Talk:Solid state drive/NVMe

Add the reason/link as to why discard is not recommended for NVMe SSDs?

The note merely states it shouldn't be done, but there is no reason or link to the source that supports the statement provided. Soukyuu (talk) 21:54, 12 February 2016 (UTC)

I would like to add that the explanation is unclear for another reason. The statement says,

Warning: Although continuous TRIM is an option (albeit not recommended) for SSDs, NVMe devices should not be issued discards.

This is confusing. Are we talking about continuous TRIM or TRIM tout court? This statement is conflating the two concepts. Moreover, this point isn't at all mentioned on the main Solid state drive wiki page. There, readers are first admonished to check whether their SSD supports TRIM/discard. Readers are then instructed on how to set up periodic TRIM. And finally, they're discouraged from using continuous TRIM.

This presents two problems. First, I feel readers should at least be warned about TRIM on NVMes on the main SSD wiki. Second - and more pertinent to this talking point - it's still unclear to me whether it's safe for NVMes to have periodic TRIM enabled or not.

Pound Hash (talk) 23:40, 22 December 2021 (UTC)

"GRUB does not support booting from /boot partitions that reside on NVMe drives, see FS#47447"

This is not true. I ran into an issue in this regard and fixed it somehow (I'm booting from /boot on NVMe at this moment). Can't remember how I fixed this. Will post back if I figure it out. Greyltc (talk) 16:28, 3 March 2016 (UTC)

Link to Intel-website does not work

The link http://downloadmirror.intel.com/23929/eng/Intel_Linux_NVMe_Driver_Reference_Guide_330602-002.pdf is dead. There is a page: http://www.intel.com/content/dam/support/us/en/documents/ssdc/data-center-ssds/Intel_Linux_NVMe_Guide_330602-002.pdf that is targeted towards developers. But I don't know it it's applicable to this article. The Wiki states that discard is not recommended on Nvme drives, but there is no information why. —This unsigned comment is by Dockland (talk) 09:17, 16 July 2016 (UTC). Please sign your posts with ~~~~!

NVME power saving patch v.2

The patchset which allows NVME devices to go to lower resting/idle power state has been introduced and on its way into mainline https://www.phoronix.com/scan.php?page=news_item&px=NVMe-Power-Savings-APST —This unsigned comment is by Frank604 (talk) 19:49, 3 September 2016 (UTC). Please sign your posts with ~~~~!

Sadly NONE of these solutions have worked for me on my (KLEVV CRAS C910 M.2 NVMe SSD 4TB) drive which uses a RTS5772DL Controller. It use to be able to come online if I boot to windows first and then soft reset back into Linux but it seems that no longer works either.

Now seeking to return it due to incompatible with Linux. It is SOLID under Windows btw except for when you boot to Linux and it fails to be ready and you soft reset back to Windows, it won't work in that situation. Need cold boot to windows and no issue, resetting from windows back to windows again also works.

All the issues seem to be with Linux in regards to this drive. I've spent over a month troubleshooting now and have decided to RMA the drive. Try and go for a different brand. Gerarderloper (talk) 04:48, 23 March 2025 (UTC)

On 4.13.11-1, I was getting ext4 fs errors that would cause the system to become unusable. I found a similar issue here and was able to successfully sidestep the issue by using the kernel parameter nvme_core.default_ps_max_latency_us=0 to disable power saving on NVMe. Should this become part of the Wiki? Maybelline (talk) 19:42, 7 November 2017 (UTC)

I got the same issue, one or two of my nvme drives(solidigm P44 Pro and samsung 970) would randomly disappear and adding the flag `nvme_core.default_ps_max_latency_us=0` seems to solve it Amfern (talk) 05:23, 13 August 2024 (UTC)

Makes sense - although it would be better to recommend setting nvme_core.default_ps_max_latency_us=5500 as the OP on that stackoverflow thread reported that this was enough to fix the problem. That still allows idle power consumption to be 0.07W instead of 3.10W, so it would be a pity if people unecessarily disabled that state. It looks like its only the deepest state that causes the problem. Perhaps also mention that it might be specific to Dell XPS 9550 (Is that the laptop you're using?) —This unsigned comment is by Hedgepigdaniel (talk) 21:54, 8 November 2017 (UTC). Please sign your posts with ~~~~!

No, I'm on an Intel NUC (6, I think?) with an Intel 500GB NVMe SSD. I will try changing the parameter to 5500 and see if it is still stable, and then update the Wiki according to my results. Maybelline (talk) 22:15, 8 November 2017 (UTC)

Using Arch Linux ISO release 2024.09.01, kernel v6.10.7, this issue still exist on Realtek RTS5765DL NVMe SSD Controller (DRAM-less). Good things that the kernel log on how to fix it 'Try "nvme_core.default_ps_max_latency_us=0 pcie_aspm=off pcie_port_pm=off" and report a bug'. So far, after couple of hours, there is no issue anymore. Ms (talk) 12:29, 12 September 2024 (UTC)

Not sure why the article referencing this marked this as out of date, disabling the power saving modes by putting nvme_core.default_ps_max_latency_us=0 in my GRUB parameters solved this for me. Before my SSD would randomly cease to function until rebooting even though it showed to be healthy after doing half a dozen tests in nvme-cli. My kernel is 6.17.8 and my device is a Western Digital Blue 4TB (WDS400T4B0E). --N9901 (talk) 04:17, 17 November 2025 (UTC)

I've got a Klevv C910 4TB NVMe drive and tried this in past without luck. I'll give it another go however.

Sometimes it can take me quite a while to get the drive to show under Linux (many reboots, cycling into windows helps), other times it just takes a couple tries, and sometimes it just works. The drive has always worked flawlessly under Windows11 so it is quite the bizarre issue, and sadly not many people have this brand of NVMe so difficult to resolve. Gerarderloper (talk) 05:04, 17 November 2025 (UTC)

Non-Operational Power State Permissive Mode (NOPPM) subsection

I added a subsection on NOPPM. I described it as a power saving feature because it has power in its name and disabling it might fix high power usage on laptops, but conversely desktop users experiencing low performance may want to ensure it is enabled. I have an Inland (Micro Center house brand) SSD based on the Phison E12 controller. I was debugging what I thought was a game load time regression with recent Proton versions, but it turned out to be slow reads of old (3+ years) data. It seems this controller won't refresh data that required ECC or read retries unless NOPPM is enabled. Jeffrey Bosboom (talk) 07:17, 20 January 2025 (UTC)