lanefu changed the topic of #armbian-rockchip to: Armbian - Linux for ARM development boards | Rockchip SoC | www.armbian.com | This channel is relayed to the equivalent Discord channel | this channel is logged
<Armbian-Discord>
<balbes150> Rocpi4x has a very low NVMe performance (presumably due to the limitation of the lines), it works at the USB 3.0 level, so they are not of interest for working with NVMe (using rk3399 in this mode is stupid).
<Armbian-Discord>
<balbes150> P.S. The stupidest cooling solution is a downward-facing radiator, its efficiency is 1\4 of a radiator with the correct position (i.e., you can have exactly the same result from a 1/4-sized radiator with the correct position up). Another huge stupidity of all manufacturers is to paint the radiator black - this is complete idiocy from the point of view of physical laws, a black radiator works 20-40% worse
<Armbian-Discord>
than a light one (the higher the height of the ribs, the worse the efficiency of black radiators for emitting thermal energy).
<Armbian-Discord>
<ManoftheSea> What
<Armbian-Discord>
<ManoftheSea> Is radiation a significant component of cooling these boards? I thought they were conducting to air
<Armbian-Discord>
<ManoftheSea> But why does the radiator position matter?
<Armbian-Discord>
<Werner> Well warm air tends to go up while cold air does the opposite. If a passive heat sink is mounted horizontally up cold air gets in from the side takes, warms with the dissipated heat and goes up. The cycle continues. If the heat sink is upside down it is less efficient which is correct since the warm air cannot dissipated as good.
<Armbian-Discord>
<lanefu> TK frequently made the case that heatsink on bottom generally in SBC context yielded better results.. likely because of surface area... being possibly
<Armbian-Discord>
<lanefu> so flip your board upside down if you dont like the heatsink on the bottom
<Armbian-Discord>
<lanefu> *because surface area mattered more than diffusion properties of hot and cold air
<Armbian-Discord>
<lanefu> then the fact that teh heat sinks are painted black... they're so big.. who cares.. they look cool and they sell
<Armbian-Discord>
<lanefu> mill a custom one for more cost if optimization matters to that extreme
<Armbian-Discord>
<lanefu> optimizing to market and sales is just as important.. or more than performance
<Armbian-Discord>
<lanefu> good companies balance the 3
<Armbian-Discord>
<lanefu> then the tuners tune the rest
<Armbian-Discord>
<lanefu> just like gaming PC builds
<Armbian-Discord>
<Werner> Yeah I guess when selling boards to the causal John Doe a black HS is probably a good idea
<Armbian-Discord>
<lanefu> yeah I still have zero nanopi boards.. sourcing them never seemed obvious i guess
<Armbian-Discord>
<lanefu> like what's stupider than painitng them black is putting them in a plastic case
<Armbian-Discord>
<lanefu> lol
<Armbian-Discord>
<lanefu> i love teh big black heat sinks htey look great without a case and can just sit there on desk, racksheslf whatever
<Armbian-Discord>
<Werner> Indeed lol
<Armbian-Discord>
<Werner> I had two, then one, then almost zero, then one again ^^. The Neo3 died unfortunately a while ago and the R4S had weird behaviour until I learned that it is extremely picky about the sd card you shove in. Refuses to boot on 64G while 32G of the same brand, same category, same speed, everything same besides capacity, boots fine.
<Armbian-Discord>
<ManoftheSea> are they actually painted? Or are they dyed, or electro-colored?
<Armbian-Discord>
<ManoftheSea> by a factor of four? In light of empirical testing that reaches the opposite conclusion?
<Armbian-Discord>
<Werner> I don't know about the factor but in basic thermodynamics. I guess easiest would be to just do some tests with board upside down and normal to proof what is correct. I also guess TK is right about his claims though since with bigger surface area air flow gets less important.
<Armbian-Discord>
<Werner> Should have brought a few boards with me here lol
<Armbian-Discord>
<lanefu> yeah i dunno if htey're annodized or painted... i suspect it varies by manufacturer
<Armbian-Discord>
<lanefu> you didn't bring your nerd travel kit? 😛
<Armbian-Discord>
<c0rnelius> i seriously doubt the color of the HS matters all that much.
<Armbian-Discord>
<lanefu> yeah not in these contexts
<Armbian-Discord>
<lanefu> if my SBC was sitting in the sun all day i'd be open to it
<Armbian-Discord>
<Werner> nope. laptop only 😅
<Armbian-Discord>
<c0rnelius> i would assume the quality and size of it matters more. like the one that comes on the nanopc-t4 is fricking pathetic. although they do make a fancy one now with a fan mounted in it. i bet its also not very good though.
<Armbian-Discord>
<ManoftheSea> "anodized" is the word I wanted
<Armbian-Discord>
<ManoftheSea> But also... I thought black better received and sent radiation. So if the radiator is hotter, black is the right color
<Armbian-Discord>
<ManoftheSea> Why does the rib height matter?
<Armbian-Discord>
<balbes150> q = σ A (T14 – T24) 🙂
<Armbian-Discord>
<ManoftheSea> Substituting in the values we get a heat flow of 6.4W due to radiation at 95 degrees C heatsink temperature and 25 degrees C ambient, so less than 25% is due to radiation under optimal conditions for maximizing radiation loss. That's actually higher than I expected
<Armbian-Discord>
<ManoftheSea> Out of 27W total dissipated
<Armbian-Discord>
<ManoftheSea> Hmm. Dull black is the highest emissivity, but at room temperatures, everything is "black". So a white anodized heat sink will absorb less from incoming hot radiation. For instance, sitting in the sun
<Armbian-Discord>
<balbes150> If you perform correct comparative analyses, a downward-facing radiator, in principle, cannot work better than an upward-facing variant. But you can deceive the user and get the opposite result if you "correctly" fake the test conditions. For example, can you take RockPi4 as a sample? put it radiator down on... a metal table and get a better result than radiator up. I have been engaged in thermal engineering
<Armbian-Discord>
for many years and I know perfectly well how many "deceptive" tests can be created to justify any condition. 🙂 But there are laws of physics that you cannot change and with correct measurements, you will NEVER get confirmation that the radiator down will work better than the radiator up.
<Armbian-Discord>
<ManoftheSea> Here, I'd have thought "sitting on a desk" was a good test. Our sbc's aren't floating in deep space
<Armbian-Discord>
<balbes150> This is another deception of manufacturers, copper is basically no better than aluminum. The key clue is the air heat transfer coefficient. Only it determines the rate of heat transfer to the environment (to the surrounding air) and it does not matter what the radiator is made of - copper or aluminum or steel. For all metals, this coefficient is many times greater than that of air, and only air limits the
<Armbian-Discord>
rate of heat transfer. As an example, you can have a multi-lane road leading to a narrow bridge, the speed (number of cars) that can travel in an hour will be determined by the narrowest point - the bridge. The same is true for cooling, the amount of heat removed from the processor is limited only by the total surface of the radiator and it absolutely does not matter what material it is made of, heat spreads "instantly"
<Armbian-Discord>
through the metal compared to air. 🙂
<Armbian-Discord>
<balbes150> color is like for women, the main thing is to be beautiful and it doesn't matter what shit
<Armbian-Discord>
<balbes150> marketers have figured you out and are using your weaknesses :))))
<Armbian-Discord>
<balbes150> All passive systems have a trick - the larger the piece of metal, the better it accumulates (during peak loads) and then gives it away (in time to reduce loads). the power allocated by all ARM is minuscule (compared to x86), so it is easy to manipulate data, producing the desired results. The example, N2, has a massive radiator and it is possible to do a tricky (perhaps unintentionally, without the desire to
<Armbian-Discord>
deceive) testing, which will show a plus at the lower location of the radiator. But if you do honest tests with exact observance of the rules, it is more gentle, it will always lose to the top one. 🙂
<Armbian-Discord>
<NicoD> The paint is true for sure. I've sanded the paint from the contact point of the heatsink of my RockPiX. From an overheating mess to never overheating again. Just a very thin layer of paint.
<Armbian-Discord>
<NicoD> I do like the heatsink on the bottom. Just for ease of use. If it cools it well enough so it doesn't overheat I'm happy. I love the M4V2 case. Ain't perfect, but it cools the beast(when replacing thermal pads with coper shims)
<Armbian-Discord>
<balbes150> then go to a store that sells home heating equipment and look at the color of the radiator from good manufacturers 🙂
<Armbian-Discord>
<c0rnelius> I'm referring to it in the context of these lil arm boards.
<Armbian-Discord>
<ManoftheSea> The ones where matching the interior design is important?
<Armbian-Discord>
<ManoftheSea> Is radiative heat a Europe thing?
<Armbian-Discord>
<balbes150> the branded radiator with a fan on the T4 is a very good design, it has worked for me for a couple of years at maximum load, a very quiet fan with automatic regulation.
<Armbian-Discord>
<c0rnelius> I've never noticed using a black heatsink instead of say a green or blue or chrome one made any noticeable diff on these little things.
<Armbian-Discord>
<c0rnelius> @balbes150 well if anything it looks neat. 🙂
<Armbian-Discord>
<lanefu> blue heatsinks are racing heat sinks
<Armbian-Discord>
<lanefu> perfect white stripe in center
<Armbian-Discord>
<c0rnelius> and yeah. the T4 needs a fan if ur doing anything of importance with it.
<Armbian-Discord>
<balbes150> black is the stupidest color for a radiator - planes (ribs\ pins, etc.) directed at each other actively absorb what the neighboring ones should emit. white (silver) - reflects, black - absorbs. Try walking in the bright sun in black clothes and look at the Bedouins - they have white clothes so as not to absorb radiation 🙂
<Armbian-Discord>
<NicoD> Before and after, no idea how they got it in their head to paint this.
<Armbian-Discord>
<NicoD> The wrong paint also creates more thermal resistance. No paint is better.
<Armbian-Discord>
<ManoftheSea> So then we just model the heat sink as a cube/sphere for radiative purposes.
<Armbian-Discord>
<ManoftheSea> If there's a hot source shining on the radiator, it's more important to reject the incoming radiation than to emit outgoing radiation. Therefore, white and shiny. If the radiator is the hottest source and is glowing, I think we want blackbody radiation - black and dull.
<Armbian-Discord>
<ManoftheSea> Which, I already said about being "in the sun". The Sun is a lot hotter and a lot more radiation than the heat sink, so it hurts the effectiveness of the heat sink to have high emissivity
<Armbian-Discord>
<ManoftheSea> If there's a fan on the heat sink, I naïvely expect that to be more meaningful than thermal convection, as well, so as to make whether the heat-sink is top- or bottom-mounted an æsthetic concern, not a thermal one.
<Armbian-Discord>
<ManoftheSea> having a good thermal paste seems a lot more important than the heat sink position.
<Armbian-Discord>
<NicoD> "In conclusion, coatings can reduce heat transfer caused by convection, conduction or radiation. The use of IR reflective pigments in an exterior paint system can prevent radiant heat transfer, providing a cooler exterior surface and less conduction heat transfer in the summer."
<Armbian-Discord>
<profbbrown> Hello. Having a little issue with my OPi 4 LTS. Fresh install of the latest Jammy CLI. The load average never goes below 1.0. CPU is 99.7% idle.
<Armbian-Discord>
<NicoD> See with htop what process is doing this.
<Armbian-Discord>
<profbbrown> The top process listed is sd-pam, using 0% of CPU
<Armbian-Discord>
<NicoD> Oh, I red the question wrong. I though you had 99.7% use.
<Armbian-Discord>
<NicoD> Is it for sbc-bench?
<Armbian-Discord>
<profbbrown> I don't know what that is.
<Armbian-Discord>
<NicoD> It's a script to benchmark sbc's that needs under 1.0 load to start.
<Armbian-Discord>
<profbbrown> No. This is my first time using the Orange Pi 4. It's just sitting idle, doing nothing. Since the CPU isn't busy, perhaps something is blocking I/O.
<Armbian-Discord>
<c0rnelius> my T4 always does this and its not on Armbian.
<Armbian-Discord>
<profbbrown> I have an Orange Pi Zero+ also and it's load hovers near 0.
<Armbian-Discord>
<c0rnelius> I think someone else was mentioning this not long ago. maybe it was you?
<Armbian-Discord>
<profbbrown> Not me, sorry
<Armbian-Discord>
<lanefu> Sdcard or emmc? How busy is journalctl -f
<Armbian-Discord>
<lanefu> SBC bench won't run of load is below 0.10 😉
<Armbian-Discord>
<NicoD> I knew there was a 1 in it 🙂
<Armbian-Discord>
<profbbrown> Doesn't make a difference sd vs emmc.
<Armbian-Discord>
<profbbrown> Just made a new bootable micro sd.
<Armbian-Discord>
<profbbrown> Totally fresh install. No packages updated or anything.
<Armbian-Discord>
<profbbrown> Seeing this in journalctl -f: Sep 28 09:14:46 orangepi4-lts kernel: fusb302 4-0022: connection has disconnected Sep 28 09:14:46 orangepi4-lts kernel: fusb302 4-0022: connection has disconnected Sep 28 09:14:46 orangepi4-lts kernel: fusb302 4-0022: connection has disconnected Sep 28 09:14:47 orangepi4-lts kernel: fusb302 4-0022: connection has disconnected Sep 28 09:14:47 orangepi4-lts kernel: fusb302 4-0022:
<Armbian-Discord>
connection has disconnected Sep 28 09:14:47 orangepi4-lts kernel: fusb302 4-0022: connection has disconnected
<Armbian-Discord>
<profbbrown> Didnt' see that before on the previous card.
<Armbian-Discord>
<profbbrown> With the previous card, I had run apt upgrade and I noticed it installed some firmware and such for the board.
<Armbian-Discord>
<Tenkawa> do you have any usb drives plugged in?
<Armbian-Discord>
<profbbrown> no
<Armbian-Discord>
<profbbrown> Only a power supply on the usb c port
<Armbian-Discord>
<profbbrown> Time to start work. Back in a couple hours.
<Armbian-Discord>
<IgorPec> any objections in this reckless move ? 🙂
<Armbian-Discord>
<profbbrown> After letting the system run for a few hours, load avg is at 1.7 with two processes taking up time: kworker/0:3+fusb302_wq and systemd-journal.
tkaiser has joined #armbian-rockchip
<tkaiser>
@profbbrown you're using kernel 4.4 right?
<Armbian-Discord>
<profbbrown> 5.15.52
<Armbian-Discord>
<profbbrown> The only thing plugged into USB is the power adapter. I'm going to try switching to a barrel plug and see what happens.
<tkaiser>
@lanefu: Any references for 'TK frequently made the case that heatsink on bottom generally in SBC context yielded better results'?
<Armbian-Discord>
<profbbrown> Switching to the barrel jack has caused the USB errors to go away. So it was something about the use of the USB-C jack to power the board.
<Armbian-Discord>
<profbbrown> However, the load average is still 1
<tkaiser>
Which 'USB errors'? As for the reason: FUSB302 configured to use wrong pin maybe?
<Armbian-Discord>
<Tenkawa> I know on some other SoCs "if" you had been using an actual usb drive with certain io schedulers could cause that in older kernels however I thought that was fixed ages ago and you have already confirmed you aren't running any usb attached drives
<tkaiser>
@NicoD what you're telling people about sbc-bench isn't accurate. With v.0.9.5 check for average load has been extended to CPU utilization since in this stupid SBC world too many broken settings cause loadavg being constantly above 1
<tkaiser>
@Tenkawa what is an otg?
<Armbian-Discord>
<Tenkawa> on the go function call in usb
<Armbian-Discord>
<Tenkawa> one of the USB modes
<tkaiser>
Sure. But in the context of FUSB302? How should this matter? It's the I2C connection that matters...
<Armbian-Discord>
<Tenkawa> not in this context actually because i2c has to still route through dts and the pd usb controller
<Armbian-Discord>
<NicoD> What am I telling people about sbc-bench?
<Armbian-Discord>
<Tenkawa> its all interconnected to the same physical connector
<tkaiser>
@NicoD: 'It's a script to benchmark sbc's that needs under 1.0 load to start.' – and I told you few minutes ago what has changed. No idea whether you will be able to pick this up.
<tkaiser>
As for your refreshing NVMe discussion here guys... you seem to ignore the difference between random I/O and sequential I/O. :)
<Armbian-Discord>
<Tenkawa> yes.. it was the interupt on line 9 causing it just as I thought
<Armbian-Discord>
<Tenkawa> which is the the PD line
<tkaiser>
For most SBC use cases random I/O is far more important than sequential. And guess what: even on an USB2 port any SSD showing shitty sequential I/O performance shines if it's about random I/O. Every SSD hanging off an USB2 port always outperforms the fastest HDD accessed by whatever protocol (even being SAS 12G or whatever that legacy protocol is called).
<tkaiser>
So choose your benchmarks wisely before drawing conclusions. And come up with sane recommendations for users like "choose an SSD that makes use of all 4 lanes when combining with RK3399"
<Armbian-Discord>
<Tenkawa> Good thing RadxaNaoki found that .. that could have gotten ugly down the road
<tkaiser>
Since Gen3 x2 SSDs ofc get bottlenecked on a Gen2 x4 SoC since Gen2 x2 will be negotiated
<Armbian-Discord>
<Tenkawa> Indeed.. already seeing it with a few x4 drives I'm testing here
<tkaiser>
Adding to that: on many of the recently (within the last 4 years) added SBC/SoCs all I/O relevant optimisations are missing. As such looking at I/O benchmarks is an excercise in ignorance
<Armbian-Discord>
<lanefu> @tk well don't want to take words from your mouth, but i recall sermons from you... but i won't be digging for citations
<tkaiser>
@lanefu: well I guess you're confusing me with willy? He reported here and there that he prefers to let the heat radiate into a desk or something.
<Armbian-Discord>
<lanefu> well i think my recollection is more about placing SoC on bottom of board being more sensible, but yeah if i'm way off i appologize
<tkaiser>
You create a ticket talking about this and add a 'fix' addressing that.
<Armbian-Discord>
<profbbrown> Looks like I need to wait for a USB PD patch to make its way into the kernel
<Armbian-Discord>
<lanefu> i'll be honest man.. i want to care.. but i don thave the brain power.. i was JUST trying to capture your concerns somewhere hoping someone else would act on it.
lanefu has joined #armbian-rockchip
<tkaiser>
The code snippet there works universally (No 'need a maintainer for this set of code'). But that's the 'I/O performance with Armbian on boards added past 2018 is crap' part
<Armbian-Discord>
<lanefu> I look forward to using that information when i get my big suite of boards... I'm probably not gonna be the guy that gets it over the line for Armbian... who knows
<Armbian-Discord>
<lanefu> im just a power user....
<tkaiser>
But there's nobody left and it seems all that stuff was a pet project of Mikhail and mine (Mikhail still today being the most important contributor to Armbian ever!)
<lanefu>
you speak the truth
<Armbian-Discord>
<profbbrown> @tkaiser If it something I can change in a config file and reboot, that's easy. Recompile a kernel module isn't high on my things-to-do-on-a-Wednesday-afternoon list. 🙂
<tkaiser>
@profbbrown that was just a hint above (about a copy&paste DT issue that is likely to happen and that you're not going to get any effective assistance in some Discord crap channel)
<tkaiser>
The symptom of having wrong pin in DT settings for FUSB302 is known, it might apply here, nobody knows since Armbian tries to maintain too many boards to provide any sufficient level of 'support quality'. It's simply impossible.
<tkaiser>
On the bright side: loadavg above 1 and CPU utilization below 1% -> time for some ignorance since it really doesn't matter.
<Armbian-Discord>
<NicoD> It's nice to see you back Thomas 😉 It's like you've never left.
<Armbian-Discord>
<c0rnelius> I only read CNX for his comments. true story.
<tkaiser>
@profbbrown you (or whoever you trust in – be it even those people here) should compare schematics of your board with device-tree. Back then when I was interested in Xunlong boards and involved in board bring-up the HW docs were pretty accurate. Today? No idea.
<tkaiser>
Just to get an idea about Armbian today (since everything happens on Discord, in the armbian-devel channel that is logged nowhere and the 'super secret' channel)
<tkaiser>
What's your motivation?
<tkaiser>
What's the outcome?
<Armbian-Discord>
<ManoftheSea> "everything"? Nothing happens on discord except Igor tells people to go to the forum
<Armbian-Discord>
<c0rnelius> what do you think happens on these super secret channels? sexy nerd grrls, hookers, flap jacks... you know, super secret stuff.
<Armbian-Discord>
<c0rnelius> sadly its none of those things. Its nerds debating heatsinks and black paint.
<Armbian-Discord>
<c0rnelius> and does the SoC distribute heat better if its facing up or down. as if this really take a genius to figure out.
<Armbian-Discord>
<profbbrown> Are the images on armbian and orangepi.org different?
<Armbian-Discord>
<c0rnelius> I don't know about you all, but I really like how super-secret channel sounds. I vote we change armbian-devel to that.
<tkaiser>
@profbbrown _If_ it's about a wrong integer value in some device-tree definition then a totally different 'OS image' won't change anything.
<Armbian-Discord>
<lanefu> @tk you have a false assumption that a lot of meaningful coordinated work is getting done. It's not. People bullshit and talk in chat and things happen as they happen in git. End of story. Nothing new
<tkaiser>
> you have a false assumption that a lot of meaningful coordinated work is getting done.
<tkaiser>
Nope
<tkaiser>
(quotation doesn't work here)
<Armbian-Discord>
<c0rnelius> In all seriousness, I think I know what @tkaiser is getting at here. Its not at all a very open platform.
<Armbian-Discord>
<IgorPec> channels are bridged, what's the problem?
<tkaiser>
Autism?
<tkaiser>
The one who called me recently an asshole somewhere here on Discord claimed that all channels are brigded but... they're not. Good luck with your project claiming 'open source' and whatever
<Armbian-Discord>
<c0rnelius> Does it really bother you that much that they are using this? Its just the flavor of the month after all. IRC in recent years did kind of take a hit.
<Armbian-Discord>
<IgorPec> from my perspective they are - if not, then this is just an infrastructure bug. nothing extraordinary
<tkaiser>
<gorPec
<tkaiser>
Autism?
<Armbian-Discord>
<IgorPec> its common, yes. but currently i am just sleepy as it was a long day
<Armbian-Discord>
<Tenkawa> @IgorPec btw I'm finally back.. we can sync up in next few days to see what I missed while I was away
<Armbian-Discord>
<Tenkawa> long trip finally completed.
<Armbian-Discord>
<c0rnelius> Austism? Isn't that just an American word for reasons.
<Armbian-Discord>
<IgorPec> hey @Tenkawa yeah, if not before in community call
<Armbian-Discord>
<Tenkawa> nod
<Armbian-Discord>
<c0rnelius> Hey Igor what rkbins are you using on that uboot?
<Armbian-Discord>
<IgorPec> i'll ... still device tree is fu* up somewhere
<Armbian-Discord>
<c0rnelius> I attempted upping the elf at one point but ran into trouble.
<Armbian-Discord>
<c0rnelius> is not matching whats going on in dtsi files?
<Armbian-Discord>
<c0rnelius> a lot of changes happened recently
<Armbian-Discord>
<c0rnelius> naming
<Armbian-Discord>
<IgorPec> yes, that makes troubles. one afternoon was not enough
<Armbian-Discord>
<c0rnelius> I feel bad for Max. I kept sitting there telling him to 'boot it again!' as I don't have the HW. I think he hates me now.
<Armbian-Discord>
<IgorPec> haha
<Armbian-Discord>
<IgorPec> without hardware is so wasteful also for the one putting things together
<Armbian-Discord>
<c0rnelius> i agree
<Armbian-Discord>
<c0rnelius> its fine.
<Armbian-Discord>
<c0rnelius> whatevert
lanefu has left #armbian-rockchip [WeeChat 3.2-dev]