Jiff

@Dantes The original case. I'm planning another use as a busy cluster for some T6, in which case I'll ditch the case and use new ice towers from 52PI to keep them really cool.

Nope, the curve starts @ 72% and I already draw more than 10% idle with one T6, plus the 512 GB Transcend NVMe who can draw up to 9 W, and anyway a second T6 will also be hooked on the same power supply plus two fans. Read this excellent article not especially tied to the RPi 4, written by an engineer building microprocessors, to understand why the lower the temperature it is, the best long life your computer (in fact all computers) will have : https://qengineering.eu/overclocking-the-raspberry-pi-4.html So I will cool two T6 as said, with two 12 V fan delivering at least 50 m³/h each. You should take the time to read the smartctl output columns headers and some articles about NVMe, this power state is the maximum you will allow your NVMe to consume, needless to say the performance is inversely proportional to this power limitation level.

@Dantes Oh yeah, thanks ! I could not found this information and as I do not have a serial-USB converter it would have taken at least a week to get one. Ok, I'll try with a newer. Thanks for your answers, I'm now entering going tinkering mode

@Dantes Thanks for the command, the T6 can indeed TFTP boot. However the wiki page you kindly also mentioned says to give the TFTP IP address to U-Boot, which I read about, but there is 1 quirks and 2 questions : * The U-Boot menu never appears on the display, it shows the FE logo, then a black screen and you're logged in, so it'll be hard to change any parm (appart displaying it's menu, I also want a longer timeout, but from what I read it is easy to solve once you stopped it before the OS boot). * so how could I stop the boot to stay within the U-Boot console when I can't even see it ? - O_o some of my readings sound like a warning : does it mean it only appears on the serial console ? (If so, I'll have to dig to find how I can have access to it:/). * FE use its own repo of U-Boot v2017.09 when the latest stable git is v2023.07.02, do you think there's a contraindication to use the newer one or is there a real good reason why they stay with such an old version ? - This is not window$ syndrome (always the latest version), but usually with 6 years difference, new versions work better with less bugs and more functionalities. And oops, by network boot, I meant BOOTP, meaning not having to inform U-Boot about the TFTP server IP, but with the command you gave me I see it is also supported

@Dantes You're right about FE images updates, this is why I'm gonna try to learn how to create my own Armbian. FYI the eMMC of the T6 is 256 GB w/ 241 GB free for the user's OS and data - I've seen there's an Armbian patch to be able to boot from the NVMe that is stable, which is an excellent good thing. What I hope is Armbian will make something using all possibilities of the T6 because it seems much easier to pack an image by yourself than with FE tools (and scarcity of docs). BTW do you know if the regular U-Boot configuration used by Armbian images allow booting from the network ?

@Dantes This time, I "tested" the speed of an Erlang dynamic server, I quoted because I used apache2 ab that is not really representative of a real load and the server has no TLS proxy at this time. The page is 23,374 byte, but what is interesting is it is a dynamic (generated for each request) page. ab -l -n 10000 -c 500 http://server.home:8000/register (which means : do 10,000 requests with a concurrency of 500) Server Software: Cowboy Server Hostname: server.home Server Port: 8000 Document Path: /register Document Length: Variable Concurrency Level: 500 Time taken for tests: 9.704 seconds Complete requests: 10000 Failed requests: 0 Total transferred: 237274119 bytes HTML transferred: 233344119 bytes Requests per second: 1030.46 [#/sec] (mean) Time per request: 485.219 [ms] (mean) Time per request: 0.970 [ms] (mean, across all concurrent requests) Transfer rate: 23877.17 [Kbytes/sec] received Connection Times (ms) min mean[+/-sd] median max Connect: 0 1 2.4 0 16 Processing: 24 474 117.4 456 862 Waiting: 15 473 117.4 455 862 Total: 39 475 116.6 456 863 Percentage of the requests served within a certain time (ms) 50% 456 66% 508 75% 547 80% 569 90% 630 95% 691 98% 757 99% 788 100% 863 (longest request) This is not bad at all, the 8 cores of the T6 goes to 100%, but event with that, the GUI is still responsive correctly. There is a slight difference with a RPi4 (same site) overclocked at 1,900 MHz (same ab test, same software, same page) : Server Software: Cowboy Server Hostname: rpi0.home Server Port: 8000 Document Path: /register Document Length: Variable Concurrency Level: 500 Time taken for tests: 61.201 seconds Complete requests: 10000 Failed requests: 0 Total transferred: 236585785 bytes HTML transferred: 232655785 bytes Requests per second: 163.40 [#/sec] (mean) Time per request: 3060.039 [ms] (mean) Time per request: 6.120 [ms] (mean, across all concurrent requests) Transfer rate: 3775.13 [Kbytes/sec] received Connection Times (ms) min mean[+/-sd] median max Connect: 0 1 2.4 1 21 Processing: 25 2959 393.1 3025 3353 Waiting: 25 2958 393.1 3024 3352 Total: 46 2960 390.9 3025 3354 Percentage of the requests served within a certain time (ms) 50% 3025 66% 3075 75% 3105 80% 3124 90% 3169 95% 3206 98% 3243 99% 3267 100% 3354 (longest request) When the RPi4 was launched, the site where people could see it in photos and read about the specs was made of RPi4×13, it could have been covered by only T6×2 (let's say 3… or not, because the T6 is fully accelerated for a bunch of cryptographic operations). I'm quite happy to have purchased this SBC, with a bit of explanations to kids and parents, it can very easily used as an every day computer for a super-price and a very low power consumption (also in enterprises), it can be an excellent multi-purpose server, with it's NPU it seems you can drive an honest AI and with hardware encoding of H.264 & H.265, I bet you can stream like a pro. The OS images are still a bit too young, but I'm drooling thinking about what Armbian will build as a mature one. Hehe, I just realized that the T6 is supporting the suspend to RAM (at the cost of ~2.4 W), thing that is unsupported by the Pi4 and will never be as I read an article 2 years ago saying it is impossible by construction. O_o I just found the source of the little buzz in the treble gamut that I couldn't localize, the power supply, which is normal as I bought a 50 W to be able to hook 2 or 3 T6 on the same supply, 2.4 W is fully in it's bad zone. I finally found a fan, some times ago I cobbled a small cardboard box with an undervolted (5 V through a USB plug) 12 V 80×80 mm fan for my neighbor's dock station with 2 HDD, he lend it to me for an hour, so I installed it upon the T6. From 40.7°C idle, it got down to 33.3°C after 10', which is impressive because with 5 V it is quite slow, and the NVMe got down from 53°C to 44°C. So it _must_ be ventilated - I will do that w/ two 12 V fans, one on top pushing the air down and one on bottom pushing the air up (to keep the NVMe at a temperature as low as possible). Touching it, it was really cold instead of a little hot without a fan.

@Dantes About the problem I made myself (no more boot), I first tried to use their update_tool, but I was afraid to make a bigger mistake as its PDF is not translated (gogol did it) and awfully missing detailed explanations, so I used a SD to eMMC image which worked like a charm. Here, there's what I consider a conception error, because if you insert such a SD into the T6 and reboot, your OS and data are dead (crushed by a new automatic installation. Of course, backups are there for such situations, but they should have used a security, even if it meant adding another button - the ferpection is not of this world… Heating test (using : sysbench --threads=8 --time=1200 cpu run - twice) : Measure of temperatures from one Little core (/sys/class/thermal/thermal_zone0/temp) and one Big core (/sys/class/thermal/thermal_zone4/temp) - This was made with a Debian bullseye desktop image. Start : T6 up for ~35' => Little : 38.8°C - Big : 38.8°C - All temperatures in Celsius. Time Little Big 20' 54.5 52.7 30' 56.4 53.6 As it was still rising, slowly but rising, I cheated around 33' : I put a metal box of 20 cigarillos open in V under the case to create some convection, that stopped the progression, so placing the T6 on a gated shelf would not be a bad idea. 40' 56.4 53.6 Now, the cooling slope : 30" 49.0 48.0 1' 49.0 48.0 1'30" 49.0 47.1 2' 48.0 47.1 2'30" 48.0 47.1 3' 47.1 47.1 3'30" 47.1 46.2 … 5' 46.2 46.2 15' 41.6 40.7 20' 40.7 40.7 - back to idle temperature These numbers look right seeing the volume and mass of aluminium alloy of the case, the level of temperature and the fact that the case (unfortunately) do not have large fins. This is a very good device, as even during this test the GUI was snappy and reactive. There is something a bit weird though, the Big core max clocking is set to 2,256,000 when it is sold for 2,400,000 ; may be it was too hot at this speed - anyway, this is a 6% loss, so it is not that significant (and if it was done to contain the heat, then it's good - chances are they did not have had a metallurgical engineer at hand to calculate the right case). This could also explain why the Little cores are hotter than the Big. I did not test with a fan, as I finally don't have a 12 V 80×80 mm at home, only server fans @ 7,500 RPM and 77 dB, sooooo, not an option. Oh, I discovered what I consider a severe flaw (well, let's say a youth sin), if you disable the auto-login of the default user (pi) into lightdm, you're doomed and obliged to re-install as it stops booting. In fact, I was asking myself why the greeting screen of lightdm wasn't visible, now I know :/ I hope this will be fixed in a not so distant future as it forbids to use the T6 as a workstation in production. Parallel to FriendlyElec, there's also some good work accomplished by our guests, their image is still not accelerated but I don't doubt it will be around this year's Q4.

@ct100 You already have all the explanations you need to understand, although to understand them you have to get out of the tunnel of creed you build for yourself : you are confusing efficiency and power factor, the first one establishes the ratio between the SPS input current and its output (the difference from 100% being the SPS losses), the second one, the discrepancy between current and tension that supply the SPS that is the direct manifestation of its reactance. BTW, GaN chargers, and by extension all chargers of this type, are only SPS with an additional logic in charge to negotiate the output tension and amperage with the device following the PD and/or QC protocols…

@ct100 Hi, you missed the part where I said that on the wattmeter there is : * one RPi4 overclocked @ 1,900 MHz + one PS/2 keyboard hook throught a PS/2-SB + a USB 2.0 HUB w/ an Ethernet 10/100 Mbps, NB : RPi4 alone is ~2.5 W. * one disk dock station using a 240 GB SSD + a 2 TB Seagate 7,500 RPM rust, for the first measure and : * one NanoPC-T6 + a USB Logitech B100 optical mouse + a 105 key USB keyboard + a USB 3.0 HUB. for the second. The measures looks good to me as they fit with manufacturers announced consumption, that said, you seem to make a confusion between the switching power supply (SPS) reactive power and its efficiency : * efficiency expressed in %age is how much the power supply eats by itself which add to its _input_ current. eg: if your device eats 10 W and the SPS efficiency ratio is 86%, that mean the whole _input_ current for the SPS is : 10 W÷0.86=11.63 W * The power factor (PF) is the other of the PHI cosinus, which express the discrepancy between current and tension - if they match on the oscilloscope, then : PF = 1.0, which is a ferpect match meaning the SPS will only draw 11.63 W from the wall socket, but the more discrepancy you have (delay or head start between A & V), the lower the PF is (which BTW is normal for all SPS because of the harmonics it generates which make part of the current bouncing on its self-inductance (high-frequency transformer)), so if your PF = 0.5, that mean the SPS is now drawing : 11.63 W÷ 0.5=23.26 W from the wall socket. However, as a private, your wall energy meter doesn't take his reflected power in account, only the active one, so you'll be invoiced only for 11.63 W and not 23.26 W BUT your meter breaker limit is not calculated in Watts but in VA (Volts × Ampères) - giving the same example (and cheating on the breaker VA power for the sake of the explanation), if your meter breaker cuts @ 20 VA, you're doomed, it'll break because 23.26 W > 20.0 W. For companies, it doesn't work like that as it is the VA power that is taken in account, hence "some" efforts (using big self-inductances and/or capacitors) to correct a maximum of the discrepancy to get a PF as close as possible to 1.0, which will avoid to pay for lost VA by too much reactance. To make an image, if you have ever used a CB, the reactive power is the SWR (Standing Wave Ratio) that express if your antenna length is correctly matched to the wave length, if it is, then the SWR tends toward 1.0, if it is too short or too long, the SWR is rising, which overloads the power output transistor and can even grill it if the SWR's really too high. This is also why changing home bulbs for LEDs (of course powered by SPS) may not be such a good idea if you use substantially more LED thinking you can do it because "the wattage is so low", it can be a benefit only at equal quantities of Lumens.

Hi @Dantes Not that I saw in my tests, USB is working well, mpv is accelerated but not vlc, the temperature is of course rising when pushing all cores @ 100% but I have to conduct 3 (same) long (45' or 1h) tests, one in the conditions of the first, case laying on the table with nothing else, then case lifted a bit and case lifted a bit with a 12 V 80×80 mm computer fan to see how it behaves. I'll see the rest tomorrow or the day after as I stupidly killed my OS running a script supposed to update the FriendlyElec eFlasher from within the original image, which is Jammy (berk) gnome (berk, berk), this nasty thing has re-written /etc/fstab with just one line pointing to the SD card , so now I have to use a USB cable (A-C & M-M) that I do not have to install a fresh image from my computer (a Debian w/ XFCE). The SD installation is out of question as the tarball supposed to contain the eFlasher is empty (or may be I did not understood that well the wiki, as the narrative is terriblly missing precision) - anyway, at this time, I'm condemned to use a FE image, but the first impression is excellent, it is snappy and opposite to things I read here and there, the HDMI connection to a full hd Iiyama 24" worked immediately when plugged. For the power button, yep, you must keep a pen at hand. About that, I remember I saw a "button" a long time ago made for this kind of problem, a self-adhesive silicon button on the outside with a plastic rod inside to push a small butto, but I do not remember where. Note that as I do not use WiFi for security matter, I did not purchase a WiFi/BT M2 card, but I may be buying a LTE M2 card later to see if I can use it as an access point to/from mobiles to an internal VoIP/telephone network using FreeSwitch. I also have to see (FE forums) if it is possible to fully boot from my NVMe or if, for the time being, /boot must stay on a the eMMC or SD card (can't remember the numbers, but it was faster than the Corsair 2 TB tested by another user here, excellent for databases). I also have to read the Armbian doc, as I don't know yet how to generate my own image at this time. BTW, I saw that 52PI made a caloduc radiator dedicated to the Orange Pi 5, so unless they create new brackets for the T6, you have to DIY, but it's worth it. This isn't of course a solution when using the T6 as a computer for people and children, but when used as a server it rocks, well at least the RPi4 model rocks : today 18~19°C => CPU 32.6°C and less than 42.3°C (w/ Arctic MX-4) if I push the 4 overclocked (1,900MHz) cores @ 100% with some dust between the blades. See : https://52pi.com/products/ice-tower-cpu-rgb-led-light-cooling-fan-v2-0-for-orangepi-5?_pos=1&_psq=cooler+orange&_ss=e&_v=1.0 I hope they're gonna also make a vertical version as the RPi4 horizontal version is less effective than the vertical one by something like 1.4°C, this might look like a few, but temperature is critical against electro-migration into CPU & GPU, read this (written by an engineer that builds microprocessors) : https://qengineering.eu/overclocking-the-raspberry-pi-4.html One more thing, the FE site always talks about a 12 V/2 A power supply, thus 24 W, but it can stand other tensions (up to 19 V IIRC) and if you plan to plug a lot of devices at once, be aware that the consumption can rise up to ~37.5 W, so a 12 V/4.2 A (50 W) power supply doesn't seem a bad idea.

Hi Dantes, better late than never… * An rpI4 8 GB RAM powered by a MW SNT-RS-50-5 + disk dock station (that has its own power supply) with a PNY CS900 240 GB SSD and a 2 TB Seagate Barracuda 7200.14 (AF) ST2000DM001-1CH164, my wattmeter shows ~16.4 W (there's some jitter). * Adding the NanoPC-T6 16 GB RAM & 256 GB eMMC with a 512 GB Transcend NVMe, a Logitech M100 optical mouse, an USB 105 keys keyboard, original Jammy image booted logged into gnome, powered by a MW LRS-50-12 , it shows ~22.2 W (there's some jitter). * With sysbench --threads=8 cpu run running, it jumps @ ~25.8-26.2 W (there's some jitter - 26.2 W with the CPU monitor on) - let's say 26 W unless you run some more daemons. So we can say it has a gross consumption of : ~5.8 W idle and ~9.6 W with all cores @ 100% (according to datasheets, both power supply have an energy efficiency of 86%). As the wattmeter is showing an average power factor of 0.42 (it stays the same with or without the NanoPC-T6), the electrical meter consumption for it only is therefore : ~2.44 Wh idle and ~4.03 Wh with all cores @ 100%, which, with a load of daemons and serving (FTP, HTTP, etc) a bit @ home should stabilize it around 2.75 Wh @ the wall meter. Talking energy invoice, using this number (2.75 Wh), that's 66 W.day, ~2.03 kW.month (30.5 days per month) and ~24.16 kW.year (365.25 days per year) if up 24/7, which is not too bad, considering it is around 3× more powerful than a RPI4. Also, compared to an RPi4, the memory transfers are 32% faster (9,053 MiB/s & 11,929 MiB/s). Conclusion : for a bit more consumption than a RPi4 and, at this time, almost the same base price, you have a very capable computer easily able to serve your own blog from home with a respectable number of simultaneous hits.