NVMe SSD slows down while laptop is on battery

[gwynbleidd997]

Hi, I own an Acer Aspire 5 2019 with a Ryzen 5 3500U and an Intel 660p 1TB NVMe SSD.
I noticed a while back, that the SSD performs worse when the laptop is running on battery namely that the sequential reads and writes are capped at ~400MB/s instead of the usual ~1600MB/s. (I obviously tested the drive while plugged in and got the higher numbers. Programs used are CrystalDiskMark and AS SSD Benchmark)

I tried Balanced power mode with the slider on Best performance, High performance mode, disabled power saving modes like ASPM in the advanced power settings...etc no dice.
The BIOS on these cheap Acer laptops is useless, there's nothing useful I can tweak.

I also have a (secondary) 860 EVO installed and the performance cap is not present with SATA SSDs it seems, only the main NVMe (Windows) drive is affected. The 860 EVO performs exactly the same regardless if the lappy is plugged in or not. Funny thing is that the SATA SSD is actually faster on battery than the NVMe one with 550+ MB/s seq.

If anyone has an idea on how to fix this, your help is greatly appreciated.

TL;DR: Intel 660p 1TB NVMe drive on a laptop capped at ~400MB/s seq. speeds. Power saving options disabled and everything.

 

[Steevo]

What about the power savings on PCIexpress?

 

[gwynbleidd997]

What about the power savings on PCIexpress?

ASPM is under PCIe in the advanced power options, so yes.

 

[TheLostSwede]

Have you tried the Intel SSD Toolbox?

Download Intel® Solid State Drive Toolbox

Intel® Solid State Drive Toolbox

downloadcenter.intel.com

 

[gwynbleidd997]

Have you tried the Intel SSD Toolbox?

Download Intel® Solid State Drive Toolbox

Intel® Solid State Drive Toolbox

downloadcenter.intel.com

I got it installed and the SSD is running the latest firmware. Though there aren't any options relating to power saving or anything of the sort in there.

 

[R0H1T]

Are you sure that's because of the power options/battery or something else? That's a QLC drive & once the SLC cache/buffer fills up the writes will slow down massively. Try to run a benchmark after a clean shutdown or restart with only the battery backup, use high performance power plan & then test it.

 

[gwynbleidd997]

Are you sure that's because of the power options/battery or something else? That's a QLC drive & once the SLC cache/buffer fills up the writes will slow down massively. Try to run a benchmark after a clean shutdown or restart with only the battery backup, use high performance power plan & then test it.

The issue is that both the read and write seq. performance is bad, not just the write. Also to be sure, I tested again on Balanced (Best Perf.) and High Performance modes, both after a system restart. I also cleared the SLC cache with Intel's Toolbox software before and after and the results are sadly the same
Take a look:

 

[R0H1T]

You could try & contact Intel's support over this? They'll usually be able to diagnose this quicker than our trial/error method. You can also do one thing, open hwinfo & then run the tests while on battery. Hwinfo will give you more detailed stats especially the polling at 2 seconds, or lower, will tell you whether the speeds are possibly dropping due to the SLC cache filling up or something else. I suggest doing a single run of writes & do all 4 tests to see if you're seeing a massive drop across the board.

 

[Mussels]

sounds like something the mobo in the laptop does to conserve energy, dropping the PCI-E lanes/gen type

 

[Caring1]

sounds like something the mobo in the laptop does to conserve energy, dropping the PCI-E lanes/gen type

Possibly a power saving feature of the CPU, low power states etc.

 

[gwynbleidd997]

You could try & contact Intel's support over this? They'll usually be able to diagnose this quicker than our trial/error method. You can also do one thing, open hwinfo & then run the tests while on battery. Hwinfo will give you more detailed stats especially the polling at 2 seconds, or lower, will tell you whether the speeds are possibly dropping due to the SLC cache filling up or something else. I suggest doing a single run of writes & do all 4 tests to see if you're seeing a massive drop across the board.

I just opened a support ticket. Will update this thread if I get an interesting response.
About HWiNFO, it seems like the drive doesn't start at normal speeds then quickly drops, as seen in the screenshot below. Looks like it never reaches more than 400MB/s to begin with.

sounds like something the mobo in the laptop does to conserve energy, dropping the PCI-E lanes/gen type

I sure hope not because there aren't any useful options in the BIOS I can tweak.

Possibly a power saving feature of the CPU, low power states etc.

According to AIDA64's Cache and Memory Benchmark, the RAM read and write speeds drop down to 8-9GB/s from ~32GB/s and latency goes from 115ns to 225ns or so.
Even then, does it affect drive performance that much?

 

[Assimilator]

The Aspire 5 is an ultrabook, which means it has extremely limited cooling capability. If that capacity is exceeded, it's quite likely that the unit could catch fire, which would be extremely bad. I would wager that as such, Acer has firmware, or perhaps even a physical hardware module, that exists to safeguard the system by drastically throttling components down when it detects a potential for overheat. You are never going to be able to override those controls - they exist to protect you, and to protect Acer from users like you.

If you have a problem with this, don't buy an ultrabook. Alternatively, accept that the form factor necessitates compromises like this. If you want maximum performance at all times, buy a desktop.

 

[gwynbleidd997]

The Aspire 5 is an ultrabook, which means it has extremely limited cooling capability. If that capacity is exceeded, it's quite likely that the unit could catch fire, which would be extremely bad. I would wager that as such, Acer has firmware, or perhaps even a physical hardware module, that exists to safeguard the system by drastically throttling components down when it detects a potential for overheat. You are never going to be able to override those controls - they exist to protect you, and to protect Acer from users like you.

If you have a problem with this, don't buy an ultrabook. Alternatively, accept that the form factor necessitates compromises like this. If you want maximum performance at all times, buy a desktop.

I understand your point and agree with you to a certain degree. But I think it is blatantly obvious that the issue here was never heat management because if it was the case, then the performance should also be capped when plugged in as well if it's running hot but as I stated earlier, it's not. Whether the laptop's been running for a while, or just up from a cold boot, the performance issue I'm having depends on it being plugged in or not, not on the temperature readings.
If anything, this probably has something to do with preserving battery life, certainly not the reduction of heat.
I'm not claiming to know better than the engineers who built the laptop, they certainly had their reasons for implementing this and I fully respect that, but as a user I would also like it to be possible to take responsibility in tweaking the configuration to make it run the way I want it to and not the way Acer decides is best. Their solution may fit the general public and that's perfectly fine, but some users may not agree with them or any other manufacturer on some of their decisions.
If I bought a laptop that has an NVMe drive that according to Intel, should run at ~1800MB/s seq. read and write, then it should perform as advertised. If not, then I believe Acer should provide their customers the option to make it run at its advertised speeds at all times, even if that means less battery life and so on. Giving the option to people and making the benefits/disadvantages clear to them, is a superior way of handling different needs than straight up locking the thing and forcing everyone to deal with it.
If an NVMe drive was too hot or consumes too much power in such an ultrabook, then why should I pay for it in the first place? wouldn't it make more sense if they scrapped NVMe and went with M.2 SATA instead? If it's gonna run at 400MB/s then I'd rather pay less for SATA SSD for the same performance (actually better perf.), then pay more and get bottlenecked whenever I use my laptop as a laptop.
Just my 2 cents.

 

[John Naylor]

While yes certainly, a notebook has to protect it's components from heat ... but I don't see as the SSD gets significantly hotter when on battery as opposed to being plugged in ... This seems to me to be an overzealous power saving feature. But here's the thing ... does any part of you everyday usage involve "running benchmarks". Since the 1st SSDs came out, we have been building systems with both SSDs and HDs / SSHDs. In each instance we make a partition on the HD / SSHD which includes a backup copy of the OS and you can boot from either from the BIOS.

In the office (Engineering where CAD is main program) and at home, I like to 'switch it up" w/o user knowledge and see if anyone notices. Unless "tipped off" they don't. So there's two questions here:

1. Is application / gaming performance impacted in any noticeable way ?
2. Is application / gaming performance impacted in any significant way ?

Those are 2 very different questions

These are the boot times:

SSD = 15.6 seconds
SSHD = 16.5 seconds
HD = 21.2 seconds (note we haven't actually installed a HD in 8 years, using SSD / SSHDs pairings exclusively) but have 1 for testing

Now certainly ... you could notice the difference between 16 is and 21 ... but only if you don't multitask. Let's say I arrive in the office in the AM after having rec'd and printed an email from a client about requested floor plan changes the night before. So if I had the box boot off the SSHD, I'd take 0.9 seconds longer to get into the OS, AutoCAD takes the exact same time to load, not sure why, license checking I imagine, and the file loading is going to be determined by network speed grabbing the file off the server. But lets say it took 5 seconds longer for arguments sake. I'm reading the email while the file loads, after that I am on the phone with the client explaining why he can't have 2 of his 9 changes.

My son built a box 2 weeks ago, and he was exclaiming how much faster his game loads with his new NVMe as opposed to his old HD ... but, most of his library doesn't fit on the SSD ... he tends to play a new game on the SSD and them "moves it" to the SSHD and you won't really notice if not paying attention. But on the other hand ... here's his routine. When he walks in from work, he turns on PC and loads the game ... then it's load discord, load his game related web sites, take headphones off charger, plug the USB transmitter for the headphones in and he's ready to play say a minutes later. Whether th gamr loaded in 50, 40, 30 or 10 seconds doesn't really matter.... and that's if he doesn't go downstairs to grab a drink / snack whatever.
but it's worth gauging whether it's worth the T & E trying to get that faster speed if, in the end, you are not going to actually benefit from it in any real way.

The best example of this is office suite benchmarks that include a script containing 100s of operations ... how significant is it if one component completes the script 45 ms faster when the reaction time for the intervening keystrokes / mouse movements exceeds 1 second ? I suspect the the power saving features associated with the CPU / GPU / MoBo will render the SSDs throttling insignificant

So while you are waiting for your answer from TS, I would use the time to see what the actual impact on what you are doing is. If a solution involves some unsupported action for example, is that worth it for saving a second of boot time ? If you could get the SSD to go pedal to the metal ... will it have any benefit due to CPU / Bus and GPU power throttling ?

 

[gwynbleidd997]

While yes certainly, a notebook has to protect it's components from heat ... but I don't see as the SSD gets significantly hotter when on battery as opposed to being plugged in ... This seems to me to be an overzealous power saving feature. But here's the thing ... does any part of you everyday usage involve "running benchmarks". Since the 1st SSDs came out, we have been building systems with both SSDs and HDs / SSHDs. In each instance we make a partition on the HD / SSHD which includes a backup copy of the OS and you can boot from either from the BIOS.

In the office (Engineering where CAD is main program) and at home, I like to 'switch it up" w/o user knowledge and see if anyone notices. Unless "tipped off" they don't. So there's two questions here:

1. Is application / gaming performance impacted in any noticeable way ?
2. Is application / gaming performance impacted in any significant way ?

Those are 2 very different questions

These are the boot times:

SSD = 15.6 seconds
SSHD = 16.5 seconds
HD = 21.2 seconds (note we haven't actually installed a HD in 8 years, using SSD / SSHDs pairings exclusively) but have 1 for testing

Now certainly ... you could notice the difference between 16 is and 21 ... but only if you don't multitask. Let's say I arrive in the office in the AM after having rec'd and printed an email from a client about requested floor plan changes the night before. So if I had the box boot off the SSHD, I'd take 0.9 seconds longer to get into the OS, AutoCAD takes the exact same time to load, not sure why, license checking I imagine, and the file loading is going to be determined by network speed grabbing the file off the server. But lets say it took 5 seconds longer for arguments sake. I'm reading the email while the file loads, after that I am on the phone with the client explaining why he can't have 2 of his 9 changes.

My son built a box 2 weeks ago, and he was exclaiming how much faster his game loads with his new NVMe as opposed to his old HD ... but, most of his library doesn't fit on the SSD ... he tends to play a new game on the SSD and them "moves it" to the SSHD and you won't really notice if not paying attention. But on the other hand ... here's his routine. When he walks in from work, he turns on PC and loads the game ... then it's load discord, load his game related web sites, take headphones off charger, plug the USB transmitter for the headphones in and he's ready to play say a minutes later. Whether th gamr loaded in 50, 40, 30 or 10 seconds doesn't really matter.... and that's if he doesn't go downstairs to grab a drink / snack whatever.
but it's worth gauging whether it's worth the T & E trying to get that faster speed if, in the end, you are not going to actually benefit from it in any real way.

The best example of this is office suite benchmarks that include a script containing 100s of operations ... how significant is it if one component completes the script 45 ms faster when the reaction time for the intervening keystrokes / mouse movements exceeds 1 second ? I suspect the the power saving features associated with the CPU / GPU / MoBo will render the SSDs throttling insignificant

So while you are waiting for your answer from TS, I would use the time to see what the actual impact on what you are doing is. If a solution involves some unsupported action for example, is that worth it for saving a second of boot time ? If you could get the SSD to go pedal to the metal ... will it have any benefit due to CPU / Bus and GPU power throttling ?

Yes, I concur. This doesn't look like a thermal issue at all.
And to your main point, the answer is no and I agree with you wholeheartedly. If I could set up a blind test and compare even a 970 EVO (3000+MB/s seq) vs the 660p (1800MB/s seq) both running at full speed in day to day tasks, the probability of me actually noticing a difference is close to 0. Same with my 660p running at full speed and at 400MB/s seq.
BUT that's not the point I'm making here. While yes the seq. speeds are probably not very important in day to day use, if a manufacturer lists a certain spec, I expect the device to work within that advertised spec, regardless of how much of a difference it makes.
If I buy a smartphone with an advertised 40MP camera, and when actually using it, it can only do 16MP on battery, even though the MP difference wouldn't make that big of a difference anyway, I would feel cheated on and I hope you would too because such behavior is unacceptable. If the phone for whatever reason cannot take 40MP shots on battery, then don't advertise that and use a smaller sensor that the phone can handle at full res at all times. Same thing here, if Acer wants to use an NVMe drive advertised to run at ~1800MB/s, they shouldn't bottleneck its performance to only 400MB/s when on battery i.e. when using the laptop as it was intended to be used, as a portable device, to conserve power.
And even if they decide to do that, at least let people have the choice to disable the power saving features and let the drive run unbottlenecked if they wanted to. Disabling those power saving options should have done the trick but here I am.

 

[Assimilator]

Another possibility is that one of the sensors in your laptop is disconnected or is reporting erroneous values to the platform monitor, causing it to throttle incorrectly. I know other hardware has this issue - for example certain Macbooks throttle down to their minimum CPU speed if they can't detect the backlight (not even joking). So assuming Asus doesn't get back to you with some helpful information (which I doubt they will) you might consider trying a repair or RMA.

 

[mgutt]

I have the same problem but with the Dell Inspiron 5405, 14" and Ryzen 4300U and I think its a limitation from the BIOS which can't be changed.

With power supply:

Without:

Usually this should not save energy, because as later the work is done, the later all other components reach their sleep states. But maybe it helps the CPU/NVMe to stay in the most efficient frequency range?!

 

[gwynbleidd997]

I have the same problem but with the Dell Inspiron 5405, 14" and Ryzen 4300U and I think its a limitation from the BIOS which can't be changed.

With power supply:
View attachment 184777

Without:
View attachment 184778
View attachment 184779


Usually this should not save energy, because as later the work is done, the later all other components reach their sleep states. But maybe it helps the CPU/NVMe to stay in the most efficient frequency range?!

One thing I did, that I forgot to update this post with is install Intel's NVME drivers manually, that fixed the issue (to some extent).
Now I get 800MB/s on battery instead of half of that which I'm happy with. At least it's faster than my SATA SSD and not gimped to perform even lower than it.
I don't know if installing the Samsung NVME drivers will do anything in your case, wouldn't hurt to try I guess but to be fair, 1.8GB/1.6GB read/write is more than enough, at least it's not 400MB/s.
Es soll ausreichend sein, mein Deutschsprachiger-Freund

 

[mgutt]

I don't know if installing the Samsung NVME drivers will do anything in your case

Sadly not. Already tested