1 Watt server challange!

[ankith26]

Don't you realise that a typical 10/100MB ethernet connection alone uses about 1/2Watt, to send out ethernet signals?

good luck with this!

If the above is true( i am sure it is) then i would use a raspberry pi zero W ... I would turn off bluetooth, and wouldnt connect any extra device to gpio or hdmi... I feel that this would work consuming less current than 200mA( at 5V voltage)...

[whiteshepherd]

Like I said a Pi 0/W is recommended which can be run at 80-100 ma in the right configuration.

https://www.jeffgeerling.com/blogs/jeff ... rve-energy

Even if you can't hit 1 watt. Fell free to post your lowest power progress so far!

[ankith26]

Sadly i don't have a raspberry pi 0 yet. My 3b consumes 2-3 watts on average (idle). didn't do anything to reduce it though

[PiGraham]

Like I said a Pi 0/W is recommended which can be run at 80-100 ma in the right configuration.

https://www.jeffgeerling.com/blogs/jeff ... rve-energy

Even if you can't hit 1 watt. Fell free to post your lowest power progress so far!

But is that "the right configuration" for a "viable server"?
How about you post your progress so far and a definition of "viable server".

[rpdom]

There are many ways you can cut power down on a Pi 0/W.
Turn off LEDs

That would be LED as there is only one. It will save a few mA.

turn off HDMI (big one)

Turns off automatically if no HDMI display is connected.

bypass linear power regulator and use a high efficiency buck

There are no linear regulators on a Pi Zero. It is all high efficiency switched mode.

I'm not saying there aren't ways of saving power, but the Zero runs about as low as it can already.

[hippy]

The goal of this "for fun" contest is to see what kind of Raspberry Pi server we can build that will run stable on 1 watt or less (either under 1 watt all the time or averaging less than 1 watt just post which).

Thanks for the clarification. So basically the challenge is to keep power consumption as low as possible when idle, less than 1W when idle, and to keep it as low as possible when active, so on average it is using less than 1W.

That seems feasible and reasonable. The thing missing in that is how this average is to be defined. All servers will have a greater average consumption if they are actively serving things every second rather than once a day. We need to level the field so all servers are subject to the same loading.

I would start with the suggestion that the average should be measured for a HTTP server which serves up a 10KB file which includes the IP address of the client requesting that, along with the date and time the request is served, and an incrementing pages served count, with requests coming in every second. I am open to alternative suggestions.

The outstanding issue would be how anyone would measure their average power consumption with any degree of certainty or accuracy.

One could run it all from batteries and see how long that lasts until it stops serving but everyone would need to be using batteries with the exact same characteristics and that doesn't seem likely.

[HermannSW]

The outstanding issue would be how anyone would measure their average power consumption with any degree of certainty or accuracy.

As posted earlier
viewtopic.php?f=36&t=238870#p1461306

powering Pi directly from 3.3V or 5V of a linear voltage regulator (if you have one) does exactly show how many milliamps are used.

[hippy]

The outstanding issue would be how anyone would measure their average power consumption with any degree of certainty or accuracy.

As posted earlier
viewtopic.php?f=36&t=238870#p1461306

powering Pi directly from 3.3V or 5V of a linear voltage regulator (if you have one) does exactly show how many milliamps are used.

That doesn't usually work because meters won't show the peaks of current being drawn when those are of short duration. One cannot determine an accurate average unless there is a complete and accurate record of current being drawn over time. What a meter or bench supply may show is not necessarily the same as it actually is. One cannot rely on those determining an accurate average.

All that would tend towards doing is determining the idle current, the idle power consumption, not the average.

It wouldn't particularly surprise me if someone connected their Pi to a meter or bench supply and that appeared to show average current consumption was always below 1W, even when it actually isn't.

[PiGraham]

Ok, as the original poster I'm going to be more clear. The goal of this "for fun" contest is to see what kind of Raspberry Pi server we can build that will run stable on 1 watt or less (either under 1 watt all the time or averaging less than 1 watt just post which). Voltage/amps combo used don't matter as long as that combo is 1 watt or less. The server can be anything and it can be any Pi though Pi Zero/W is recommended. Like I said it doesn't mater what you run.

Well that's easy then. I can run a "viable server" that averages that averages less than 1 milliwatt using a Pi 3B+ just by controlling the uptime duty cycle.

[whiteshepherd]

The goal is to see how creative people are in being power efficient. 1 watt is the suggested goalpost either all the time or average (post which you achieve). BE CREATIVE.

As far as the HDMI already being off if not used it's not. If you're running a headless Raspberry Pi, there's no need to power the display circuitry, and you can save a little power by running /usr/bin/tvservice -o (-p to re-enable). Add the line to /etc/rc.local to disable HDMI on boot. This will save an average of 25ma. Disabling LED/s will be 5ma per LED. Runnin at different voltages may produce better results. Again BE CREATIVE and post your results! How far can you push the bar and have things like a running web server, ftp, or any other services you find useful, etc???

For myself to measure average power used I have a 5V 1 Farad super capacitor which I'll put in parallel with the Pi 0W to average out spikes. I'll then use a ohm meter to measure/average my power used over a 5 min sample time. You can use a different method. Post how you chose to measure power used and again BE CREATIVE.

The how point of this post is to see how creative people can get!

[hippy]

The goal is to see how creative people are in being power efficient ... The how point of this post is to see how creative people can get!

That's somewhat different to how the challenge was first presented, but that's okay.

I think pulling the DRAM off the chip / board, running from internal cache RAM, would save some mA, and would be considered "creative"

Not that I'm going to try and pull that off ("boom, boom"). I'll probably just wait to see what others deliver and achieve, then look to see if they've missed any tricks I can think of which could reduce consumption further.

[HermannSW]

All that would tend towards doing is determining the idle current, the idle power consumption, not the average.

I really doubt that. I already stated what all is running, two ESP32 cores, Wifi, and camera. The CIF image (400x296) is live video, updated at 25fps. So the 1st core has to receive 25 frames per second from ov2640 camera, and then send as mjpg stream to browser over Wifi. The second ESP32 core handles Wifi. The camera is running the whole time and consuming current as well. If you think that is idle, then your definition of "idle" is not the one I use. The 25fps live mjpg stream you see on right side in the browser is the one taken with the ESP32 ov2640 camera:



So I will connect a small VGA USB camera to Pi ZeroW when back home on the weekend and post the number measured with it, and for comparison with v1 and v2 camera as well. I am pretty sure that Pi ZeroW powered with same linear voltage generator at 3.3V or 5V will consume much more current for running live 25fps browser video than the ESP32-CAM 186mA (0.61W) ...

[hippy]

All that would tend towards doing is determining the idle current, the idle power consumption, not the average.

I really doubt that. ... If you think that is idle, then your definition of "idle" is not the one I use.

I wasn't suggesting your set-up is idle. What I am saying is that measuring, how you seem to be doing it, would tend towards showing what the idle consumption were on a Pi server if it were only seeing brief spikes in current when it is activated to serve something, and otherwise idling, waiting for those.

Maybe what you are using is somehow special but I have never known a meter to accurately measure or show very brief voltage or current spikes.

[PiGraham]

Maybe what you are using is somehow special but I have never known a meter to accurately measure or show very brief voltage or current spikes.

A typical DMM won't show spikes, but if you want the average you don't want the spikes.

[drgeoff]

I'll then use a ohm meter to measure/average my power used over a 5 min sample time.

That's not creative. Just plain wrong.

[Gavinmc42]

I use i2c high side current ADCs.
Log the current 100+ times per sec and then average.
It's not perfect but better than any multi meter.
Very useful for RF devices that have low duty cycles.

You could make a webserver that displays it's own power usage

[whiteshepherd]

I'll then use a ohm meter to measure/average my power used over a 5 min sample time.

That's not creative. Just plain wrong.

Why is it plain wrong? The capacitor would help average out the quick spikes and the meter could give me my amp draw.

[ankith26]

Ok, as the original poster I'm going to be more clear. The goal of this "for fun" contest is to see what kind of Raspberry Pi server we can build that will run stable on 1 watt or less (either under 1 watt all the time or averaging less than 1 watt just post which). Voltage/amps combo used don't matter as long as that combo is 1 watt or less. The server can be anything and it can be any Pi though Pi Zero/W is recommended. Like I said it doesn't mater what you run.

Well that's easy then. I can run a "viable server" that averages that averages less than 1 milliwatt using a Pi 3B+ just by controlling the uptime duty cycle.

Did you just type 'milliWatt' or its a typo.. i think you meant "watt" right?

[PiGraham]

Did you just type 'milliWatt' or its a typo.. i think you meant "watt" right?

Not a typo. I was making the point that we could get arbitrarily low average power by controlling the duty cycle of uptime. There is no specification for uptime given. We could choose 1 minute per week. Allow another minute to boot and assume 2W instantaneous consumption.


The average consumption would be 2 mins uptime x 2W / minutes in a week
which comes to 3.96 mW. 2 mins once per month would be ~1mW.


There are plenty of options for microwatt power devices that could bring the Pi server up periodically without adding significantly to the overall power consumption.

The 'challenge' needs a specification to make any sense.

[hippy]

A typical DMM won't show spikes, but if you want the average you don't want the spikes.

But how is the actual average determined if those spikes aren't being recorded ?

If a server idles at 100mA, sees a 1ms spike of 800mA when accessed, the average current if accessed once a second is -

( ( 999 * 100 ) + ( 1 * 800 ) ) /1000 = 100.7mA

If it is being accessed more quickly, 500 times a second -

( ( 500 * 100 ) + ( 500 * 800 ) ) / 1000 = 450mA

Without recording those spikes the average will always appear to be 100mA. That's what I meant about tending towards only getting an average of idle current, not an actual average current.

It also shows the rate of accessing the server has a great bearing on things.

Now let's assume that the Pi hardware is modified and the idle current is reduced to 10mA, but access current is still 800mA. The average isn't 10mA as seen without spikes, but -

( ( 999 * 10 ) + ( 1 * 800 ) ) /1000 = 10.8mA

( ( 500 * 10 ) + ( 500 * 800 ) ) / 1000 = 405mA

So those spikes have to be recorded and taken into account.

[hippy]

Did you just type 'milliWatt' or its a typo.. i think you meant "watt" right?

Not a typo. I was making the point that we could get arbitrarily low average power by controlling the duty cycle of uptime.

Indeed. I slashed a notional third off my server consumption by powering it down for 8 hours overnight.

[drgeoff]

I'll then use a ohm meter to measure/average my power used over a 5 min sample time.

That's not creative. Just plain wrong.

Why is it plain wrong? The capacitor would help average out the quick spikes and the meter could give me my amp draw.

An ohm meter measures resistance.

[PiGraham]

A typical DMM won't show spikes, but if you want the average you don't want the spikes.

But how is the actual average determined if those spikes aren't being recorded ?

If a server idles at 100mA, sees a 1ms spike of 800mA when accessed, the average current if accessed once a second is -

( ( 999 * 100 ) + ( 1 * 800 ) ) /1000 = 100.7mA

If it is being accessed more quickly, 500 times a second -

( ( 500 * 100 ) + ( 500 * 800 ) ) / 1000 = 450mA

Without recording those spikes the average will always appear to be 100mA. That's what I meant about tending towards only getting an average of idle current, not an actual average current.

It also shows the rate of accessing the server has a great bearing on things.

Now let's assume that the Pi hardware is modified and the idle current is reduced to 10mA, but access current is still 800mA. The average isn't 10mA as seen without spikes, but -

( ( 999 * 10 ) + ( 1 * 800 ) ) /1000 = 10.8mA

( ( 500 * 10 ) + ( 500 * 800 ) ) / 1000 = 405mA

So those spikes have to be recorded and taken into account.

There will be inacuracies, of course. But a meter with a low pass anti-aliasing filter matched to the sample rate will continuously average the signal so spikes contribute to the reading. You should see the difference between the two activity patterns you mention on most current meters. You should see it on a moving coil ammeter.

What you describe would occur due to aliasing if the input is not filtered. Aliasing would mean the sampling missed the spikes entirely some of the time.
If you wanted to be very accurate you could use a true RMS power meter

[whiteshepherd]

Spikes is why I also suggested trying a super capacitor (a few farad) if you didn't have something to sample at a high rate. The super capacitor (though not perfect) should catch/smooth the spikes (if they are frequent enough) as the super capacitor increases the load to recharge between spikes?

Another thought using super capacitors though I would have to figure out the math. Since the Pi will run from V5 to 3V another way to measure power used (over a short window) I could use two of my 3500F caps in series and charge them to 5V then running just on the caps see how long it takes for the voltage to drop to 3V. Based on the time taken to hit 3V could mathematically get a good average of current draw.

[whiteshepherd]

That's not creative. Just plain wrong.

Why is it plain wrong? The capacitor would help average out the quick spikes and the meter could give me my amp draw.

An ohm meter measures resistance.

Most ohm/multi meters can also measure voltage and current draw if you didn't know....