NAS, please check this url for alternatives”.
Sadly, many competing designs lack an enclosure, making it a big pain to assemble a finished product.
Not surprised if they’re running in 32-bit mode, then they’re booting in armv7 mode and the extensions are not exposed. It was the same with the nanopc-t3 (octo-A53) with older 32-bit only kernels and was automatically fixed with the recent ones booting in 64-bit mode.
The VC4 is not just some random 3D GPU like old and boring Mali 400 but the main CPU on every Raspberry Pi running a proprietary realtime OS that controls the hardware. Completely proprietary and not documented but essentially defining the ‘RPi experience’. If you try to boot without ThreadX (that’s the RTOS called ‘the firmware’) close to nothing will work any more, if you try to avoid all the proprietary VC4 stuff contained in Raspbian it gets pretty boring wrt the usual ‘RPi experience’ (no HW accelerated video decoding, no 3D stuff and so on) and even the whole GPIO stuff is totally focused on the way Broadcom did it on their old SoC.
To my knowledge this situation is somewhat unique since VC4 is not an ARM SoC but something else where just one or more ARM cores have been added later to the SoC die (the ARM(s) being secondary CPU cores with a somewhat strange integration attempt to interact with the VC4 main CPU). How to ‘migrate’ from something like that to any other modern/normal ARM SoC? Close to impossible without totally ruining backwards compatibility and then RPi customers simply might ask ‘why should I choose your new and totally incompatible RPi when I can choose an ODROID too?’
BTW: pretty amazing experience to join RPi forum: while the RPi people themselves are absolutely honest about their platform limitations and constantly emphasize on the nature of things (VC4 being somewhat ‘special’) none of the fanbois does take notice. It’s really an impressive amount of ignorance present there 🙂
As one of the RPi engineers just put it: ‘$random-hardware-feature would require a new SoC. A new SoC would mean a massive redesign of the HW and years of SW effort’. That pretty much sums up both the success story and the problem they face: Using the same old and boring hardware like the last 6 years obviously results in mature software support situation today. And you don’t want to throw your only asset away by choosing a totally incompatible SoC as hardware basis even if this means that your current offerings are laughable when looking at hardware features. Their customers do not care anyway…
I think the most news worthy piece of information in this press announcement has been entirely overlooked. There is now an official release window for RPI4:
“The Raspberry Pi 3 Model B+ […] will remain in production until at least January 2023.”
This means you can expect RPI4 in 2022.
I’m not sure we can derive availability of Raspberry Pi 4 from long term availability of RPI 3B+
The Raspberry Pi 3 was launched in 2016, but the older Raspberry Pi 2 or Pi model A+ boards are still for sale.
If they are planning for the future, I expect them to (at the very least):
1. Start the development for new board prototype (RPI4) from now. Choose among the best cheap ARMv8 with quite open architecture.
2. Max memory cap at least 4GB.
3. USB 3 (better 3.1), GB ethernet, wifi (better if it has integrated wifi on the SoC <- my opinion)
4. 28nm at the minimum.
5. 64bit OS.
The have 6 year long experience with RPI, so for new board, I think they can pull it < 2 year from now
My crystal ball works very well!
‘“PI next” will launch next year (2018).’
‘ All indications are it will be an incremental update: more Pi3+ than Pi4; a product “refresh”.’
So far they released every year around the same time something ‘new’. For those people who ignore that it’s always the same (VC4) the replacement of one secondary slow ARM core with 4 faster secondary ARM cores was a ‘big update’ while for those familiar with the platform it was just an incremental step.
Maybe next year we’ll see Wi-Fi using two antennas as one possible further ‘innovation’ or a ‘new’ Zero using the BCM28370B combined with the MxL7704 PMIC allowing for four cores still keeping consumption low (should be now easy with the PMIC if they limit DVFS to lower OPP on a Zero). On the other hand the consumption numbers published so far for the RPi 3 B+ do not look promising at all with almost 1W more idle consumption compared to the predecessor: https://www.raspberrypi.org/magpi/raspberry-pi-specs-benchmarks/
You should keep in mind that ‘charity went business’ in the meantime. I read somewhere in their forum one third of their sales go to commercial customers which usually want to get an idea about a product’s long term availability first. No wonder they announced 5 years of 3B+ availability yesterday.
Currently they use the most closed architecture possible which everyone ignores for whatever reasons. Prior to any secondary operating system being able to boot the VC4 SoC they use now needs to load a proprietary RTOS called ThreadX to initialize the hardware (and to control it later). Only after the primary OS finished booting parts of the hardware are allowed to be controlled by the guest operating system (be it Linux, Win IoT, a BSD, whatever). All the proprietary / closed source stuff running on the VC4 is integral part of the ‘Raspberry Pi experience’.
Why exactly should this change? Why should they migrate to a ‘quite open architecture’?
I noticed the same there. You see a lot of “yay”, “I can’t wait to get mine” or “I need one” while for anyone with a little bit of technical background, this one is a regression from the previous ones.
In fact the amount of clueless fanboys is very likely their best asset : I’m pretty sure it’s possible to sell them an incompatible upgrade provided it’s announced as the new best feature. Even better if the finger is pointed at incompatible stuff for being responsible for the incompatibilities. Also it would be the right moment : history has shown that a lot of products experienced small improvements till version 3.x then completely changed in 4.x. Probably that people get bored after 3 times the same stuff.
Integrating some hardware abstraction in some libs and frameworks used in applications, and making these new versions a requirement for future extensions would be a first step in that direction, breaking compatibility with the most legacy stuff.
Now a real difficulty would be to choose another SoC, especially since it’s possible that Broadcom sells them their outdated stuff for almost nothing. SoCs like i.MX8 offer a wide range of options that could be really interesting to increase the product range, but they’re likely among the most expensive ones as well.
RPF going iMX8.. Our species will populate Mars first ; )
Yes, thanks. Clearfog Base bit on the expensive side.
As I mentioned earlier RPI stick with this SoC because only they have it, it stops competition, or other SBC makers boards being fully compatible with the RPI software.
It is same as having a cool Ferrari looking car, with turbo chargers, but look under the bonnet, engine cover, and its just, a 800cc 4 stroke engine with the combustion cylinders bored out. Liable to breakdowns, overheats and does not deliver what the looks promise.
Looking at Linkedin explain many things: https://www.linkedin.com/in/ebenupton/
Might explain why they started with VC4 8 years ago. But not necessarily what’s happening next (e.g. Broadcom releasing an VC5 with four A55 cores or something like that to be used on a Pi 4 or whatever will follow)
Most probably I don’t get the hint? 🙂
I’ve seen people start ‘dirty’ but when they success, they start a new (clean). Hope RPF can follow their step 🙂
I see. Didn’t think that way. True that only RPI using broadcom & VC4 so far. Really hope others like hardkernel, or orangepi could be alternatives beside RPI.
Hmm… why a regression? It’s just another ‘incremental update’. After dealing 36 hours with new board and the interesting RPi vendor community my preliminary (very subjective) conclusions as follows:
1) Ethernet speed improvements: not ten times better as usual when switching from Fast to Gigabit Ethernet but up to 3.5 times when USB storage is not involved at the same time (94 Mbits/sec vs. 330 Mbits/sec) and 2 times faster when USB storage is in use (9.5 MB/s with Fast Ethernet vs. 19 MB/s with GbE — the NAS use case)
2) Wireless speed improvements: from less than 40 Mbits/sec max now in 5 GHz band +100 Mbits/sec possible. No MIMO due to single antenna solution so especially with larger distances performance will drop a lot
3) Improved bootloader behaviour to boot from USB storage or PXE without any local SD card storage needed
4) I2C accessible PMIC added to allow for fine-grained DVFS, better thermal design: Negligible peak performance ‘improvements’ (less than 20% so you won’t notice) but higher sustained performance without heatsink/fan
1) Higher consumption (both idle and under load). One user already figured out that using latest kernel with EEE support reduces consumption by a whopping 450mW (the switch on the other end of the cable has to support EEE too!), switching to Fast Ethernet is then another 100 mW less. So users who don’t need the tiny network improvements save over half a W by downgrading to Fast Ethernet via ethtool 🙂
2) No cheap/passive PoE possible any more due to Gigabit Ethernet. Only expensive IEEE 802.3af will work. The PoE HAT is said to cost an additional $25
3) If I understood correctly the Microchip LAN7515 is internally 2 cascaded hubs. But I might be wrong and won’t ask further since already lost interest
Everything else is the same. So RPi 3 B+ is at least not for me (the use cases where we use Raspberries all only require the proprietary VideoCore IV stuff so still the older boards are the better choice).
At least next year RPi folks can replace their single antenna Wi-Fi with something MIMO capable. 2×2 already will do and double or even triple wireless performance. At the same time they upgrade BT from 4.2 to 5 then. And maybe they come up with an eMMC socket as another ‘innovation’. I’ve tested a bunch of ODROID and Pine64 eMMC modules with RPi already. Slow as hell due to slow SDIO interface but at least worked: kaiser-edv.de/tmp/9gv9du/RPi_Rocket.jpg
Im wondering what sbc i could get to replace my boring raspi3 as a libreelec box….
Do i really have to get x86? That would be sad and a capitulation for arm…
@tkaiser did you check the performance of fast Ethernet, is it maxing out the 100? Not that it would matter 😛
Talking about bandwidth you get on all Fast Ethernet equipped Rasperries what you can expect: between 90 and 95 Mbits/sec at the TCP/IP layer and with an OS image using good settings (e.g. OMV using the Armbian tweaks) ~9.5MB/s NAS throughput. Latency is a bit of an issue but that’s expected with USB2. No idea whether this will change now with the new board and GbE (it’s still USB2 connected to the host — on other platforms with USB3 GbE chips connected via SuperSpeed to the host this usually isn’t that much of an issue any more)
Maybe RPF want to go to VC5 but the driver/firmware are really crap to use with Linux and don’t want to show an image of “regression” or performance lost with his flagship.
I find a presentatios called “Status of Broadcom’s vc4 and vc5 Drivers” from 2017-09-21 by Eric Anholt
The thermal images here are worth a look: https://medium.com/@ghalfacree/benchmarking-the-raspberry-pi-3-b-plus-44122cf3d806
Nice improvement or let’s better say an impressive demonstration how broken the thermal design of both BCM2837 and RPi 3 are compared to how it’s done everywhere around and now finally with BCM2837B0 / RPi 3 B+ too.
To sum the changes up: ‘They have flipped the chip (SoC) upside down, the back of the chip is completely flat and can make excellent thermal contact with the heat spreader that is mounted directly to it, the heat spreader in turn is thermally attached to the ground plane of the PI, so that heat is transferred to all corners of the PI. This much better thermal arrangement makes that the chip isn’t such a hot spot as it was before’ (source)
Power supply paths to the ARM cores have been improved, now there’s an I2C accessible PMIC allowing for fine-grained DVFS and the thermal sensor inside the SoC moved from the VC4 close(r) to the ARM cores (the huge distance between thermal sensor and the A53’s on the BCM2837 was a real problem since if you measure temperatures at the wrong spot it becomes difficult to implement a correct thermal throttling strategy).
BTW: Please don’t waste your time looking at the benchmark numbers in the above link. Partially both numbers and ‘methodology’ are simply bizarre 🙂
In fact they generate those clueless fanboys. As many know censorship is an integral part of the ‘Raspberry Pi experience’ at least in their forum and communication channels. Say something negative there and it gets deleted. Say something that could result in their users realizing that there’s a larger world around and it gets censored away.
Happenend twice today after I explained in their forum what’s important when choosing a heatsink using the picture of a ‘competitor’, an Orange Pi Lite. The 2nd time I chose to ‘archive’ the thread prior to some RPi censor doing his work: https://archive.fo/6kzg0 vs. https://archive.fo/B8dNn now — original link: https://www.raspberrypi.org/forums/viewtopic.php?f=63&t=207863&start=225#p1286503
Asshole behaviour as expected 🙂
But that’s part of the RPi success story. A CEO always telling fairy tales about ‘education’ and ‘charity’, everyone babbling about ‘open source’ while the RPi platform is as closed as possible with a proprietary and totally undocumented RTOS running as primary OS on the primary CPU (the VC4) controlling the hardware, censorship in their forum to keep their users dumb, not fixing serious issues like crappy Micro USB for powering, telling misleading BS about PSU amperage ratings instead of explaining that under-voltage is the problem and so on… Finally done with this crap platform…
I hear what you say on RPI forum, having experienced my self, and watched other being abused by RPI forum, managers and staff.
RPI have a upgrade path, but the logical chips to use, are available to other makers too.
It really is all about exclusive SoC chip, gives RPI hold over customers. While blocking competition from other using the same SoC.
RPF’s other market is the Zero line, so a upgraded SoC for that line is another branch they could make, maybe even use their quad A7 SoC again in a upgraded Zero.
Why would you choose x86? Get yourself a cheap s905/s905x box or board, that’s pretty much the best libreElec choice either way.
When I asked a developer a week or so ago, they said S905 or a Rockchip. The latter was a “it’s not quite there now, but it’s heading in the right direction–given it being in many chromebooks–and the chips are fairly good.”
So, an ODROID-C2 or one of the Rock64 boards might be good candidates.
I’m still using a C1, but my TV is only 1080p.
RPi trying to be competitive to any recent ARM SoC released within the last 5 years is somewhat succesful. They now implement DVFS in a better way than before: https://www.raspberrypi.org/forums/viewtopic.php?f=63&t=208057&sid=1562232267b5eb9fc220cf3c7673b627#p1287591
Fear of deletion is why you post on raspberry forums on Sundays? 😉
The increased energy consumption looks quite wired to me. Might they have switched back to a linear voltage regulation?
The increased consumption is due to Gigabit Ethernet now and the more capable Wi-Fi chip. If you switch off the USB hub (kills all USB ports and Ethernet) then you’ve 1W less. Switching between Fast and Gigabit Ethernet also affects consumption and for whatever strange reason even if no Ethernet cable is connected the following call to bring down the eth0 interface results in 200mW less:
When you also kill the wireless stuff with the following command idle consumption is at around the same level as the old Pi 3:
Though then your Pi has only serial console, audio and GPIO stuff any more.
Now I got finally banned and the moderator lied as always about ‘usually no censorship happening here’ and ‘usually nobody is banned’ while even books mentioned their censorship problem years ago and recommend to stay away from their official forum: https://books.google.com/books?id=DpFsHhgmHFYC&pg=PA277&lpg=PA277&ots=UMSZpw4RWb&sig=jO4ntG5g5mOb38WubYxLNcprnRc&hl=de&sa=X&ved=0ahUKEwjhktmSofbZAhWC2CwKHXspANkQ6AEIRDAD#v=onepage&f=false
Now that first people outside the little censored RPi micro world start to test we get also Wi-Fi numbers:
So single direction traffic good for ~230 Mbits/sec while official announcement talks about 102 Mbits/sec.
The above test was with a capable 802.11ac AP 1m away from the Pi 3 B+ while the numbers the Pi Foundation presented were made in some random way with an 802.11n AP for reasons unknown to me.
So under best conditions in 5GHz band with 802.11ac 6 times higher Wi-Fi throughput compared to the old RPi 3 that is limited to 802.11n at 2.4GHz band. Of course in real world scenarios always environment matters: 5GHz should be less overcrowded but 2.4GHz could be better over longer distances with walls in between, but only 802.11ac (5GHz) allows for beamforming if the AP supports it and so on…
If their next ‘incremental update’ in 2019 then implements at least 2×2 MIMO this might double the throughput over short distances and will give even greater benefits over larger distances compared to today.
After several months delay, the RPi POE HAT is now up for sale: https://www.raspberrypi.org/products/poe-hat/