pfry

The Orange Pi RK3399 has 5V and 3.3V available on the 40-pin header (and nowhere else that I can see offhand). Pinout is in the manual; it's a standard .1"/2.54mm header - lots of connector options. The mSATA interface is a bit inconvenient (uncommon); it also has 3.3V power available - perfect for an mSATA device; not much use if you're adapting it to a 2.5" device. I didn't check to see how much power is available, but I'd expect it to be fine for typical devices.

"Were", at least, likely rectified by now. There were a few posts earlier this year.

I haven't tried the most recent kernels, but video was a bit spotty with the mainline as of a few months ago. Looks like Rockchip's status page: http://opensource.rock-chips.com/wiki_Status_Matrix ...hasn't been updated recently. Anyway, the 4.4.x has worked well for me, but I... have basic needs (e.g. NVMe not disappearing). Rather like IBM (up to the '80s, at least), nobody gets fired for choosing Samsung SSDs. Pricey (compared to the NanoPC-T4), but reliable.

Nice work. You're even crazier than I am. And your soldering skills are obviously superior to mine.

Get a NanoPC-T4, which has handy pushbuttons? The NanoPC-T4 schematic sheet 17 has its "boot" button (ground EMMC_D0, essentially); sheet 26 has the "recovery" button (ground ball AH 26/ADC_IN1). EMMC_D0 is pin 1 of the NanoPi M4's eMMC header - a bit tough to reach, given its surface mounting on both the M4 and the eMMC module. The best way would be via a (presumably custom) module inserted between the M4 and the eMMC module, but that's a bit of work, unless you plan to do it a lot(!). Using the solder pad(s) should be possible if you have a finely pointed contact and a steady hand. Good luck, as a bad slip could cost $100. As for recovery mode, finding a particular ball or trace would make dealing with the surface-mounted eMMC header look super easy. Kinda odd - am I missing them, or does the M4 have no buttons at all? Edit: Hm. Not sure how suppressing the eMMC on boot helps you. But hey.

Found one: https://www.infinitecables.com/power-cables/internal-pc-power-cables/12-inch-lp4-female-to-lp4-female-internal-power-cable/ The linked site is unfamiliar, so caveat emptor. Search term (e.g.) was "lp4 cable f-f"; you may find other parts/sources. I'd verify the pinout before application. If you're careful (or you have the factory tool) you can dismount the contacts from the housing(s) and re-order them. Naturally, you can build your own as well.

I was just referring to the RK3399 boards (given the extra regulators for the A72s and GPU, it's apparent that the RK808/818 was designed for a less power-hungry SOC), but yeah, it's pretty obvious to me now (!) that the implementations vary. A lot, considering the relatively minor variances, mostly in the peripherals (audio codec, W-Fi, eMMC, etc.). The Realtek Ethernet phy is pretty much a standard, at least. What a tangle. Nothing that can't be solved with enough time and money. It's too bad both are tough to come by.

Scroll down to sheet 27 of the schematic. The 1.8V power is for the PCI-e phy (power/enable/both? - I didn't look that closely), not the slot itself. If the RK808 wiring is board-specific it could explain a lot of the difficulties with individual boards (I haven't compared 'em).

The board power design is interesting. I'd expected a more (even more?) uniform reference implementation across RK3399 designs. The RockPro64 uses a 5V 3A regulator to power... well, the board, essentially. The NanoPC-T4 uses a 3.3V 8A regulator for the same purpose, so it has no additional regulator for its M.2 slot, while the RockPro64 has a 3.3V (in this case) 2A (3A peak) regulator for its 3.3V bus (which includes the PCI-e slot). So Friendly makes 25W available on the NanoPC-T4 where Pine64 has 15W for the RockPro64 (the power domains appear to be equivalent); in addition, 2-3A seems low for a 3.3V domain that includes a PCI-e slot (alone rated at 3A). I'm not an EE and I haven't done any testing, so my opinion is worth less than the time it takes to read this. Still, it's interesting. The Samsung 970EVO consumes what? 3.3V 1.8A peak? I doubt the PCI-e -> M.2 card has any on-board regulation (to utilize the 12V power); other devices on the 3.3V bus include the USB, Wi-fi, Ethernet and SPDIF phys, and the eMMC - none likely to be a heavy draw. Aside: the NanoPC-T4 also has an 8A regulator for 5V (USB and audio). I'm paranoid, so I glued heat sinks to the SY837 and 838, the NB680 (3.3V regulator) and the RK808 on my NanoPC-T4. They don't even get warm. The RK3399, on the other hand...

I prefer the Supermicro for build quality: https://www.supermicro.com/products/accessories/mobilerack/CSE-M14TQC.cfm It may be possible to locate an OEM version for less money. I'd also lose the 40mm 15000RPM fan (may or may not be an exaggeration) and use something different to push air through it. One of my favorites is the Sanyo Denki (same brand that comes with the Supermicro), e.g.: https://products.sanyodenki.com/en/sanace/dc/counter-rotating-fan/9CRA0912P0G001/ The counter-rotating fans give you good static pressure for use with filters. Note that you'll want to use a PWM controller with it* - it's hilariously frantic at full speed. Motorized impellers have generally better performance than axial fans, but are more involved to mount and duct. * I usually just use a potentiometer to ground. PWM presents a variable voltage, usually 0-5V; any fan that can run with an open PWM line has a built-in voltage source, which can be shunted to ground through the pot - no additional components needed. If you wish to control it via software, all you need is a PWM output. Probably more than you want to get into, but what the hey.

Whoa, I'm not. I'm a total leech on y'all's work, and I have no desire to cross-compile an image. (I prefer PC-style native installations like LFS and Gentoo - from an end-user perspective Armbian is the least-customized distro that I use.) I've had no issues with my NanoPC-T4, and the super-easy eMMC+NVMe install is unique (afaik). Can't beat it with a stick. I can't contribute much, as all I have in excess is Internet bandwidth and old PC hardware (and somewhat off-the-rails forum posts).

? I'm guessing you didn't look closely at the splitters. They're just SATA power "Y"s: each has one SATA power "device" connector and two SATA power "cable" connectors (I'm too lazy to look up the accepted nomenclature). Plug them into the power cable that came with the hat and you get "Molex" (really AMP Commercial Mate-n-lok) -> 4 SATA power "cable" (there's no wrong way to plug them together, so long as you only plug compatible connectors together and do not loop the cables; but the best way is to plug each splitter into the original cable, rather than plug one splitter into the other). If we're still not connecting (so to speak), grab a friend and run this thread by 'em.

I'm the last person who should accuse someone else of overthinking an issue (as though it's a bad thing), but you may be missing an opportunity to get your boards running the way you want it. If you grab two of the the splitter cables linked by [frauhottelmann] (they're far from the only option, but StarTech is a safe bet, and those are "shipped from and sold by Amazon"), you should be good to go (for one board/drive set), since you said you have a 12V power supply. You can then grind on the details to your heart's content, with a bit of added information/experience. You might decide to stay with what you (will) have. Pre-built systems are convenient. You'll certainly not save time (in the short term) with a custom solution, and if you save money, you're a better man (person?) than I. The benefits lie elsewhere. Not the least: you can help others here. (Speaking of overthinking, I won't run a NAS without ECC RAM and a RAID of some sort. That gets expensive, in both time and money. But cosmic rays, man! Cosmic rays!)

I got it wrong: I assumed the 12V input on the hat was a simple passthrough, but it's not - it has a 5V 8A regulator as well (and I make an example of myself again with a stupid answer to a simple question, because I didn't do my research). Interesting, as your posts in the "Nanopi-M4 SATA HAT" thread seem to show that you can run 5V-only devices without an external supply (so I'm only half-stupid). [frauhottelmann] has the best recommendation for you (two splitters plus 12V supply). Two power supplies and a cable chain is a bit untidy, so if you're bored in the future you could whip out a dual-output supply and custom cabling. Or not. Looks like you're using Crucial MX500s. Crucial doesn't have an easily-locatable power spec, but the photo appears to read "5V 1.7A". I'd definitely recommend using the 12V power input on the hat with four of those. If you run saturation benchmarks on them you could heat up the regulator on the hat, but I doubt it. (One way to find out, and it'd be tough to tell how it's doing without a wee thermocouple or an IR thermometer.) It has two USB ports, too, with the tiny, irritating JST PH2.0 connectors. Nice that they put the pinout on the (back of the) board.

Bah! I should have said that the power output connector on the hat is a standard, old-style PC 3.5" power connector, a Commercial Mate-n-lok (I forgot about the silly spelling there, too), but most likely you'd need SATA connectors for the devices - they're also readily available (and more easily locatable via search, as it's a bit tougher to find the precise Mate-n-lok). I really expected to hate the SATA power connector, but the crimp-style has lots of room to solder and add heat shrink tube. (The IDC-style is designed for 18g wire, a bit bigger than you'd need, especially if you're making a 1->4 splitter.) The M4 seems to be another device where you can clobber the power supply if you try hard enough. In terms of 5V current, the RK3399 SOC and board needs 2-3A, 2 USB 3 can consume 2A, 2 USB 2 1A, the SATA hat itself is negigible, but 4 SATA devices 2-4A = 7-10A. Not a likely scenario, but you could do it with, say, a USB HDD, a USB Blu-ray, a couple random USB 2 devices, four power-hungry 5V SATA devices (e.g. the Samsung 830 Pro was rated at 1.6A) and the slightly overclocked RK3399. Squeezing 8A out of that scenario would be easy, at least for a short time. Not that you need to worry about it so much, but be aware that power could be an issue as you plug more stuff into the board.

Heh. There are stupid questions, but at times you have to ask 'em, if only to serve as an example to others. I've been that example plenty of times. Find a split or breakout cable, or make your own. I prefer to make my own: you can get parts from suppliers like Digi-Key, Mouser, Newark, etc. They're AMP Commercial Mate-n-lock. Oddity: The power output connector on that hat is reverse gender. The supply is normally female (sockets, not pins), so finding a pre-made 1->4 splitter cable might be tough. You could get a couple SATA (I assume your devices use SATA power) splitters and make a cable chain. Also: most consumer-type 2.5" devices are 5V only, so the 12V in is not needed (I don't bother to wire 3.3V or 12V on SATA unless I need it), and you'll want to check the specs on your NanoPi M4 to make sure you don't clobber its 5V supply (most devices are <1A; HDDs will draw more than their rated power on spinup, maybe 2-4 seconds - not enough time to cook connectors, and regulators will normally compensate for voltage drop, but hey). Given that the NanoPi M4 uses 5V in, you could bypass the board and power the storage devices directly. It's just a matter of wiring.

I don't know if this'll be useful to you, but what the hey... When you say "the NVMe SSD we currently have", does that indicate a sample size of one? I can't comment on the error, though, beyond the obvious "It doesn't appear to work." (My board is currently out of service.) Your Toshiba device appears to be M.2 2230 form factor with no on-board DRAM: more compact (and potentially difficult to mount on the T4) and generally more expensive and slower than other options. If you don't have an investment in it/them, I'd look for: - M.2 2280 form factor, M key; - PCI-e 2.1 or later (basically every M.2 NVME device); - x4 width, as it's available (x2 is acceptable); - On-board DRAM buffer. Lots of folks use Samsung SSDs. I personally chose a Phison E8-based device (Liteon MU X), a v3 x2 device like the Toshiba but with on-board DRAM, because it was cheap ($28 on eBay for 128GB, new/sealed). As always with cars and computers, "How fast do you want to spend?" The Phison E8 can be outperformed by a good SATA SSD (or better NVME SSD, obviously) in saturation benchmarks, but that's an odd target application for a $110 ARM dev board. Are you constrained by time (product research), money (choosing a known-good but potentially more costly item), or both (potentially having to purchase and test multiple devices to find an appropriate solution)?

Not to nitpick, but to nitpick... e.g. the NanoPC-T4 uses a 12V input, with an NB679 regulator for 5V (USB and audio) and an NB680 for 3.3V (basically everything else). You're right, though -- most of the regulators are rated down near or below 5V for input. The RK3399 reference design uses the RK808 plus a bunch of additional regulators (for the big array, GPU, etc.), all on that 3.3V supply. The NanoPC-T4 can consume 30W fairly comfortably (2 USB 3 + 2 USB 2 = 15W alone, plus SOC, plus M.2, etc.); the RockPro64 has a PCI-e slot, which should provide 12V power over and above the 3.3V (the NanoPC-T4 only has an M.2 slot -- no 12V). You're not likely to plug your 10 boards full of USB HDDs/DVDs and 25W PCI-e cards, but I still think 100W is low. Heh -- I have fans that draw more power (not that I run them at that level -- they frighten me, in addition to deafening me). I'm curious what kind of draw you'll see under a good compute load. Soldering insulated crimp connectors without making a burned, blobby mess is kinda nifty. I always cut the insulators off and go heat-shrink happy. And while there are edge cases where solder can contribute to failures, I prefer to risk those to solve others. (At least when dealing with itty-bitty computer stuff. I can swing like an ape all day on a #2 welding cable with a crimped lug.)

Wow, that worked at all? Impressive. That the fixed buck converters would run at 5V and 240% current. Not to mention your poor 90W PS -- I'd expect it to squeal like a pig. Hm... looking at the PicoCluster site, they list "Power: Internal (100W)", which sounds a bit low -- with default clocks, I've seen measurements (in board reviews) of 1.1A or so under load (>13W), and lacking any other data, I'd call that (say, 140W for 10) a starting point. If you want 2A available at each of 10 boards, you're looking at 250W. Decent modern power supplies should be stable at low loads, and you should be able to get >90% efficiency at 25-100% load (assuming you want to pay for it -- your MeanWell LRS is good for about 86%).

It should be tough to ignore, given that it's monkeying with the duty cycle of the fan's power input. Do you have a way to measure it? Hm... I've never measured a PWM signal with a multimeter - I've always either just plugged into a PWM-capable board or wired up a voltage divider (mostly just a pot -> ground, as the fan has a voltage source if it runs with the PWM disconnected). Beyond playing with the little resistor pigtails that come with various PC fans, I've never tried to control fan speed using the supply voltage - too unreliable, and I prefer to work with a nice, low-current speed control (if possible). The NanoPC-T4 uses a nice big MOSFET on that fan supply line, so current draw shouldn't be an issue. Looks like a higher duty cycle on GPIO4_C6/PWM1 should slow the fan, if I'm reading it right. (Am I?)

RK3399: Junction: 125C; maximum recommended operating temp is 80C for 2.0/1.6GHz (for the K version, but hey). The board has a thermal protection circuit based on the RK3399's over-temp signal (assuming the TSADC interrupt is not set too high or disabled), so it should just die and return to life after reset. That isn't to say that the thing isn't a burning weenie roaster, but surprisingly (to me, at least) the only really hot running component is the SOC itself. The RK808 and various regulators don't seem to get too hot*. Looking at the little enclosure... looks like a nice fit, but a bit tight. In open air (no movement/flow) I get +50C under load with a large (for the board) heat pipe and fin stack, so I'd expect throttling or shutdown with the small heat sink inside the wee box. I'd also expect a reduced component life, but not a hard failure. Probably just bad luck. Never can tell, though - I've seen a few failure modes I didn't expect. *Given my setup, where the giant (er, relatively speaking) heat sink on the SOC reduces feedback. I can get maximum sustained performance, but at the expense of additional time, money, and space.

Interesting. I figured the Rockchip devices would be - by and large - reference designs, given that most features are integrated into the SOC. Excepting things like DRAM type and speed, audio CODEC, Ethernet phy, etc. Certainly the power components seem standard (one board swapped the SY837/838, meaning the big array and GPU would get the other's voltage setting, although I imagine they're similar enough). You should be able to flash images using Rockchip's AndroidTool under Windows, so long as you can read Chinese (it has English language files, but I didn't see a way to choose the language), although I've read of issues with Win10. It should also be possible to choose the boot device: (http://opensource.rock-chips.com/wiki_Boot_option) ...but only after U-Boot is loaded from eMMC (or SD, if no eMMC present/enabled). And Armbian seems to do its own thing, partition-wise, and I haven't quite figured out what/how. Good luck.

I'd say it's easy enough to modify a compatible fan... You can purchase the housings and contacts (in the U.S., from Digi-Key, Mouser, Newark, Arrow, etc.), clip your fan leads, graft on the new contacts and insert them into the appropriate slots on the housing. In practical terms, I find the small JST connectors to be extremely difficult to work with: I crimp and solder all contacts (smaller than, oh, 10AWG or so - I'm very paranoid about connection quality), and I'm just not skilled enough to avoid flowing solder into the contact area. I had the same issue with PicoPSUs. But if you're good, or know someone who is, the contacts and housings are dirt cheap. Me, I just wired a Molex KK (cheap knockoff compatible) receptacle (for standard PC fans) into the 12V power supply for the board. (I also don't trust wall wart-type supplies, so I use a Delta open frame.) Note that my fan will run whenever the 12V supply is energized, and I have neither speed control nor tach out. To actually address your question, I wasn't able to locate either a fan or a pre-built adapter cable with the JST GH connector (you can find a few advertised, but the ones I found did not actually have a compatible connector - examine closely before buying). Good luck.

It's around 56mm, and very tight. I stuck a 4-position housing with a couple blank sockets (to hold it in place) on the ADC connector to avoid any possibility of contact with the heat sink bracket (the fifth pin is a ground, and I happened to have the 4p housing). Threaded inserts are 2.5mm for the heat sink (M.2 is 3mm, i.e. standard). I trimmed a couple #4 nylon washers for the back of the board (to clear components). Photo (2MB JPG): http://www.tailbone.net/p1000325.jpg I'd cover the GPIO header, too, but I ditched all of my 40p housings years ago (unless I still have some 40p -> 80 wire ATA housings somewhere...). So the Thermalright HR-05 (non-IFX, old mounting hardware, long discontinued) works with minor modifications. Something like the Enzotech CNB-S1 would likely work if you trimmed the mounting ears (and didn't mind spending $30); I don't believe there are any high/tall components within about 20mm of the CPU center, but that's the limit. The Enzotech R1 is too large (in diameter), for instance; the Zalman Flower would be as well. There's not much of a selection any more, since PC chipsets are simpler devices these days. (I have an old K10 with an nVidia MCP55 Pro chipset with an HR-05 IFX on it that runs a bit hotter than the RK3399.) As always, YMMV.

Finally fired up my NanoPC-T4. Reading this thread, I figured it had thermal limits in either the power components or the RK3399, so I stuck heat sinks on some of the power components (the Silergy 837/838 and the main 3.3V regulator, and the RK808) and an old Thermalright HR-05 (single heat pipe with a fin stack about the size of the board) on the RK3399. The power components seem to run OK, but it's fairly easy to push the RK3399 +45..50C ("openssl speed -multi 6") in an 18C room (no air flow) (the CPU thermal sensor seems fairly accurate). So the RK3399's over-temp signal (NanoPC-T4 schematic, sheet 13) is likely the source of the shutdown issue when overclocking. A couple things: - 6W TDP? At 2.0/1.5, I'd guess higher - maybe 10W. Higher than an N3150 Atom (not comparing anything other than TDP). - Is there a way to clock the thing up without compiling a new kernel? - Probably the wrong thread, but has anyone set up to boot from an NVME SSD? I firgure I'll try it, as soon as I wrangle a USB cable (to cheesily update the U-Boot "parameter.txt" on the eMMC, something I could probably do from the console or via dd, but hey). - Gluing tiny pieces of aluminum to tiny chips is a pain. And the giant fin stack somewhat defeats the purpose of the tiny board, assuming one cares more about board size than the burning weenie-roasting SOC. Me, I'd prefer Mini-ITX. - Photo of the silly thing (warning: 2MB JPG): http://www.tailbone.net/p1000322.jpg