Which chart?
Here are the charts from Anandtech and TechPowerUp. As you can see the Anandtech chart had the P31 Gold at only 3 mW (in line with the rated 4 mW) whereas TechPowerUp measured a much higher 0.365 W (365 mW).
Hm, interesting. And that is on the chart labeled ASPM enabled. How is Anandtech measuring though? TechPowerUp is using as separate programmable device that measures without the full system, just the SSD itself.
Anandtech says they use an “Quarch HD Programmable Power Module (PPM)” and attached this image (Source):
TechPowerUp says they use an “Quarch QTL1999 programmable power module” and attached this image (Source):
So they seem to be using the same testing equipment.
Yep. That would be because ASPM is just the PCIe power saving mechanism. That is sadly disabled on many desktop mainboards. But can be HEAVILY messed with by the BIOS in blacklisting certain devices from even using it or only using specific levels.
But it is independent of the power state transitions of the SSD itself. The 4mW are if the SSD is put to its lowest power level and ASPM will basically disable the PCIe link.
“ASPM” alone is almost worthless.
This would be the actual power states of an NVMe drive (Solidigm P44 Pro) as Linux receives them:
St Op Max Active Idle RL RT WL WT Ent_Lat Ex_Lat
0 + 7.50W - - 0 0 0 0 5 305
1 + 3.9000W - - 1 1 1 1 30 330
2 + 1.5000W - - 2 2 2 2 100 400
3 - 0.0500W - - 3 3 3 3 500 1500
4 - 0.0050W - - 4 4 4 4 1000 9000
You’d have to confirm each one of them and then also test with a specific trace of activity what the OS actually does.
For example Linux uses that data to come up with “Idle Times” that it needs to wait before going a power state down. And those seem to have currently only discrete values with large distances. Here is what I see Linux using with that SSD:
Autonomous Power State Transition Enable (APSTE): Enabled
Auto PST Entries .................
Entry[ 0]
.................
Idle Time Prior to Transition (ITPT): 100 ms
Idle Transition Power State (ITPS): 4
.................
Entry[ 1]
.................
Idle Time Prior to Transition (ITPT): 100 ms
Idle Transition Power State (ITPS): 4
.................
Entry[ 2]
.................
Idle Time Prior to Transition (ITPT): 100 ms
Idle Transition Power State (ITPS): 4
.................
Entry[ 3]
.................
Idle Time Prior to Transition (ITPT): 100 ms
Idle Transition Power State (ITPS): 4
And with the WD SN850 I had before my P44 Pro, the lowest level Idle time becomes 2000ms, with appearently no step in between. Because the WD SSDs are so extremely slow to transition.
And while I have not found that data in Windows, I could see the system taking a lot longer to go down in power consumption when idling compared to with the P44 Pro.
1 Like
And sadly they also say, testing the specific power modes would be to much work. So they just print the power states table as I have. But they use different systems and I believe one is a notebook that actually supports all the power saving.
Where do they say that?
Given that the graphs they release don’t perfectly align with the rated/reported power modes (they’re not far off but not aligned) they seem to be measuring something.
Note that the above tables reflect only the information provided by the drive to the OS. The power and latency numbers are often very conservative estimates, but they are what the OS uses to determine which idle states to use and how long to wait before dropping to a deeper idle state.
That is what I meant with testing the power modes. You could try to test the latencies of the transitions and actual power draw.
And then they only test full idle, in which every drive should use the lowest power state. But they do not test real-world situations like browsing the web on a notebook, which could potentially use a lot more of the middle power states. And where my FW 12th with WD SN850 was WAY worse than with the P44 Pro. But nothing I do is perfectly accurate in this regard, because I have nothing perfectly reproducible and I can only look at the whole system battery power draw.
Here for example the WD SN850’s table
St Op Max Active Idle RL RT WL WT Ent_Lat Ex_Lat
0 + 9.00W 9.00W - 0 0 0 0 0 0
1 + 4.10W 4.10W - 0 0 0 0 0 0
2 + 3.50W 3.50W - 0 0 0 0 0 0
3 - 0.0250W - - 3 3 3 3 5000 10000
4 - 0.0050W - - 4 4 4 4 3900 45700
And when closing out of all programs, the System with the SN850 took like 2mins to go down to 3. sth Watts. While with the P44 pro it takes it like 30 secs to hit < 3 W of whole-system power draw. Which I presume is caused by Windows accessing the drive far too often, such that the WD drive rarely if ever hits its lowest state. And also the lowest active state is way more power hungry. But you could not actually tell that from just the idle consumption numbers from Anandtech.
But then again, such intermittent idle scenarios also involve the OS and what it decides heavily. So nothing you could longterm ascribe to only the SSD itself.
1 Like
Oh wow, amazing information @Ray519, hadn’t done all that research myself, thanks! Amazing information on how that all works, in general.
Maybe I will stick with the P44 Pro, since it’s well proven with it’s quicker middling power states and such.
Yeah, I purchased based on the little information I could confirm from manufacturers (Solidigm and Samsung list at least some sleep state power draw numbers) and the few tests that looked at some part of the sleep modes. Not much out there.
So far I have used the Samsung 980 (non-Pro) in a weaker system, because it is also DRAM-less which should save additional power when mostly idling (same as other drives like the Solidigm P41). And the P44 Pro. Samsung 990 Pro was also on my short list in the high-end but still good idle power management class.
There may be a lot more power efficient SSDs out there, but like I said, takes someone reading out the power state tables, using it in an actual notebook and observing the power consumption closely to confirm. The numbers of the T500 operating look good. And its not their own controller, so even if Crucial might not care the developer of the controller might have…
And also, the OS has probably a lot of headroom to improve. The WD SSD seems terrible with its latencies, but on the other hand it lists only 4ms entry, 46ms exit instead of 1ms, 9ms and Linux changes its idle-time from 100ms to 2000ms. That seems like Linux is actually wasting a ton of power. (Also I own WD SN700 drives that have slightly different numbers (4ms, 40ms) but result in the same 2000ms idle time. So it is not just an unlucky coincident).
I bought an SK Hynix P41, apparently it consumed less power than my previous WD Black SN770