H3 idle power / BIOS version

[fvolk]

BIOS 1.11:
1.8W at idle, no errors in kernel log so far...

[xnd]

...
Power consumption is a bit lower (~1.1—1.2W in average with unplugged keyboard and HDMI):

(or https://i.imgur.com/zziIaGf.mp4)

My new measurements with BIOS 1.11 on my H3:

- idle power consumption is the same as before.

With default BIOS settings: 2.6 — 2.8 W in average (after powertop --auto-tune)
With enabled ASPM, SA GV: 1.1—1.2W in average (after powertop --auto-tune) + cracking noise (as described in my previous post with 1.08 bios)

dmesg without errors ✔
unplugging HDMI causes no system failure ✔
my powertop html report: https://output.jsbin.com/turuzab/1
my whole dmesg log: https://pastebin.com/43BdiwP3

CPU frequencies using command sleep 3 ; cat /proc/cpuinfo | grep 'cpu MHz': ✔

Code: Select all

cpu MHz		: 800.000
cpu MHz		: 800.017
cpu MHz		: 790.881
cpu MHz		: 800.006

my HW configuration:
» RAM: 1x 16GB Crucial 3200MHz (CT16G4SFRA32A)
» SSD: 1x SATA 480GB KINGSTON (SA400S37480G)

Code: Select all

DietPi benchmark results:

Benchmarks completed: (max 12.95 W ⚡️)
 - CPU performance : Duration = 2.15 seconds (lower is faster)
 - CPU temps       : Idle = 36 °C | Full load = 42 °C         
 - RootFS I/O      : Write = 312 MiB/s | Read = 444 MiB/s     
 - RAM I/O         : Write = 1807 MiB/s | Read = 4837 MiB/s 
 
Benchmarks completed: (12.90 W ⚡️ max)
 - CPU performance : Duration = 2.12 seconds (lower is faster)
 - CPU temps       : Idle = 32 °C | Full load = 38 °C
 - RootFS I/O      : Write = 310 MiB/s | Read = 443 MiB/s
 - RAM I/O         : Write = 1771 MiB/s | Read = 4851 MiB/s

However unplugging and plugging HDMI causes some DMAR error messages in dmesg log:

Code: Select all

~$ sudo dmesg:
[ 5291.579468] r8169 0000:01:00.0 eth0: Link is Down
[ 5294.832335] r8169 0000:01:00.0 eth0: Link is Up - 1Gbps/Full - flow control off
[ 5318.382933] DMAR: DRHD: handling fault status reg 2
[ 5318.383029] DMAR: [INTR-REMAP] Request device [00:02.0] fault index 0x0 [fault reason 0x22] Present field in the IRTE entry is clear
[ 5376.060983] r8169 0000:01:00.0 eth0: Link is Down
[ 5379.386817] r8169 0000:01:00.0 eth0: Link is Up - 1Gbps/Full - flow control off
[ 5395.904603] DMAR: DRHD: handling fault status reg 2
[ 5395.904684] DMAR: [INTR-REMAP] Request device [00:02.0] fault index 0x0 [fault reason 0x22] Present field in the IRTE entry is clear
[ 5399.540940] DMAR: DRHD: handling fault status reg 2
[ 5399.541021] DMAR: [INTR-REMAP] Request device [00:02.0] fault index 0x0 [fault reason 0x22] Present field in the IRTE entry is clear

//edit: I also unplugged/plugged in the ethernet cable, not just the HDMI.

[odroid]

Thank you sharing the latest BIOS test results. Basic functionality looks fine.

But I have a doubt how disconnecting/connecting the HDMI cable causes the Ethernet disconnection/connection event.
Can you check the Ethernet cable and RJ45 plug quality? Was there accidentally shake or touch the Ethernet cable while you unplug/plug the HDMI cable?
Is this a new appearing issue only after flashing the BIOS 1.11?

[xnd]

I also unplugged/plugged in the ethernet cable, not just the HDMI sorry, I could have written it more precisely.

update: when I plugged HDMI cable exactly these 7 lines appeared with [2669.x] and [2670.x]:
↓

Screen 2023-01-16 at 09.56.46.png (842.98 KiB) Viewed 978 times

and when I unplugged HDMI cable, no new lines appeared in dmesg.

[JanUl]

You posted your DRAM clock was increased to 2933Mhz from 2400Mhz or 2666Mhz to other topic.

1.08 made Kingston memory work, buts set the speed to 2400.
I made an edit to the post that yes 1.08 makes it no longer crash, but the speed is wrong: viewtopic.php?p=359630#p359630
I'm reluctant to test the Kingston memory again, because it takes a lot of time and adds mating cycles of connector - not of the H3, but from the much more expensive box I "borrow" it from for testing.

1.09 gets the speed right, but the idle power is much higher and I don't see why.

I flashed now down to 1.06 again.
The non-Kingston memory of 2666 (Samsung) and 2933 (SKhynix) get initialized correctly AND idle power is 1.xW - as it should be.
Consumption varies a bit, I guess to due memory size, memory brand, H3 or H3+, and how hot the PSU currently is.
The only BIOS changes from default are 1) disable secure boot, 2) set 3x PCIe ASPM to "auto", 3) enable PXE boot

So I guess I will freeze this configuration now, and continue with the H3(+) evaluation....

Could you please post? sudo dmidecode -t processor | grep Speed

[fvolk]

Could you please post? sudo dmidecode -t processor | grep Speed

I'm now on BIOS 1.11 and with my H3 that plays NAS:

$ dmidecode -t processor | grep Speed
Max Speed: 2900 MHz
Current Speed: 2400 MHz

?

[JanUl]

Thank you for reply,
I bought H3 recently coming with BIOS 1.2 After some time I decided to upgrade BIOS suo I pushed the latest version. The idle power consumption jumped from 2.7W to 4.7W meaning also reasonably higher temperatures. First I was wondering why this happened. Then I activated ASPM everywhere and deactivated ports which I do not use (all SATA, I LAN port). Now I am at 3 which is good.
With bios 1.2 there was not necessary to deactivate all SATA. I believe the whole reason will be in CPU state. However at the time of asking you the output of the command neither now I know what exactly means the current speed (mine is 2700 MHz, max 2.8 or 2.9GHz, on H3 non +), because
watch -n.1 "grep \"^pu MHz\" /proc/cpuinfo"
shows values up to 2000 MHz, mostly around 800 - 900 MHz.
Anyway, now I am ok with 3 W, so I will see than what else I can do with it.

[fvolk]

The first BIOS releases had various issues, however all the serious issues and incompatibilities got fixed meanwhile - please use the latest one.
As power-saving also depends on all the devices working correctly, I believe Hardkernel does not ship the BIOS with all power-saving options enabled to the max, so customers do not complain about weird errors and incompatibilities and blame the H3(+) - although their problems may actually come from cheap non-brand SSDs etc. The more advanced customers can enable everything that's possible and get a ~2W idle NAS.

The "Max Speed" reported by dmidecode is the hardware limit, which is correct for the CPU.
The "Current Speed" reported by dmidecode appears to be wrong.

I have my cores deliberately limited to 2 GHz:
$ cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
2000000
...so that I never have to use a fan for cooling, even under long running loads.
Therefore the "Current Speed: 2400 MHz" reported above by dmidecode must be wrong.

The current CPU freqs seem to be correct in /proc/cpuinfo.
You can also show the current core frequency in "htop" utility: Setup(F2) -> Display options -> Also show CPU frequency
(but note that some versions of htop were buggy and showed wrong freq)

[xnd]

I've added second RAM module (32GB Crucial 3200 Mhz (PC4-25600), CL22, CT32G4SFD832A) to my H3 with 1x 16GB Crucial.
Now idle power consumption increases ~0.5 W, so it's about 1.6 — 2W (my previous measurements: viewtopic.php?p=363931#p363931) (I'm still on BIOS 1.11, kernel 6.0.0-0.deb11.6-amd64)

UPDATE: I forgot to unplug HDMI. Now it's 1.3 — 1.5 W