RockPi4 shutdown itself while benchmarking

#1

Hi,

I have a strage behaviour of RockPi4 shutting down itself, while demanding benchmarks running. I’m using official big heatsink for RokcPi4 with PWM-controlled FAN. Below is the picture of my configuration :

rockpi4
rockpi4.JPEG1040×585 47.4 KB

While benchmarking, the fan is manually set to 100%, and the temp on rockpi is maintained at 70-75 deg (thermal_zone0) and below 60 deg (thermal_zone1). Below is the temp charts.

rockpi4_temps
rockpi4_temps.png652×818 62.9 KB

The missing probes are at points, where RockPi4 shut down.

Could anyone advice, how to prevent it from shutting down? Or what can be the cause of it?

#2

What OS are you using? When it cuts out is it completely dead until you reconnect the PSU?

#3

Hi NekoMay,

Thanks for Your interest in my thread.

I’m using latest armbian buster (Armbian 20.08.1 Buster with Linux 5.8.6-rockchip64):

Linux rockpi4 5.8.6-rockchip64 #20.08.1 SMP PREEMPT Thu Sep 3 18:03:42 CEST 2020 aarch64 GNU/Linux

Yes, the board is dead untill PSU is reconnected. The used PSU is 5.1V, 3A KW18-051E300.

I’m using stress-ng app. The call I’m using is :

sudo stress-ng --cpu 16 --io 8 --vm 8 --vm-bytes 256M --timeout 21600s

and this test is shutting down the rockpi4. I’ve also tried to check 7zr b, and this test works without shutdown, but it apply less cpu load.

I’m trying to debug this issue further, so I’ve changed stress-ng parameters, to test only CPU load. After the 6 hours CPU-only test PASSED without the shutting down. The CPU load and temperature were high (all cpu’s @MAX, temperature below 75 deg).

6hrs_cpu_only
6hrs_cpu_only.png578×720 29.5 KB

Used app call was:

nohup sudo stress-ng --cpu 16 --timeout 21600s

So next, I will try to add or test only memory with -vm 8 -vm-bytes 245MB

If anyone have any ideas what to check, please let me know.

EDIT : I’ve just checked and in benchmarks thread other users report ~18-19W of power consumption. So my 5.1V 3A PSU could be a little to weak. I think I need to find a 9V PD charger instead.

#4

We don’t recommend 5V PSU, please use 9V or above instead.

#5

Ok, I’ve just ordered the PD 3.0 charger witch 27W rating. I will test it ASAP. So, with higher voltage and power rating I shouldn’t have anymore the poweroffs? Even if I disable the fan (and left it only with a big passive heatsing)? The CPU should throttle to drop the temp down?

#6

yes, the CPU will reduce the freq if the temperature is high.

1 Like
#7

Ok, thanks. I will update the post when I test my board with new PSU.

#8

Looking forward, since I have the same issue with random shutdowns on high loads. If that’s the solution it’ll be great to have my Rock Pi 4 nice and stable! (Though I don’t see why it shouldn’t work with a powerful 5V PSU…)

#9

In my case the power rating of PSU is just to low. 15W is below the possible consumption of the board, so I really need to change the PSU to one with higher rating.

Although, I also don’t see the reason why 5V PSU with higher rating (that met the board power requirements) is not recomended. As far as I know, the RockPi4 still can be powerd by GPIO header power lines, that are 5V rating MAX, so this kind of powering scheme is supported?

#10

Yeah, that’s what I mean. The PSU I’m using is 5V @ 4A, which is 20W; maybe it’s right on the edge of what the board draws, but my poweroffs are much more random (however, they don’t occur if I keep Armbian from overclocking and don’t run X11… my load is Rosetta@Home on only the A72 cores).

#11

Because the wide voltage input for ROCK Pi 4 is 5.5V to 20V, 5.1V is at the risking edge. Powering from the GPIO is different, it must be 5V exactly.

1 Like
#12

I see. I just check, and the getting started page has this info included.

https://wiki.radxa.com/Rockpi4/getting_started#get_start_specs

Was it added recently? I don’t remember it, and I’m sure that it wasn’t there when I checked it during my search for appropriate PSU for RockPi4. Although, I think that 5V/3A was included in supported PSU configurations before, and that’s why I decided to buy this one.

#13

I also confirmed with a test - one attached to 5V / cca. 3A and second one to PD PSU, first powered off and second one survived stress testing: http://ix.io/2z9Y

EiyeQuYWoAAWSC1

3 Likes
#14

I have a new PSU, Baseus Type-C PD 30W. According to seller description it should support PD 3.0. I’ve also bought good quality (60W) PD compatible USB Type-C cable. Is there any way, from API to check if RockPi4 negociated higher voltage mode, and which one exactly?

EDIT : Or maybe You could provide me a photo of the board with selected testpoints? This way I could measure USB voltage with a multimeter.

EDIT2 : Ok, I’ve measured the voltage on 0 OHM jumper near the USB Type-C connector, and it shows the 20V. With old PSU, it shows the 5V, so, so far It looks OK.

#15

Good to know! Let it run the stress test for 24-48 hours and see if it keeps stable!

#16

Ok. Now work on RockPi4 looks rock-solid :smiley: The board was working stable at high stress for more than 24h. Below is the chart with core temp for this time period.

24hrs_all_9V
24hrs_all_9V.png764×724 71.2 KB

I’ve changed the fan speed manually a few times to check how high it can get, and if throttle occurs (high peak @85deg). I’ve also set fans to 25% (temp @~80-81deg), and to 100%, to see how low it can get at this stress scenario (50 deg).

For me, with fan @35-45% PWM is sweet spot of cooling performance vs noise - that the core isn’t throttled, and the temp is still at aprox 75deg, and always below 80 deg (heavy load), while the noise is almost unnoticable. At idle, temps drops at this setting to ~35-40 deg.

Thank You all for interest in my thread, and help in resolving this issue.

#17

Unfortunately, my issue is not resolved; my unit still shuts down even under moderate load using a PD PSU (delivering 20V @ 350mA), but only if X11 is running. So I don’t know what is up with mine.

#18

Perhaps, your USB-C cable has too much resistance to deliver enough power.
Buy a thick 18AWG USB-C cable. Make sure the cable is not too long because a long cable has more resistance than a shorter cable.

Or, buy an E-marked USB-C cable capable of delivering 100W through USB-PD.

#19

No, that’s not the issue.

#20

What about using QuickCharge charger?