![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
How to Measure SATA power draw
- Thread starter Computurd
- Start date
Red Falcon
[H]ard DCOTM December 2023
- Joined
- May 7, 2007
- Messages
- 12,454
This has a few methods:
http://www.xbitlabs.com/articles/storage/display/hdd-power-cons.html#sect0
http://www.xbitlabs.com/articles/storage/display/hdd-power-cons.html#sect0
thx, i did see that...but i will not beat around the bush...i am not very knowledgeable with assembling things of that nature..
you would imagine there would be some sort of in-line monitor or a device you use to power the drive that would give a measurement?
you would imagine there would be some sort of in-line monitor or a device you use to power the drive that would give a measurement?
Red Falcon
[H]ard DCOTM December 2023
- Joined
- May 7, 2007
- Messages
- 12,454
Just look up the drives' specs and see what the max power draw is on each, then add them up!
oh this is for some specific testing, i need actual power draw under load so that i can compare different devices to each other.
being able to measure 5V and 12V would be very beneficial.
but this would be to test the mfr specs more than anything. see the real usage under different loadings.
being able to measure 5V and 12V would be very beneficial.
but this would be to test the mfr specs more than anything. see the real usage under different loadings.
The way you measure current in a DC circuit is to break the connection, insert a resistor, and measure the voltage across the resistor. That is why in that link that Red Falcon posted there was a circuit board that you plug the drive and power supply into, there really is no way around that. You don't really need all the same stuff on that circuit board though, you just need a way to "break" the connection and insert the probes from a multi-meter.
You could probably take the power cable, cut it in half, solder/heat sink all the wires back together except for the ground wires. Bring the ground wires out to a couple of pins that you could stick a jumper on so you can still use it if the multi-meter is not connected. You could probably even very carefully slit the cable and find just the ground wires and just cut those so you don't need to worry about reconnecting the 5v and 12v wires.
You could probably take the power cable, cut it in half, solder/heat sink all the wires back together except for the ground wires. Bring the ground wires out to a couple of pins that you could stick a jumper on so you can still use it if the multi-meter is not connected. You could probably even very carefully slit the cable and find just the ground wires and just cut those so you don't need to worry about reconnecting the 5v and 12v wires.
@P.A.
Things you will need:
A decent multimeter.
.2 ohm resistor
hope
Imagine the SATA power connector, connected to a 2.5" SSD. You'll only be worried about the 5v power, so forget about the others. Splice the resistor in line on the 5v wire. So then splice in some leads before and after the resistor. Now, you can attach the multimeter leads to your before and after the resistor leads. The voltage before the resistor - the voltage after gives you the voltage drop. Using this, you can divide by the the resistance to get amps. Multiply the amperage by the nominal 5v output -- usually like 5.03V or something. Multiply the amps x 5v rail voltage to get watts, which is what you want.
For a 3.5" drive you need 5v and 12v, so you'd need twice the stuff. This is the simplest way to do this, and while it's not lab-accurate, it's pretty good nonetheless. There are other ways, and more complicated methods I've been exploring, but this is the best way.
I've ordered some resistors to do this, but I was planning on modifying a Molex to Sata II connector for this purpose. That way I could keep the ghetto-ness to a minimum, and not have to hack up sata strings.
Things you will need:
A decent multimeter.
.2 ohm resistor
hope
Imagine the SATA power connector, connected to a 2.5" SSD. You'll only be worried about the 5v power, so forget about the others. Splice the resistor in line on the 5v wire. So then splice in some leads before and after the resistor. Now, you can attach the multimeter leads to your before and after the resistor leads. The voltage before the resistor - the voltage after gives you the voltage drop. Using this, you can divide by the the resistance to get amps. Multiply the amperage by the nominal 5v output -- usually like 5.03V or something. Multiply the amps x 5v rail voltage to get watts, which is what you want.
For a 3.5" drive you need 5v and 12v, so you'd need twice the stuff. This is the simplest way to do this, and while it's not lab-accurate, it's pretty good nonetheless. There are other ways, and more complicated methods I've been exploring, but this is the best way.
I've ordered some resistors to do this, but I was planning on modifying a Molex to Sata II connector for this purpose. That way I could keep the ghetto-ness to a minimum, and not have to hack up sata strings.
Just get a Killawatt and plug your PC into it, it plugs into an outlet. You can measure real time power with one drive, then add some more and remeasure.
Power supplies also vary this is some testing I've been doing. Lowest idle power to date is Intel DH61WW and i3-2100, 18W with Seasonic 300W Bronze.
Gigabyte H55M-2SH
Core i3-550 @ 3.2GHz @ 1.216V
8GB Gskill Sniper @1.5V
500GB Black WD 2.5" drive
Win8 beta -
Ultra 400W PSU - power saving mode 44W idle
Power drops to 32W idle with Seasonic 300W Bronze
Power drops to 30W idle with FSP 400W Gold
Switched CPU to 1.15V and RAM to 1.4V = no difference
My Sempron test server with 6 2TB Samsungs idles at 41W, makes out about 85W, runs Lubuntu off a flash drive for the OS. The Asus 785G board is not efficient, also uses the same power if both cores are unlocked or just one core. Ugggg.
Power supplies also vary this is some testing I've been doing. Lowest idle power to date is Intel DH61WW and i3-2100, 18W with Seasonic 300W Bronze.
Gigabyte H55M-2SH
Core i3-550 @ 3.2GHz @ 1.216V
8GB Gskill Sniper @1.5V
500GB Black WD 2.5" drive
Win8 beta -
Ultra 400W PSU - power saving mode 44W idle
Power drops to 32W idle with Seasonic 300W Bronze
Power drops to 30W idle with FSP 400W Gold
Switched CPU to 1.15V and RAM to 1.4V = no difference
My Sempron test server with 6 2TB Samsungs idles at 41W, makes out about 85W, runs Lubuntu off a flash drive for the OS. The Asus 785G board is not efficient, also uses the same power if both cores are unlocked or just one core. Ugggg.
Last edited:
How would you measure the .3w idle of an Intel 510 120? I mean, if you were talking about GPUs, maybe, but not .3 to 3w
You do not need a resistor. You simply set the multimeter to display amps. Put the 2 meter probes across a gap in the power wires.
But the whole project seems worthless. Buy 100 drives set them up in a RAID. Watch the Killawatt change as the drives spin up and down. (Divide by 100.)
But the whole project seems worthless. Buy 100 drives set them up in a RAID. Watch the Killawatt change as the drives spin up and down. (Divide by 100.)
Ok, I drew up a quick picture. When I originally responded to your post I was picturing a sata data cable in my mind, not the power connector. Also I didn't realize you want to know how much power each voltage was using, like 12v and 5v which is why I said to leave the colored wires and instead snip the ground wires and probe those grounds with the multimeter.
The serial ata power connector actually also has a 3.3v line too.
A picture is worth a thousand words.
No, 225 is just a placeholder number showing you should see something, it could be anything. Like lorem ipsum as a placeholder for text.
Computurd ---
Are you sure you are competent to do this?
It seems to me that you want information that serves no purpose. It is not clear how much error is involved in using a multimeter. The sampling time is fairly long compared to the spin up time so you will miss the peak value. I don't see the numbers you getting being any more informative than the published numbers.
And you don't seem to get it.
Are you sure you are competent to do this?
It seems to me that you want information that serves no purpose. It is not clear how much error is involved in using a multimeter. The sampling time is fairly long compared to the spin up time so you will miss the peak value. I don't see the numbers you getting being any more informative than the published numbers.
And you don't seem to get it.
Your right George, I dont get it 
That is why I am asking. There is this thing called 'learning'. Whereas one seeks knowledge that they do not have. Forums tend to be a good place to do this, as you learn from others who possess said knowledge without having to learn a whole ton of unnecessary info to get the knowledge that you require.
Surely you aren't one of those people that believe published specs? Maybe you dont get it?
SSDs dont have spin up time in the normal sense, btw. I do not see how the spin up time is relevant, as i have not mentioned it. The numbers will be much more informative than the published numbers.
With Enterprise-class SSDs and HDDs one of the major considerations is power draw. Over a course of time power can be a substantial amount of the TOC. The power usage depends upon different factors, especially with SSDs. Now I hope that you have learned something, that is what forums are for in the first place, eh?
I am looking to do some monitoring that would be similar to this monitoring on this page;
http://www.tomshardware.com/reviews/mk4001grzb-endurance-benchmark,3132-7.html
fortunately i have found that the method using resistors is accurate to within 5 percent, and that the inline method is much less accurate.
That is why I am asking. There is this thing called 'learning'. Whereas one seeks knowledge that they do not have. Forums tend to be a good place to do this, as you learn from others who possess said knowledge without having to learn a whole ton of unnecessary info to get the knowledge that you require.
Surely you aren't one of those people that believe published specs? Maybe you dont get it?
SSDs dont have spin up time in the normal sense, btw. I do not see how the spin up time is relevant, as i have not mentioned it. The numbers will be much more informative than the published numbers.
With Enterprise-class SSDs and HDDs one of the major considerations is power draw. Over a course of time power can be a substantial amount of the TOC. The power usage depends upon different factors, especially with SSDs. Now I hope that you have learned something, that is what forums are for in the first place, eh?
I am looking to do some monitoring that would be similar to this monitoring on this page;
http://www.tomshardware.com/reviews/mk4001grzb-endurance-benchmark,3132-7.html
fortunately i have found that the method using resistors is accurate to within 5 percent, and that the inline method is much less accurate.
Your right George, I dont get it
SSDs dont have spin up time in the normal sense, btw. I do not see how the spin up time is relevant, as i have not mentioned it. The numbers will be much more informative than the published numbers.
fortunately i have found that the method using resistors is accurate to within 5 percent, and that the inline method is much less accurate.
Read your first post - SSD and HDs. HDs have spin up time.
I think my multimeter is accurate to plus or minus .5 of the least significant digit. Some people need much better equipment.
As for the power consumption being important. You might consider what the lifetime cost of power is. About $1/year for a 1w SSD. Add $1 for the A/C cost. Lifetime power cost is unimportant.
Yes, I learned a lot. You asked a dumb question. You were not happy about the answer. So you are still dumb.
if you had read the link you would see that under load they consume much more. 6.61 under load. it is tiring to talk with people who do not research what they say, when i even provide links.
in an enterprise scenario all power must be tripled in cost as well. they have to have battery backups, and generators as well, to power said devices in case of power loss. The overall power consumption of the datacenter is exceedingly important, one of the prime considerations of every single decision made in design and use. you have to provide redundancy for every single watt.
I was lucky enough to attend a speech from the head architect of Microsofts datacenters. if you want to learn something, here is the pdf from said speech...maybe you could educate us more on how power isnt important after you read this.
http://www.lsi.com/AIS2011/Documents/DatacenterEfficiencyforLSI.pdf
also you are taking a very narrow uninformed stance here. think of these in the thousands, for years, in datacenters. then triple the cost that you are estimating.
that is why it is enterprise testing.
tell that to the multibillion dollar companies that buy these products (Facebook, google), with power being one of their chief concerns.
Im not doing this testing on consumer SSDs.
i never said anything about HDD spin up times georgie. never once were they mentioned. i will attempt to keep it simple;
HDDs yes,
spin up time, no.
Well, quite honestly power is just one part of the equation. Heat dissipation and cooling is another huge item... BTU computation is just as important. You might want to read up on something like the HP power adviser app http://h20000.www2.hp.com/bc/docs/support/SupportManual/c01861599/c01861599.pdf which goes into all the points you wish to hit in enterprise resource planning. When it comes to the enterprise parts from reputable manufacturers you CAN take their measurements as a generally accepted because of just these needs. When computing your power needs, it is not just a simple "triple" your numbers. In a complete end-to-end power plan, your UPS's should be able to power your equipment until your supplementary power comes online as well as a generous margin, but im most cases you are not going to use them for continuous power. They are also there for graceful shutdowns should the secondary power not come online.
cooling is an important consideration, of course.
Cooling also figures into the power consumption figures. The amount of watts that a device consumes generally equates to the amount of heat that it generates. The heat that must be dealt with also incurs other power considerations of its own. So if you are drawing higher wattage with device A vs device B, not only do you have to provide redundancy for that power, you also have to cool that power that is generating heat as well.
of course these are among the many factors that make SSDs so attractive in datacenters.
On a side note, the power consumption for cooling is dropping, as many more datacenters are running at higher temps. Intel is recommending that they run at 80 F since the equipment can handle it. This was another thing that was discussed at length in the presentation. This also has led to other components becoming less desirable in these scenarios (supercaps, batteries for RAID controllers, etc.) that are less desirable in high heat conditions.
Datacenters are not allowed graceful shutdowns unfortunately, so their requirements for power backup are much more stringent than a normal users.
The purposes of my measurement is for product evaluations, not for actual deployment. The results will be considered by professionals in that area though, so the figures will need to be fairly accurate.
Thanks for the link to the HP Advisor, i will read that completely when i get time
i agree, I am just trying to explain the concept that it isnt as clear cut as "the device draws 1 watt at idle, therefor power is no consideration"
this is a huge concern, one that has hidden costs.
Cooling also figures into the power consumption figures. The amount of watts that a device consumes generally equates to the amount of heat that it generates. The heat that must be dealt with also incurs other power considerations of its own. So if you are drawing higher wattage with device A vs device B, not only do you have to provide redundancy for that power, you also have to cool that power that is generating heat as well.
of course these are among the many factors that make SSDs so attractive in datacenters.
On a side note, the power consumption for cooling is dropping, as many more datacenters are running at higher temps. Intel is recommending that they run at 80 F since the equipment can handle it. This was another thing that was discussed at length in the presentation. This also has led to other components becoming less desirable in these scenarios (supercaps, batteries for RAID controllers, etc.) that are less desirable in high heat conditions.
Datacenters are not allowed graceful shutdowns unfortunately, so their requirements for power backup are much more stringent than a normal users.
The purposes of my measurement is for product evaluations, not for actual deployment. The results will be considered by professionals in that area though, so the figures will need to be fairly accurate.
Thanks for the link to the HP Advisor, i will read that completely when i get time
i agree, I am just trying to explain the concept that it isnt as clear cut as "the device draws 1 watt at idle, therefor power is no consideration"
this is a huge concern, one that has hidden costs.
Last edited:
Red Falcon
[H]ard DCOTM December 2023
- Joined
- May 7, 2007
- Messages
- 12,454
SSDs only eat about 2-3 watts tops, and enterprise-grade SSDs such as the PCI-E models eat no more than 5-10 watts tops.
As for HDDs, 15K RPM drives will obviously eat the most electricity, but still, they are only at about 10-15 watts on average.
Most HDDs in the 5400RPM-10K RPM range only eat about 5-10 watts.
Are you running a very large data center where power is a concern?
Um, no. The questions Computurd is asking are not dumb.
Sometimes getting a technical readout on the actual equipment being used is far more important than the tech specs listed by the manufacturers, depending on the situation.
It really depends on the user's needs and requirements.
As for your statement, that in itself might be a little on the 'dumb' side.
I don't fully understand your project. Is this a simple thought exercise, or are you planning on designing a data center? Is this a datacenter for just your company, or are you going to be opening a colo? Are you looking to lease square footage at an existing colo? What is going to be the duty cycle of the boxes? What type of servers? N+1, N+2, N+3 power supply redundancy? All 120v? 12v? Complete hot aisles? Alternating hot/cold? Conventional servers? Blades? 32GB/server or 128GB? 80% efficiency on your power supplies or 93%? Quite honestly, the power draw and dissipation of your SSD's in a general server rack is going to be the least of your worries.
actually under load they eat more. as you can see from the Toms link, they can eat much more. over at other review sites, monitoring also shows that the real power draw under different loads can be much higher.
at this link, go down to the power monitoring of 5 different enterprise SSDs, and you will see the reason they are checking the mfr specs. they are never right!
http://www.storagereview.com/smart_storage_systems_xceediops2_enterprise_ssd_review
Red Falcon
[H]ard DCOTM December 2023
- Joined
- May 7, 2007
- Messages
- 12,454
Wow, I'm amazed it took so much power, but it also runs off of the 3.3v line as well.
Most of the SSDs I'm aware of that are SATA only require the 5v line for power.
Measuring current draw - as others have stated - is fairly simple.
There are two ways to do this: The first is to measure current directly using the DC Ammeter built into a multimeter. The problem with this is that most handheld ammeters won't do anything above 200mA so that's not going to be an option for a hard drive since you will simply damage the instrument.
The other way is to use a shunt resistor.
As we all know, V = IR, which states that voltage is equal to current times resistance (Ohm's Law). A shunt resistor and the load are connected in series, which means that the current through them will be equal (Kirchoff's Current Law). You take advantage of the fact that the resistor is a very simple way to turn a current reading into a voltage reading (voltage is much easier to measure).
However, you still need to consult the specifications of the drive under test - the reason is because you need to know how much power to expect if you are going to size your test equipment properly.
You _need_ to know the nameplate ratings for power (or current) consumption for both the +5V and +12V rails. This way, you can calculate the power that will be dissipated in your shunt resistor so that you do not set it on fire.
Resistors are rated in power capacity. Let's say that you have a load that draws 0.6 Amps, and a shunt resistor of 0.2 Ohms. The amount of power dissipated in the resistor is equal to ( I^2) / R, so (0.6^2) / 0.2 which is equal to 1.8 Watts. Therefore, you need a minimum of a 2 watt resistor otherwise it will overheat and then catch on fire.
Also, you need to calculate the voltage drop across the resistor, to ensure that the supply going into the drive under test will be acceptable. Since we know both the resistance of the resistor ( 0.2 Ohms) and the current flowing through the resistor (0.6 A) we can simply use V = IR to calculate the voltage drop across it (0.6 * 0.2) which is equal to 0.12 V. 0.12V is about 2.4% of the 5V supply rail so this voltage drop shouldn't affect the device under test excessively (the supply has fluctuations anyways so the drive must be prepared to deal with that).
There are two ways to do this: The first is to measure current directly using the DC Ammeter built into a multimeter. The problem with this is that most handheld ammeters won't do anything above 200mA so that's not going to be an option for a hard drive since you will simply damage the instrument.
The other way is to use a shunt resistor.
As we all know, V = IR, which states that voltage is equal to current times resistance (Ohm's Law). A shunt resistor and the load are connected in series, which means that the current through them will be equal (Kirchoff's Current Law). You take advantage of the fact that the resistor is a very simple way to turn a current reading into a voltage reading (voltage is much easier to measure).
However, you still need to consult the specifications of the drive under test - the reason is because you need to know how much power to expect if you are going to size your test equipment properly.
You _need_ to know the nameplate ratings for power (or current) consumption for both the +5V and +12V rails. This way, you can calculate the power that will be dissipated in your shunt resistor so that you do not set it on fire.
Resistors are rated in power capacity. Let's say that you have a load that draws 0.6 Amps, and a shunt resistor of 0.2 Ohms. The amount of power dissipated in the resistor is equal to ( I^2) / R, so (0.6^2) / 0.2 which is equal to 1.8 Watts. Therefore, you need a minimum of a 2 watt resistor otherwise it will overheat and then catch on fire.
Also, you need to calculate the voltage drop across the resistor, to ensure that the supply going into the drive under test will be acceptable. Since we know both the resistance of the resistor ( 0.2 Ohms) and the current flowing through the resistor (0.6 A) we can simply use V = IR to calculate the voltage drop across it (0.6 * 0.2) which is equal to 0.12 V. 0.12V is about 2.4% of the 5V supply rail so this voltage drop shouldn't affect the device under test excessively (the supply has fluctuations anyways so the drive must be prepared to deal with that).
I've recommended a 10w 0.20 Ohm resistor for the 5v rail and something with a higher wattage rating for the 12v rail (and/or double up the resistors). I know that most enterprises SAS drives probably hit 20w just spinning up on power-on, and how that load is split on 5v and 12v rails is not know to me.
I'm primarily interested in 2.5" SSDs which will only use 5v, so it becomes a lot simpler.
To my knowledge, most cheap DMMs don't have 250mA, usually they're 2A -- so it bears repeating not to try that.
I'm primarily interested in 2.5" SSDs which will only use 5v, so it becomes a lot simpler.
To my knowledge, most cheap DMMs don't have 250mA, usually they're 2A -- so it bears repeating not to try that.
I didn't think a current clamp would work in the sense that you'd have to isolate the 5v lead from the others, and then when you're testing 3.5" drives you need 12v and 5v. So you'd need two Flukes.
AFAIK current clamps only work for AC, aside from requiring cleanly isolating the wire, and accuracy isn't quite up to what I think OP was looking for. Myself when i'm measuring DC motor spin load (for other things like fans) I use an old analog meter that can handle up to 2A. It's so friggen old that I can't read much of anything off it but it's old school automotive diagnostic type stuff, but it's very fast to react (analog) so it's quite easy to get a pretty accurate read on even a short spike. There must still be analog meters that can handle 2ish Amps ?
FWIW I hate digital multimeters. Aside from their durability and added functions most people don't need, I just don't see the point. I guess they're cheaper too.
FWIW I hate digital multimeters. Aside from their durability and added functions most people don't need, I just don't see the point. I guess they're cheaper too.
Ok I'll admit they are pricey, I'm just used to having to buy a lot of tools/gear for work myself.
They do make clamps and probes that do DC, accurate low current measurements, multiple same phase conductors, multiple probes, graphing/logging etc etc if someone felt like going all out. I do understand you aren't looking to buy a bunch (or any) lab gear though.
And yeah I agree that your method will be more practical, was just throwing it out there.
They do make clamps and probes that do DC, accurate low current measurements, multiple same phase conductors, multiple probes, graphing/logging etc etc if someone felt like going all out. I do understand you aren't looking to buy a bunch (or any) lab gear though.
And yeah I agree that your method will be more practical, was just throwing it out there.
It seems that Comuturd is trying to make a name for himself by gathering data that has no purpose (except to a manufacturer).
I expect that efficiency and sizing of the power supply will make a larger difference in power consumption than which SSD is used.
But the difference in price of SSDs is going to dwarf the difference in electrical consumption (even if you ignore the performance needs).
----
It was clear that Computurd knew little when the first response provided a nice link about silly-scopes and multimeters that should have settled everything for him, but he continued on.
To do a reasonable job of testing and analyis will cost $2-5000. It would take a data center with 1000 drives to recoup that. Knowing the answer is not worth the cost of finding the answer.
Trepidati0n
[H]F Junkie
- Joined
- Oct 26, 2004
- Messages
- 9,269
Using the multimeter method will "lie" as well since it is filtered in a non standard way. With something like a drive, you need a scope and the appropriate style of probes.
Red Falcon
[H]ard DCOTM December 2023
- Joined
- May 7, 2007
- Messages
- 12,454
Honestly, Computurd does have access to some pretty high-end enterprise-grade tech that not many of us have a chance at touching, so this could actually be significant.
Also, if he wants to make a name for himself, who are we to stop him?
Maybe he will find something cool that will help his company or the tech industry.
In either event, I'm interested in finding out how he is going to do this and what it will accomplish.