That is a very interesting development. I know it's not getting anywhere close to x86 (_64) open specs or some other ARM-powered SoCs, but QualComm constantly shooting themselves in the foot by closing it off and not even selling the SoCs hasn't been very productive in the low-power computing community (and industry). I get that the first instinct of NDA-driven businesses is to ignore makers or community-based hardware designs (and sales), but with the Pi selling about 10 million, it should be getting harder to ignore what used to be 'the little guys'. With innovations basically taking place at hacker's homes, not being available basically makes you useless in the sector where your chips could be great.
For Broadcom, the Pi might not have been such a big deal, a single home use router with a Broadcom SoC might easily sell 50 million units, and there are plenty of home use routers for sale with Broadcom SoCs in them (let's say, we have 100 brands/versions, that's at least 5000 million chips sold over ~10 years), but the brand name and support from the community (which is bigger in numbers as well as diversity than any of the big tech companies) is pretty much priceless when you hit that millions number and up.
I hope QualComm and their Snapdragon isn't the last one to nudge closer to availability, there are plenty of others that would be welcome to join. While being able to buy the chips themselves isn't the biggest improvement, the fact that they are actually interested in helping with some open drivers is a major improvement!
When I worked for QuIC [0] I had a chance to ask Rob Chandhok how I could connect maker spaces [1] to Qualcomm so they could buy SoCs and all the glue from Qualcomm on tape with data sheets without an NDA. He put me in touch with someone who eventually hooked me up with a Canadian VAR who sells (or sold) some sort of dev kit q10 or something for a grand a pop. It is a market that Qualcomm just doesn't care about. I was upbraided for using non-Qualcomm silicon for a demo - I asked where the Qualcomm version of the RPi was, and that was that.
At least for me Dragonboards are not suitable since they have no ethernet port (this is probably due to the fact that the 96boards specification for "Consumer Edition (CE)" (http://www.96boards.org/ce-specification from http://www.96boards.org/specifications/) does not require ethernet, but WiFi (which I consider as a bad idea)).
5000M home use routers over 10 years doesn't sound right. According to https://en.wikipedia.org/wiki/Global_Internet_usage , only about 10% of 7.2B people had a fixed broadband internet connection as of 2014. And these connections are typically shared between multiple people, so assuming 1 router per 2 people, that's only 360M routers. Making the conservative assumption that they are replaced every 2 years (in my experience it's actually less frequent), that's still only 1.8B chips. And that's for the entire market, not just Broadcom.
Well, I based my loose figure on the Linksys WRT54G which sold 50 million units in 11 years. So in a best case situation, there would globally be 100 similar models in that timespan also selling about 50 million. On top of that, you usually have a CPE provided by the ISP on which the open market models are added on top of, theoretically making the market twice as big. And then there's at least one chip at the other end of the line (i.e. a line card) and possibly more since there are multiple hops in between a carrier core network and the end-user.
In the past, a router might have lasted about 5 years, but currently, with the speeds almost getting an order of magnitude higher, almost any router that's 2 years old or more is no longer suitable.
So that's why my guess with the SoC sales for broadcom-ish chips is at 5000M :) And then there's the STB's, mobile phones, smart devices, TV's, modern DECT phones, cars etc. Even every 802.11 device (be it PCI, USB or otherwise) has a SoC, as the radio's can't work on the host CPU due to timing.
As the x86_64 world continues to get more locked down, it's going to be interesting to see whether a market niche opens up for a less-constrained platform.
It's confusing to me that you're calling it "less-constrained." The signed-bootloader world really caught fire in SoCs like Snapdragon and indeed the 410E and 600E include SecureMSM/TrustZone. It hasn't (yet) found big popularity in x86_64 designs -- even when the processors offer it many customers don't leverage/enable it.
The "locked down" designs come from two big motivators: DRM and security. Since many snapdragon phones are sold with subsidies, the signed bootloader allows the network operator to restrict the phone's "owner" in various fashions. The security provided by a signed bootloader is real -- it's much harder for a malicious actor to create ransomware that targets mobile phones like they do PCs.
I'd expect/hope that you can decide not to opt-in on the Snapdragons' SecureMSM features, but it doesn't seem any "less-constrained" than Intel or AMD's offerings.
So, to be honest, I don't know all that much about the ARM world, and I hadn't heard of TrustZone. I looked it up, and it seems that the OEM has a great deal of discretion in setting it up, to the extent that there's a reference implementation on GitHub [0]. And if I buy a bag of loose SOCs, which I can apparently do now, I'm the OEM.
Right, kinda like what happens when you buy a Xeon or an Opteron. You can enable AMD SKINIT / Intel TXT & UEFI Secure Boot if you like. Intel didn't force this upon the consumers, the OEMs requested the processor features.
is wrong. I know that vendors producing servers want that Intel ME (because otherwise remote servicing servers would be much harder), but they surely did not require from Intel that Intel ME cannot be disabled.
But IME is distinct from "Intel TXT & UEFI Secure Boot" ("this"'s referent).
> Intel didn't force [Intel TXT & UEFI Secure Boot] upon the consumers
is true. Both of those things can be disabled on most hardware. In fact, I disabled them on the laptop the I just bought, because I wanted to install OSes that aren't SecureBoot signed.
The locked-down hardware architecture definitely enhances security, as far as protecting the embedded firmware and IP. As a consumer, I don't want a clipper chip in my pocket. It's a sad situation for hardware these days, because if you don't comply with the unwritten rules, you'll go the way of Joseph Nacchio and Qwest.
We might as well use an unencumbered ISA as well, like RISC-V. I'm hopeful that that project will take off. It seems to be doing quite well so far. There is still an uncomfortable amount of proprietary junk and questionable hardware modules in the ARM world as well, even if it's a lot easier to get a plain SoC and turn it into a computer.
Yep. I'm actually really excited to get my hands on a Pulpino[0], just because it's based on RISC-V. Everytime this topic comes up, I flash back to "The Coming War on General Purpose Computing"[1] and think "we can't let computing become totally locked down". Hence I'm always interested in finding ways to support more open platforms, and something like an open ISA like RISC-V seems like an important element of that.
There as so many blobs and secret controllers in the x86 platform it's not even funny. Since Core 2 Duo there is basically no way to run a CPU+PCH (or 'chipset') with free software. An average x86 PC has 10 'hidden' controllers doing all sorts of magic, and you can't touch it. The Intel Management Engine and Boot Guard make it extremely hard to use the chips without Intel's approval, NDA and a truckload of money.
The biggest example is Intel and their taking away of ability to do custom chipsets for their chips. That move torpedoed NVIDIA's Ion. If you want custom, then you go ARM.
The first two responses are right on the money -- "trusted" modules, locked-down chipsets, can't write or use open firmware. This is why we get so excited when a random ThinkPad model might have a firmware bug that allows us to bypass the TMM.
It's really unfortunate too, because NXP nuked quite a bit of the NDA free support for the i.MX6 after they acquired Freescale. There's now tons of dead links, and they even removed stuff like a bunch of the SD Card images for their Sabre boards. When we asked where that stuff went we were directed to some crappy Indian outsourcing firm that kept it behind a $19,000 pay wall. I'm really happy that one of our more seasoned engineers essentially locally mirrors everything he can download when he starts a new design just for this circumstance.
Hopefully that will change in 1Q2017 when Xilinx finally releases to production their new Zynq UltraScale+ MPSoC[1]. It has a quad-A53 cluster and a dual-R5 cluster. It is still in early-access so I can't say much more beyond that (we're designing it into a system so I'm constrained by the early-access NDA).
Yeah, working in the SoC industry (closer to the silicon side), I'm regularly frustrated at the gulf between the state of the art and what's usefully available to the public.
I'd love to have a multi-core ARM64 with integrated PCIe, SATA (NVMe!), USB3, IOMMUs, maybe a few Cortex-Rs around for some real-time/MCU-type work with GPIOs and stuff... All of which I've seen and worked on in the industry. But none of that is open to the public :(
So serious question since you worj in the industry. Given the amount of time, energy, and money that would go into that design - if it were completely open source, what would prevent some Chinese firm coming along and making a knockoff for half the price?
Having the register specs and behavior of the various blocks inside the chip is a lot like having the API of a library. It lets you use it, but doesn't tell you how it's implemented internally, which is where most of the secret sauce happens. This register/spec is what a lot of people like me wish for.
There are two levels of "open source" for hardware: the first is where the registers and behavior are documented, allowing programmers to write their own drivers for the hardware, but for moderately complex stuff doesn't get you far for copying it. The second level is where the implementation details of the chip (the Verilog or VHDL description of its behavior) are open-sourced. Only the latter would let you "easily" create a knockoff, and even that isn't within reach of any household hacker, as the base cost of fabbing a chip is still in the hundreds of thousands of dollars, but yeah that would be within reach of an undercutting competitor.
(there's even a third layer of open source where the Verilog/VHDL is turned into the final layout tuned for the fab, think of it like having the source-code vs the final assembly for a given architecture. That step is costly too)
While not fast by PC standards, this chip will still run circles around Snapdragon 600E. I'd think that AM1-based computers are the ideal cheap NAS setup.
I'd spend a bit more money though to get older, better supported hardware (AM3+ / FX-6300). I'm sinking ~$400 into hard drives anyway so what's another $50 for more performance and better hardware support? (AM3+ is supported by FreeBSD and therefore FreeNAS / Nas4Free).
But if you're going for a Linux build which generally has good support for these things, AMD's AM1 is a strong contender.
I'm currently running a Synology 2013j which has 512MB of RAM and a 1.2GHz Marvell Armada 370, and the only thing I'd like to speed up is PAR2, otherwise it does the job of feeding AppleTV just fine. The quad-core 1,5Ghz cores should definitely help here.
AM1 looks like a 25W CPU plus what the chipset consumes, which I don't consider to be very energy efficient anymore.
The price for the whole system (RAM, PSU, case) will probably be quite a bit higher as well.
But it looks like good option if you need a standard platform with comparatively high single thread performance.
Its certainly not energy efficient. Which is why I said it only satisfies "C" :-)
But 25W is still more energy efficient than a full-blown computer.
> The price for the whole system (RAM, PSU, case) will probably be quite a bit higher as well.
$20 for 4GB of RAM + $40 case. You really can build the core components for less than $150.
Don't forget the $10 USB boot drive. USB-boot seems like a superior option over hard-drive boot. That way, your entire hard drives can be used for data-storage. USB-boot can be easily backed up separately and kept independent of your data.
Or spend a bit more on a bigger case + PCI SATA3 port extensions to grow to 6+ Hard Drives.
----------
Of course, the hard drives you put in will raise the costs by several hundred dollars. (Four 5TB Drives will cost $600 for example).
> But it looks like good option if you need a standard platform with comparatively high single thread performance.
I've contemplated a Windows-based NAS before, but the $100 Windows tax is just too much.
But I do want to play with Storage Spaces. It seems like Storage Spaces + ReFS is Microsoft's answer to BTFS / ZFS. And I do mean the two features combined. ReFS only does bitrot checking so it isn't quite a ZFS replacement. But combined with Storage Spaces the Windows solution has distinct advantages.
--------------------------
I've ordered a Synology for work, and it works good too. Its simple and solid, works with everything I expected out of the box. At home, I can afford a little bit of time to baby these machines (ie: build custom). Plus, you can build 4-bay systems for very cheap.
Lots of good configurations out there. I think people forget about standard PCs because they think they're expensive... but for the costs of ~$100 (like a Dragonboard), the standard PC / AM1 platform is indeed competitive. Although it does use more power.
>>The new SDK will only work with the latest Snapdragon 820 processors from the latter half of 2016, and the company isn’t saying if it plans to expand its availability. [1]
IMO the article is clearer and more explicit than the press release. It also summarizes data sheet features which are only referred to in the press release.
> Officially both SoCs support Android and Linux but for some reason the 410E also supports Windows 10. We will ignore that option almost as thoroughly as Microsoft ignores ARM users and focus on the Linux side.
Still, he reports real news events despite his clear bias. So I find that reading his page every now and then is worthwhile (like today's article). He is prone to hyperventilating however.
Unless you want to run Android the Pi remains the best bet for Linux support. I have an Odroid and have dabbled a bit. However its become clear that because of ARM, Google and their partners complete closed nature its always going to be an uphill struggle to get Linux working properly on their platform.
You get old kernels, missing or poorly functioning drivers, and blobs so even experts cannot verify anything. The recent Amlogic scandal with fradulent CPU speeds discovered only recently is case in point.
A whole army of open source enthusiasts and developers have been trying diligently to work with ARM but they only end up being bounced back and forth between ARM and its partners some for years on end, with both feigning lack of authority to make things work.
There are plenty of binary blobs, closed drivers and security blobs that will not go away unless ARM supports Linux with action beyond lip service and announcements. Given some of the issues like GPU drivers are pending for more than 3 years Linux is clearly not a focus for ARM.
Like some others suggested I would look more closely at the 25W AMD Athlon 5350 or Celeron 1037/1900 for Linux support for things like HTPC, NAS and homebrew routers. You can get CPU and motherboard for $50-60.
Big news here is that they are officially supporting Freedreno. Far as I'm aware they're the first embedded vendor with a modern graphics offering to support an existing quality free driver. The motivating factor I've heard for people developing these drivers has always been long term support and flexibility in embedded products, and this is an excellent answer to those questions.
For Broadcom, the Pi might not have been such a big deal, a single home use router with a Broadcom SoC might easily sell 50 million units, and there are plenty of home use routers for sale with Broadcom SoCs in them (let's say, we have 100 brands/versions, that's at least 5000 million chips sold over ~10 years), but the brand name and support from the community (which is bigger in numbers as well as diversity than any of the big tech companies) is pretty much priceless when you hit that millions number and up.
I hope QualComm and their Snapdragon isn't the last one to nudge closer to availability, there are plenty of others that would be welcome to join. While being able to buy the chips themselves isn't the biggest improvement, the fact that they are actually interested in helping with some open drivers is a major improvement!