On Fri, 24 Jun 2016 06:32:33 +0300 li...@wrant.com wrote: > Fri, 24 Jun 2016 12:10:11 +1000 <bytevolc...@safe-mail.net> > > On Fri, 24 Jun 2016 04:30:39 +0300 > > li...@wrant.com wrote: > > > > > > What is more important is the level of engineering information > > > available from the manufacturer (PC Engines) web site including > > > tech specs, manual, BIOS updates, accessories, enclosures, diag > > > boards and also: Schematics! > > > > I certainly dig schematics! Don't forget, they use coreboot. > > Yes, these are block diagrams of important sub-systems with chip > pin-outs, signal names, voltages, logic arrangement. It's not the > complete electric schematic diagram as in a service manual, and > certainly not system design documentation, but is is engineering > level sufficient and: public access!
The schematic here (http://pcengines.ch/schema/apu2c.pdf) actually has many of the the components too (mostly the support components around the chips, and power regulators and filters), but they are "joined" with the corresponding signal pins. But as you say, engineering level sufficient and public access; definitely a win. I mentioned coreboot mainly because it is great to have hardware that comes with it pre-installed, so it is essentially a guarantee that it works. > > Comparing this to the paper manual I got with my expensive 2011 Atom > D525 system board from SuperMicro, I found what I got lacking for my > engineers purpose, despite having connector pin-outs and some > voltages. Also, some time later I found out much cheaper Atom > mini-ITX boards form competitors as a whole, I would have went with a > PC Engines if I knew about them then. I originally held off on PC Engines because they only had 100Mbps connections at the time, and the specs were somewhat mediocre. The APU1 came in, but it only had a dual-core CPU and I have heard horror stories about it heating up a lot. Then the APU2 came in, and it was difficult to refuse; 1GHz quad-core, 4GB RAM, USB 3.0, supports OpenBSD, and 100% open source and public-access schematic. And best of all, it means I can finally have a modern system that doesn't have UEFI, which seems avoidable in laptops these days unfortunately. > > So, are you saying that coreboot is serial compatible BIOS? As in > textual interface exposed on the serial port, and no menu like the > other historic Award/Phoenix/Ami PC BIOS-es? Does it give access to > all the BIOS options over the serial port as in pre-boot system set > up via RS-232? None video and keyboard dependencies any more for > complete system management? Is it? This one is ASCII only, and doesn't appear to use any funny control codes to juggle text around the screen and move the cursor all over the place; if it does use control characters, then they are very basic ASCII ones for presentation. Though I am not sure if this is vanilla Coreboot; it would have been modified to suit the hardware. There is a "menu", but it's one of those "type the number of the item you wish to enter" menus. The only options I could see are the boot device order and a few settings here and there. The general attitude of this BIOS is "load the OS and get out of the way", which is how it should be. > > > > > > > This seems to be by far more friendly to both engineer & consumer > > > users. > > > > > > PC Engines APU2 product line > > > [http://www.pcengines.ch/apu2b2.htm] > > > > > > 1) How do the APU systems go as pricing to comparable systems from > > > other similar (industrial class, desktop enclosure) > > > manufacturers? > > > > I have two APU2C4 boards. > > I've never seen these in action, nor had chance to use any coreboot > device. I don't think this is has vanilla coreboot software on it either, given that it only has a serial port and no video output. This is the only coreboot device I have used. > > > The price is not bad, and the ALIX/APU boards are not loaded with > > consumer-grade "ooh, aah" bullet-point rubbish, unlike some of the > > VIA boards which are (quite worryingly) also marketed towards > > medical devices. > > The SuperMicro BIOS experience over serial port (the -F models have a > BMC/IMPI controller onboard) is not that great. It is the traditional > Award style BIOS transposed in a screen interface 1:1 without a proper > serial connection functionality factored in. It is not a serial BIOS, > it is serial exposed historic old school BIOS. Not necessarily bad, > it has borders, colours, much like servers from other commercial > vendors. This BIOS is 100% text-based, and it is a "what goes out stays out" approach as far as its UI is concerned. In my opinion, every single BIOS should be like this. Basically you need to make sure the serial console is connected before starting the system up, otherwise you'll end up missing the BIOS setup menu and the system will either be in the 10 second delay (that allows you to make a selection before it boots into the OS), or it will end up in the OS itself. Other than the above minor trap, I would say this simple UI is essentially "printf()-style" terminal-agnostic. > > It works, much better than compared to the KVM options you get with a > dedicated server hosting plan, but is not susceptible to scripting, > nor automation or anything computer (program) operated. It has a > bug, that the sensors stop providing data over time and if you do not > restart the BMC when you reboot you get a system unusable long beep > unable to boot surprise, quite disappointing. A complete remote > usage no-go. There is NO FIX, and the work around is to poll the BMC > over IPMI and reboot the BMC if it gets to crash which is frequent. > Expensive and trouble. It may not be happening on other systems, but > for me it is just enough. When you start the APU up for the first time, there will be some blips from its internal piezo. It then starts a memory check, which I highly recommend letting it complete (with the serial plugged in and terminal running throughout the test). The test took about 1 hour on mine, if my memory serves me right. > > I know of lots of other SuperMicro embedded and server mainboard > quirks. Does a PC Engines APU have any quirks that we should know of > in advance? Mostly physical traps. One thing I found is that these boards are very sensitive to brown-outs, even short ones. For isntance, when powering multiple boards connected in parallel, I found power-cycling one of the boards will actually cause the other one to reset, due to the initial in-rush current. Pascal Dornier has told me they can handle up to 14V, though I haven't tried this yet. Also, when assembling, the heat spreader they give you also is not easy to put on either, especially if you are not an octopus. When I got mine, the heat spreader packet came with two thermal pads; The original APU1 requires both of them, so maybe that explains it, but I kept the other one as a spare in this case. > > > I am just a little disappointed in the way the components used are > > just the consumer/low-end commercial grade versions (0-70degC > > operating temps, etc). However, it certainly saves cost! > > Interesting, when you source sufficient enough amounts of components > per order the price difference is negligible compared to the PCB > fabrication. I definitely would not need the industrial / medical > price tag for my gw. The components are really just half the story. I sometimes get worried about some of those medical devices; they still run a dodgy unpatched version of Windows, so it really doesn't matter at that point whether the components can withstand -55degC to +105degC and are military grade etc. Some clever PCB designs actually build capacitors, resistors and inductors on the PCBs themselves. > > > In addition, the clips for the mSATA/mPCIe slots, given that the > > use of metallic screw points would improve grounding to the devices > > and would be a lot more robust and resilient against vibration; > > with screw posts, there is the option of using rubber washers too. > > And, screw posts would cost an order of magnitude less, considering > > the cost of assembly too. > > OK, it is not fit for vehicle mounting, yet those do not come with > Ethernet ISP cabling :-) I suppose the device is quite fine for a > tabletop enclosure. Nothing a bit of thick double-sided tape can't fix though, I guess. I recall the old ALIX Fortunately I have yet to encounter a miniPCIe or mSATA device with a PCB so thin that the clips cannot hold them. It would not surprise me if such devices do exist though. > > > > 2) How is the OpenBSD experience on the APU systems, do they have > > > serial RS232 console (serial BIOS), do they expose all the > > > hardware to OpenBSD? > > > > The serial (RS232) console is set up to use 115200 baud by default > > (I am unsure if this is changeable), so make sure this is > > in /etc/boot.conf: > > > > stty com0 115200 > > set tty com0 > > > > I would in fact recommend writing "installXX.fs" directly to a USB, > > and then mounting /dev/sd#a to edit boot.conf to add those lines in > > right at the top, before installing. It makes life easier. > > > > I have not had the opportunity to test the GPIO support though; the > > watchdog timer is not supported by OpenBSD, so whatever you do, do > > not enable the watchdog timer yet. > > > > > Thank you for providing valuable technical feedback on these what > > > appears to be the smarter choice over the other low power > > > devices. > > > > It is definitely the smarter choice; on average, I notice around > > 5-6W consumption, although that does vary depending on load. > > I will have much less difficult choice the next time I need to deploy > the (home) office gateway router and network micro server system. > Thank you. You definitely won't be disappointed if you are using it in that way! In fact, I have found it to perform better than some of the "well known" brands (Netgear/TPLink/D-Link) of routers. My suggestion would also be to obtain the board with 4GB RAM instead of just 2GB; a difference of US$13 is a bargain for an extra 2GB in RAM. Allows one to allocate a reasonable-size non-volatile /tmp partition. And even if you don't use the extra RAM, it's there when you need it; the RAM cannot be upgraded (at least, not easily). Throw in the 16GB SSD that PC Engines sell (US$17), and you have a winner; the specs on that SSD are not bad too, and they use a PHISON controller. For such a gateway 16GB is plenty.
