Thank you for sharing! Did you notice any drawbacks in terms of performance?
CPU wise they both idle at a few percent, no changes there.
Boot up is much faster on the T530 with the SSD vs the laptop with mechanical HDD.
Excellent! How about esp home compiling? I have serious wait times every time I modify a configuration with my box. Compiling takes up to 8 minutes
Still the odroid is more than the double efficiency compared to your linked intel cpu.
(Multi-core / watt performance)
Performance / watt ratio
Intel: 187 pts / W
Rockchip 399 pts / W
So intel should be much better keeping a coffee hot as it produces more heat (which often results in more fan noise too).
What is it that compiles 8 hours minutes on your HA server - ESPHome? I just compile on my 8-core (non intel) workstation and that takes hardly more then 2 minutes.
I just want best of both worlds 
Peak Performance AND efficiency 
I was playing around with ESPHome Pylontech Integration (Pylontech Battery ā ESPHome) yesterday and it was a pain in the buttā¦
I am currently looking into Intel N100 Chips which, according to online reviews, reach 95% of their speed at a 10W P2/P1 power limit (they can go up to 15W) and have a TDP of 6W. So low idle consumtion but performance that is more or less on par with a i3-8100.
What really scares me off with the odroid m1s or similar systems is that the mass storage is not interchangebale. My current HA Installation already takes up 110GB and itās not getting smaller, the more I add. So this seems like a one way street for me.
Odroid M1S with 4GB Ram and 64GB mass storage costs 95 EUR / 103 USD in Germany while I can get a N100 equipped mainboard (Asus Prime N100I-D D4-CSM) for a few bucks less.
Sure, I would have to add a PSU, SSD and RAM but I have those lying around anyway so no change in cost. I do not need/want Wifi, so thatās not an issue, too.
Also this rig might be able ot play Doom Eternal @640x480. Soā¦
What do you think?
To my knowledge the odroid takes m2 SSDās in a dedicated slot. You might mistaken it for a HA green which only has onboard eMMC an only allows extending storage via USBs?
IMHO it makes no sense to have a powerful ha server with high idle consumption just to speed up compiling while still maxing out the usage. As told you can easily compile esphome nodes on any other client too
Pollin sells the 4/64GB kit with case and PSU for 80 EUR
Thank you very much for pointing that out. I missed that. As soon as I read ā64GB emmcā iāve stopped reading
Thanks for the tip!
Can you please elaborate on the compiling part? I am really really new to all of this.
I am having performance issues with backups, updates and some statistics already.
An update of HA with a backup takes about 30-45 minutes atm.
While the system updates, my automations for solar power obviously do not work since HA is not available. This pushes updates to night time. Which disrupts the power feed from battery systems controlled by HA to my grid.
Which pushes updates to points in time where both the sun ins down and the batteries are empty
I do realize that my initial request was a more power efficient system and I recognize your recommendation as a really solid platform.
When I was starting with HA about 12 months ago I had a limited number of devices, automatisations and most of all statistic data. I was more than happy with the performance of the old Celeron 1037U back then. I was amazed how smooth it all runs.
Utilizing more and more of HA, embedding live CAN-BUS communication with solar chargers which do generate TONS of data etc. is slowly but steadily slowing down the HA Installation.
This is why I am looking for a more powerful aka future-proof option. I will be extending the data input to HA very soon (new CAN-BUs solar data, ~30 more sensors/relays. 6 AC Systems etc.)
Is there a more powerful version of the ODROID-M1S that retains the performance per watt ratio?
Comparing the power per watt ratio itās very hard to compare the ROKCHIP vs the newer intel chips since they can draw much more than the TDP might suggest.
Looking at the same source with the Geekbench 5 Multi-Core Benchmarks:
ROKCHIP: 56.2 pts / Watt (5W TDP)
Intel N100: 147.46 pts / Watt (15W p2 turbo limit usage)
Intel N100 @90%: 199.08 pts / Watt (With 10W max limit)
1 Like
Just install esphome on your (powerful) client like the docs describe (docker or pip)
That would probably a device with a RK3588(S) instead of the RK3566 that the odroid m1(s) and the HA green has.
Downside is that it doesnāt look like any device with a RK3588(s) is (yet) supported by HaOS.
I think that was always the actually. If a intel CPU has TDP of 15 watt for example the cooling needs to be designed for a 15w āheating sourceā - so I understand 100% of this 15W goes into heat (wasted energy) and the couple of more Watts are actually doing something.
Suggest that finally intel seems to produce (efficent) processor again and not only heaters 
1 Like
Unfortunately thatās not the case with intel. The TDP specs are more or less useless since the chips can power in their P2 or P1 states for as long as they desire if you have sufficient cooling.
For example:
A i5-13500 (https://www.intel.de/content/www/de/de/products/sku/230580/intel-core-i513500-processor-24m-cache-up-to-4-80-ghz/specifications.html) has a TDP of 65W but can continously(!) draw 154W if you simply set the BIOS or XTU limit to this value.
The System will throttle if cooling is insufficient or the motherboard cannot sustain the power delivery.
This makes comparing benchmark results so absurdely hard since a properly cooled I3 will outperform an improperly cooled (or configured) I9-13900K
Same goes for the N100 chipsā¦ usually you only get about 10% more performance out of the upper 50% wattage. So for example the i5-13500 mentioned above will perform at 90% at around 100-110W and the additional 54 Watts only result in about 10% more performance.
In essence, the TDP Value is just the mini the minimum cooling capacity youāll need for the cpu to run at base clockspeed.
Speeking of the N100, it seems to reach 90% of its speed at 166% of itās TDP which is 10 Watts. Throw another 5W at it, youāll get 10% more. Needless to say, this is not efficient.
On the other hand it opens up scenarios where you want that extra 10% of speed with no regards to power efficiency so I personally quite like the road intel has taken as it gives me a choice.
I sell specialized Workstation notebooks with I9-HX CPUs for that very reason: you can run it at a cool base clock speed but fire up a second dedicated PSU if you need workstation class power on the road. We modify an aluminum notebook casing for this with heat transfer to the entire frame. The CPU can boost up to 170W then and sustain this for minutes before we get our wieners and steaks to roast on the bottom of the case
Iāll look into that, thanks! 
That seems excessive, the old system took me 2 minutes, at most. Ill try again closer to the weekend.
first quick update on the Intel N100 experiment.
Iāve equipped the board with:
- 480 GB M.2 Samsung 980 EVO
- 4GB DDR4 3200 MHz Ram
- 80mm Noctua Fan (not needed actually)
- LAN Connection
- USB Dongle for Keyboard/Mouse
- BIOS in default settings with Audio enabled (for example)
and installed Windows on it (just for the sake of a GUI^^).
The BIOS of the PRIME N100I-D D4-CSMļ½Motherboardsļ½ASUS Global motherboard offers three Power modes.
- āPerformanceā (default)
- "Non-Turbo Performance (wtf?)
- Battery Saver
The system is being powered atm by an ultra-sketchy 25 year old 12V DC/DC Pico-PSU that is connected to the cheapest Multi-Voltage AC/DC Adapters I could find.
So probably something in the area of ~50-60% power efficiency in regards to the power supply alone.
Initial tests show:
- ASUS sets a 6W continous and 25W P2 Limit that can be sustained for 25 seconds
This getās modified by the Bios setting aboveThis is only valid during boot and post. CPU Frequency gets limited until Windows has started. Then the 6W/25W limit takes over and the CPU turbos either way- With just the smallest of fan youād probably be able to run the CPU at 25W continously (but thatās not important for us here)
Power Consumtion in Performance mode:
* Load: 25W Package Power (CPU), 35W Full System
* IDLE: 1.2W Package Power (CPU), 8.6W Full System
Power Consumtion in Battery Saver Bios mode:
* Load: 15W Package Power (CPU), 25W Full System
* IDLE: 0.8W Package Power (CPU), 7.5W Full System
EDIT: Further Testing reveals: BIOS Setting is irrelevant (see above). Power Consumtion after Windows Boot is the same in Performance and Battery saver mode:
- Load: 15W Package Power (CPU), 25W Full System, Sometimes 35W for very short amount of time (or when the GPU kicks in full time)
- IDLE: 0.8W Package Power (CPU), 7.5W Full System
Keep in Mind, that the bullshit power supply construction Iāve got going takes 4-5 Watts aloneā¦
In regards of real world performance there is barely no difference between the performance and battery saver mode since the CPU reaches 90% potential at 10-15W anyways.
I am VERY happy with the initial results and will do a test with a proper PSU soon.
Mean Well 12V PSUs are around 13 Euros and claim an 88% Power Efficiency rating. If i can get IDLE Consumtion in the range of 5 Watts I am very happy since I would then have the best of both worlds: High performance when needed but low Idle Consumtion
Iāll keep you guys posted.
3 Likes
0.25A * 12V == 3W IDLE
Measured between 12V adaptor and the PicoPSU (DC PSU for ATX)
4.8W IDLE measured on the AC Outlet in front of the 12V AC/DC Power Unit.
Setup:
- N100 Asus Motherboard
- 25 years old 200W Pico-PSU
- Mean Well LRS-50-12 AC/DC Power Supply set to 12.2V (12.2 seems to be the sweet spot)
- 480 GB M.2 Samsung 980 EVO
- Noctua 80mm PWM Fan running at ~350RPM
(Will replace with a SATA drive for more serviceabilty - NO Keyboard/Mouse
- NO VGA
Connected VGA + Input Devices add another 2-4 Watts!
Thoughts?
I also found out that VGA adds about 2W. Was actually surpised. But I guess this wont be a problem, as normally there is no monitor. Wifi, if youāre using, it will add about the same. The 5GHz types in particular are quite power hungry.