Measuring power draw from 6-20R receptacles.

Status
Not open for further replies.
P

preacceleration

Member
Location
Washington State
Our data center is using 6-20R receptacles to power Bitcoin miners at 208V.
We don't have metered PDUs and need a way to measure their power draw.
They run 24/7 so if I measure it once I can calculate their electricity usage for the month.


I was thinking about using a clamp multimeter with a line splitter to measure the power draw, but couldn't find a line splitter for 6-20R.

Any bright ideas of how the power draw could be accurately measured?
 
gar said:
How do you expect to measure power with a multimeter?
Click to expand...

He probably thinks he can multiply the current he measures by 208 and have power. And if the systems have power factor corrected power supplies that will be pretty close (especially if he has a true RMS clamp meter), but otherwise it could be pretty far off.
 
preacceleration said:
Our data center is using 6-20R receptacles to power Bitcoin miners at 208V.
We don't have metered PDUs and need a way to measure their power draw.
They run 24/7 so if I measure it once I can calculate their electricity usage for the month.


I was thinking about using a clamp multimeter with a line splitter to measure the power draw, but couldn't find a line splitter for 6-20R.

Any bright ideas of how the power draw could be accurately measured?
Click to expand...
keep in mind that power is measured in watts and billed in watt/hours. Your clamp on meter is measuring amps, doesn't consider power factor or other distortions that may effect true watts. If the load measured were pure resistance load you should be able to determine watts or watt/hours with pretty fair accuracy just with voltage and amps readings.
 
kwired said:
keep in mind that power is measured in watts and billed in watt/hours. Your clamp on meter is measuring amps, doesn't consider power factor or other distortions that may effect true watts. If the load measured were pure resistance load you should be able to determine watts or watt/hours with pretty fair accuracy just with voltage and amps readings.
Click to expand...

The miner's power supplies are PFC, with PF > 0.99. Do you think the distortions will be an issue?
 
preacceleration said:
How would I make my own splitter?

I could take some of the 6-20P to C13 power cables I have and separate the conductors in them, does that sound like a good idea?
Click to expand...
It may not be code compliant if you intend to leave it in place rather than just using it for testing, but it seems like a good idea.
If you want compliance cut the cord and rejoin the ends inside a j-box. Only open the cover for testing.
Or is there some other place on the branch
circuit, such as the breaker box, where you can access the wires?
 
preacceleration said:
The miner's power supplies are PFC, with PF > 0.99. Do you think the distortions will be an issue?
Click to expand...
Shouldn't if you are using a True RMS meter. If product "features" in essence state regardless of waveform, all the better, as some are better than others. If it don't state something similar, look for a frequency spec'; higher rate is typically better.

preacceleration said:
How would I make my own splitter?

I could take some of the 6-20P to C13 power cables I have and separate the conductors in them, does that sound like a good idea?
Click to expand...
You could... for "temporary" testing purposes only. However, keep in mind that to get true power, you need not only measure true rms current, but also true rms voltage... which TTBOMK requires conductor contact.

If I were going to be doing this measurement on a regular basis, I'd get a kwh meter similar to one I linked earlier, mount it in a box with display exposed, cut such a cord (or two), and wire it into box using strain reliefs at the penetrations.
 
FWIW, this item might help your cause...

http://www.tequipment.net/ElencoST-74.html

It can be found from other suppliers, but beware that you may get something other than pictured. All advertisements I've seen state its an ST-74.... and it is the same model number as one from same manufacturer with 5-15P/R. Also note in picture that while the R appears to be 6-20, the P is not, so 1) you'd need an adapter cord, and 2) seems to indicate it is not certified for 20A use.
 
Smart $ said:
Shouldn't if you are using a True RMS meter. If product "features" in essence state regardless of waveform, all the better, as some are better than others. If it don't state something similar, look for a frequency spec'; higher rate is typically better.


You could... for "temporary" testing purposes only. However, keep in mind that to get true power, you need not only measure true rms current, but also true rms voltage... which TTBOMK requires conductor contact.

If I were going to be doing this measurement on a regular basis, I'd get a kwh meter similar to one I linked earlier, mount it in a box with display exposed, cut such a cord (or two), and wire it into box using strain reliefs at the penetrations.
Click to expand...

How could I measure the voltage at the load? We have identical brand new secondary transformers, with identical paneling, so is it likely that the voltage will vary at the load at different places in the datacenter?

Also, I looked at the kwh meters. I'm not sure if you're suggesting that I implement permanent metering, or are showing how I can make a measuring device using the meters.

The documentation says its suitable for 1P + N. I think the loads are connected phase-to-phase at 208V, so would these meters work for that application?

And do they even measure instantaneous power draw, or do they only measure cumulative power usage?

Thanks for the help and clarification.

Robert
 
preacceleration said:
How could I measure the voltage at the load? We have identical brand new secondary transformers, with identical paneling, so is it likely that the voltage will vary at the load at different places in the datacenter?
Yes, a result of voltage drop, but perhaps not drastically... yet even the source voltage can vary regardless of voltage drop.

Also, I looked at the kwh meters. I'm not sure if you're suggesting that I implement permanent metering, or are showing how I can make a measuring device using the meters.
Both.

The documentation says its suitable for 1P + N. I think the loads are connected phase-to-phase at 208V, so would these meters work for that application?
Decide what and how you want to do it, then we can look at spec's.

And do they even measure instantaneous power draw, or do they only measure cumulative power usage?
See below.

Thanks for the help and clarification.

Robert
Click to expand...
There's a lot of details you are not being specific about here, IMO. Let's start with exactly what you desire to achieve.

First, if you want something to compare with what you are getting billed for, you'd want to measure as close to POCO meter as possible. Measurements closer to the load will omit I?R wiring losses to that point in the system.

Next, an instantaneous power reading (true or otherwise) will not be a good comparison if the load fluctuates during operation.

Watts is a measure of power. KWH is a measure of energy (accumulative power usage). One power measure times time will give you long term WH/KWH only if power usage is constant.
 
Smart $ said:
There's a lot of details you are not being specific about here, IMO. Let's start with exactly what you desire to achieve.

First, if you want something to compare with what you are getting billed for, you'd want to measure as close to POCO meter as possible. Measurements closer to the load will omit I?R wiring losses to that point in the system.

Next, an instantaneous power reading (true or otherwise) will not be a good comparison if the load fluctuates during operation.

Watts is a measure of power. KWH is a measure of energy (accumulative power usage). One power measure times time will give you long term WH/KWH only if power usage is constant.
Click to expand...

Bitcoin miners run at full capacity 24/7. Ideally their power draw is constant and continuous. In addition, once we've measured enough of them we'll be confident that a miner of a certain type will have a specific power draw and won't need to measure it. We still may need to remeasure in the case a customer wants to over/underclock their miner, or their are firmware updates to the miner which affect power draw.

I'm looking for an accurate and quick way to get an accurate "at-the-wall" measurement of the wattage of the miners, with the assumption that their loads are constant and continuous, given that we have 6-20R outlets.
 
Since in this situation you are looking at the relative profitability of the miners, I would think that you would want to get a high accuracy in the measurement, especially if the numbers will be used for extrapolation.
It is possible to modify a commercial product like the KillAWatt to work with other than 120V 15A. This would give you an instantaneous reading, with pretty good accuracy.
But a revenue grade energy meter left running for an hour would give a more authoritative number. My only concern would be whether it has a fine enough resolution readout for the job.
For example, if the server draws 2kW, a resolution of .1kWh over one hour would only be a precision of 5%.

I would also be concerned about the need to shut the server down to insert and remove the meter.
A "permanent" installation of a voltage test point and clamp on meter box might make more sense.
One more concern is whether periodic events in the OS or application would cause the power consumption to vary over time.
 
preacceleration said:
Bitcoin miners run at full capacity...
Click to expand...
I admit I am not familiar with the process or any purpose-dedicated equipment, but it's a common misconception in general that full time data processing equipment draws constant power.

Giving GoldDigger's reply into consideration in addition to yours, I would think you can get the measurement you are after more easily on the wiring side of the receptacle, if relatively accessible and safety hazards can be mitigated to industry practice. Use a clamp-on power meter you feel is up to the task spec' wise, with peak hold feature and clip-on leads for the voltage connection. You can then take an instantaneous measure, invoke peak hold and leave it set for awhile, then defeat peak hold and get another instantaneous. Wouldn't hurt to get a certified calibrated meter either.
 
preacceleration...

It is relatively easy to increase the accuracy of any clamp-On kW or kWhr meter!

Instead of clamping a single current-carrying conductor just wrap several turns of the conductor around the jaw of the meter! Simpler yet would be to purchase an Ampere-Multiplying-Adapter, Model ALS-1, manfactured by Universal Enterprises, at an Electrical-Supply house! It can safely increase the kW or kwhr meter reading by a factor of 10!

Regards, Phil Corso
 
Phil Corso said:
preacceleration...

It is relatively easy to increase the accuracy of any clamp-On kW or kWhr meter!

Instead of clamping a single current-carrying conductor just wrap several turns of the conductor around the jaw of the meter! Simpler yet would be to purchase an Ampere-Multiplying-Adapter, Model ALS-1, manfactured by Universal Enterprises, at an Electrical-Supply house! It can safely increase the kW or kwhr meter reading by a factor of 10!

Regards, Phil Corso
Click to expand...

All the line splitters I've seen are for 120V 15A, maybe a few that go up to 250V. Biggest problem is none are 6-20R and I can't find adapters either.
 
150101-2409 EST

preacceleration:

I don't believe you have clearly defined exactly what your goal is.

Why do you want to know the power used, with what accuracy, and do you need to know the energy used? You have already been given many good ideas.

To make a splitter it is unlikely you will find 6-20 plug and socket components for cords at most stores. Consult the Hubbell catalog, and then find a supplier that has stock or will order what you need. Use Nylon coated #12 stranded wire to interconnect the plug and socket.

If you use one of the clamp-on meters suggested, then you may want to make a 10 turn coil to increase the sensitivity of the instrument.

If you don't need high accuracy for initial experiments, then because your PF is essentially 1 measure the RMS current and voltage and multiply together for power. This would allow you to see how power varies with computer activity.

No way do you want to know instantaneous power this would be hard to use. A one second averaging of power might be realistic for your initial use.

If short time average power shows swings of 50%, then I doubt you really want to work with a short time average power. Thus, Wh (energy measured over a long enough time period to be a stationary random process) is probably the measure with which to work. We need to know the reason for your measurement.

A TED system adapted to work with 208 input and connected to a computer could be useful. To do this you would use a small 208 to 120 transformer. There are more details than that, but not real difficult to do. If you could find 1000 system components, then 1 second resolution and integrated over a 24 hour period could be saved before having to save the data to another file.

.
 
Status
Not open for further replies.
Top