Ideas for SSD Power Consumption Testing

[mechtech]

Interesting thread.

just curious on the background for this? More for laptops and battery conservation?

I would suggest power consumption when watching a movie that’s on the drive.

Other than that I would suggest max power usage as in what causes max power usage and how many watts is it……and when would a heat sink and or forced cooling be recommended at X amount of watts.

cheers

 

[mplayerMuPDF]

Active idle consumption (ASPM disabled) please. Some NVMe SSDs use considerably more idle than even (still spinning) idle 2.5" laptop HDDs if ASPM/L1.2 is not working. And NVMe power management can be an issue, especially on Linux, I think, so it is nice if the worst case scenario is taken into account. With SATA SSDs slumber/DevSlp not working is much less of a problem due to active idle draw being a lot less.

 

[W1zzard]

Active idle consumption (ASPM disabled) please. Some NVMe SSDs use considerably more idle than even (still spinning) idle 2.5" laptop HDDs if ASPM/L1.2 is not working. And NVMe power management can be an issue, especially on Linux, I think, so it is nice if the worst case scenario is taken into account. With SATA SSDs slumber/DevSlp not working is much less of a problem due to active idle draw being a lot less.

Yeah I’ll have „idle desktop“ which is aspm off, because that is the default setting in all (most?) bioses, and „idle mobile“ with it enabled

 

[chrcoluk]

Active idle consumption (ASPM disabled) please. Some NVMe SSDs use considerably more idle than even (still spinning) idle 2.5" laptop HDDs if ASPM/L1.2 is not working. And NVMe power management can be an issue, especially on Linux, I think, so it is nice if the worst case scenario is taken into account. With SATA SSDs slumber/DevSlp not working is much less of a problem due to active idle draw being a lot less.

Indeed, my DC P4600 uses more at idle than the spindle I took out when I put in (was only a wd green though). So idle measurements would be greatly appreciated.

 

[Wirko]

This would probably take too much of your time to include in your regular SSD test set but maybe you can do it with some SSDs:

On a drive that's (mostly) full, delete all files at once. Observe what's going on with the power consumption afterwards. Deleting should trigger the erasing of data blocks on the SSD, an operation that potentially takes a lot of time and consumes a lot of power, even if the controller is idle or nearly idle.

 

[Mussels]

This would probably take too much of your time to include in your regular SSD test set but maybe you can do it with some SSDs:

On a drive that's (mostly) full, delete all files at once. Observe what's going on with the power consumption afterwards. Deleting should trigger the erasing of data blocks on the SSD, an operation that potentially takes a lot of time and consumes a lot of power, even if the controller is idle or nearly idle.

That's not how SSD's work, they get marked as free space but don't get erased, specifically to avoid wear and tear as you described it
All that would do is mark the files as safe to be deleted, and have the files marked as junk - ready to be erased as the space is needed

TRIM and active garbage collection would work the way you described if a small amount of data was written to every cell, filling none of them

 

[Wirko]

That's not how SSD's work, they get marked as free space but don't get erased, specifically to avoid wear and tear as you described it
All that would do is mark the files as safe to be deleted, and have the files marked as junk - ready to be erased as the space is needed

TRIM and active garbage collection would work the way you described if a small amount of data was written to every cell, filling none of them

If a SSD is, like, 90% full then most blocks are mostly or completely full. What happens if you delete all files? The OS sends trim commands for all sectors, save some filesystem metadata. The SSD processes them, determines that most blocks now contain zero user data and marks them as "to be erased". That should happen immediately, unless the controller is unable to process a giant amount of trim commands in real time.

The rest is up to the garbage collection. I'd expect GC to begin its job shortly after that (seconds? minutes?) if the drive is idle, making it ready for further writing. Erasing while writing is avoided if possible because it drags down the performance by a lot.

 

[bug]

It would be nice if you could measure the energy needed to finish each of the real-world benchmarks.

 

[Mussels]

If a SSD is, like, 90% full then most blocks are mostly or completely full. What happens if you delete all files? The OS sends trim commands for all sectors, save some filesystem metadata. The SSD processes them, determines that most blocks now contain zero user data and marks them as "to be erased". That should happen immediately, unless the controller is unable to process a giant amount of trim commands in real time.

The rest is up to the garbage collection. I'd expect GC to begin its job shortly after that (seconds? minutes?) if the drive is idle, making it ready for further writing. Erasing while writing is avoided if possible because it drags down the performance by a lot.

They'd get marked as empty and nothing would be done.
What you're talking about only happens as the equivalent to defragmenting, allowing files to be compacted into less flash modules

You'd need to delete 90% of the files, somehow spreading the deleted files perfectly between all the modules

A complete erase, a quick format - just treats them all as empty and writes nothing

 

[Argyr]

Just received my Quarch for SSD power consumption measurements (https://quarch.com/products/hd-programmable-power-module/)

Any ideas, requests, suggestions what power-related tests I should add to my SSD reviews? Gimme a wish list, let me know what you'd love to see, don't worry if it can be done

Idle consumption and... prolonged mp3 playback? I always wanted to know how much power I'm using with my daily dose of 10-14 hours of mp3 playback

 

[W1zzard]

Less than 100 mb per hour for mp3, not sure if this will even be 1% over idle per hour

 

[Chris_Ramseyer]

For the most part, the only time people truly care about power consumption from an SSD is in a notebook environment. When I was a reviewer I worked out a deal with Lenovo to get four identical notebooks (Y700-17) with average battery life (for the time). That way I could test MobileMark on up to 4 systems at a time. That took care of when I had every capacity size available and didn't slow my reviews too much. You will want to get a battery sponsor because the batteries would loose too much capacity after 10 or so runs where it drains all the way down.

Best Storage For Notebook Battery Life

What type of storage is best for laptop battery life? Should I get a hard disk or solid state drive? For SSDs, NVMe or SATA? Let’s find out!

www.tomshardware.com

In my experience with the Quarch, 4-corner and idle power consumption wins don't always equal the best notebook battery life. Garbage collection, when it happens, and how long it takes plays a massive role. Some companies will use a GC method where the drive will run cleanup right after data writes and others will handle the cleanup later and let the queue build up before cleaning the flash.

 

[Mussels]

As one method to test power draw, use a USB NVME enclosure and a device that lets you measure the USB power draw
You'd get figures fast and accurately, assuming theres no key NVME power saving features that get disabled on USB

I've got this cheap USB 3.1 Gen 2 enclosure (10Gb/s)

And a quick search on amazon brought up pass-through measurment devices that aren't meant to impact speeds

Would allow external testing and possibly multiple at once, speeding things up?


The only flaw would be that at least with this model enclosure many NVME drives cant hit their maximum speeds - so you cant do true load testing for max power draw, but idle and constant loads are certainly doable

 

[W1zzard]

accurately

/doubt

maximum speeds

that's the problem

I have the Quarch ready to go, tests figured out, just need to find time to retest 30 drives

 

[mashie]

Total energy to get a bencmark/copy job completed. A faster drive I would expect to use more power but for a shorter duration than a slower drive.

I don't know if it is possible to set the read speed artificially, if so it would be interesting to see how power is or isn't linked to read and write speed. Say how much power is needed to sustain 100MB/s read for example?

 

[Wirko]

I don't know if it is possible to set the read speed artificially, if so it would be interesting to see how power is or isn't linked to read and write speed. Say how much power is needed to sustain 100MB/s read for example?

Copy data over gigabit ethernet, done.

But the OS, or any app, may read the data from SSD in many short bursts or few long bursts. That may affect the average consumption greatly, and the user has no control over that.

 

[Mussels]

/doubt


that's the problem

I have the Quarch ready to go, tests figured out, just need to find time to retest 30 drives

3am "cant sleep" thoughts are not always good ideas

 

[qubit]

You could do occasional long term tests on what it takes to kill an SSD, like Anandtech did a few years ago. I think it was them, anyway.

 

[1freedude]

/doubt


that's the problem

I have the Quarch ready to go, tests figured out, just need to find time to retest 30 drives

Too bad I did not see this earlier....

Efficiency, as in, heat to watts comparison. Does the temp rise linearly with power consumption.

 

[A Computer Guy]

Just throwing out some ideas.
Power consumption vs. Temperature of the drive
Power consumption vs. When SLC cache has been exhausted
Power consumption vs. Multiple I/O operations happening at the same time