16GB Intel M10 Optane M.2 SSD $15.50 New?

[Glock24]

Does anyone here got them?
Does the support Sata?

I have one 16GB I removed from a new laptop some time ago. It's NVMe, not SATA, won't work as SATA.

What I've noticed is that after sitting a while it reports errors, requiring a full erase. This has happened twice. I just use it for tinkering with some SSF devices.

 

[rmt]

Keep in mind the M10 series has only two PCIe 3.0 lanes. That's not exactly fast but of course still beats SATA and makes for an OK ZFS SLOG or similar, especially if you have unused PCIe slots or unused M.2 board slots.

 

[alecz]

I bought 4 of these with the intent of using them as USB (2.0) boot drives and as ZIL via PCIe adapters in an old Dell T610 server (only USB 2, and plenty of PCIe).

Three of them died after trying to use them in some no-name RTL9210 based USB-A adapter. They checked out fine in the PCIE adapter, but failed on USB.

I managed to use the 4th one in a Orico NVMe USB-C adapter with cable connected to the server.
While Proxmox boots off the Optane, after a few minutes of activity, the OS hungs, reports errors from the drive and remounts it as read-only.

If I reconnect the drive to USB3 on the desktop, I can run Gnome Disk utility Benchmark without issues.
I might try Proxmox on the Optane in the NVMe on my desktop to determine if the issue is Optane or the USB adapter, but I think the latter.

So while these drives are excellent as ZIL, using them as USB Boot drives is not reliable.

P.S. I can run an OS of a usb2 powered 2.5" HDD for days with no issues, but not the Optane.

 

[Samir]

Would be interesting if some genuine ones bought domestically would have the same issues or not.

 

[alecz]

The ones I bought (were from China via eBay). They were sold as brand new, were sealed and the Windows NVME utility reported nearly zero usage (what ever little usage could have been attributed to factory testing).

I seriously think the issue is with the USB bridge and I attempt to confirm that by running proxmox from the PCIe adapter for a while.
So far badblocks show no issues with the last working Optane drive.

 

[nexox]

I bet it's a power saving issue, as I recall from the last time I listed all the features on mine, the M10 Optanes don't support as many low power modes and features as a typical desktop M.2 drive, seems like something that might not be well-tested in the USB adapter. You could try using an enterprise M.2 drive in the adapter, they also tend to have limited low power features implemented, though probably not exactly the same as the Optanes, so that test might be meaningless.

 

[BackupProphet]

These are supercheap and goes for like 5 USD ea. Excellent SLOG devices.

 

[UhClem]

... Excellent SLOG devices.

and

...
So while these drives are excellent as ZIL, ...

The specs for this (M10 16GB) are Seq-W 150 MB/s max & 4KB_IOPS-W 35k max.
Does that warrant the above quotes?

 

[BackupProphet]

and

The specs for this (M10 16GB) are Seq-W 150 MB/s max & 4KB_IOPS-W 35k max.
Does that warrant the above quotes?

Yes, because slog is about latency. You want fsync to happen as fast as possible.

 

[UhClem]

Yes, because slog is about latency. You want fsync to happen as fast as possible.

Won't sustained writing of data to a pool be severely constrained by the poor write performance (35K * 4KB = 140 MB/s maximum) of the M10 as the ZIL device?

 

[BackupProphet]

100MB/s sync writes is actually very good.
Compared to Samsung 990 PRO which does 5-10MB/s sync writes

 

[e97]

Grabbed 2 for a ZIL/SLOG mirror.

Will see how it goes for $10.

 

[alecz]

...

The specs for this (M10 16GB) are Seq-W 150 MB/s max & 4KB_IOPS-W 35k max.
Does that warrant the above quotes?

using a simple Python script, I measured 2545 fsyncs/s in my KINGSTON SA2000M81000G NVME and 9901 on the 16G Optane. So at least 4x fsync speed which is what ZIL needs.


If you use them over Gbit network, 150MB/s is sufficient, if you need more, you can always stripe the ZIL and get 300MB/s writes. That's what the 32G Optane version is anyway (hence the double write performance)
(with a stripped ZIL, you only risk losing the last few seconds of writes if you have a crash/powerfailure/hang AND the ZIL dies at the same time)

For posterity, I have ran Proxmox from the Optane drive in a PCIE adapter for more than a day and it is solid; absolutely no issues with this drive.
So, don't use them via USB adapters unless you are positive the adapters are of good quality.
This being said, it's probably not worth using them as boot drives (due to the USB adapter issues I encountered) because they take precious M2, PCIE space on the motherboard.

 

[AlexGee]

does anyone know fsync numbers for other Optane drives? Is it similar between the Optane models, or other models much faster

P1600X with 58GB, or 118GB for example (40/80$)

 

[nexox]

The p1600x is much faster (243k write IOPS for the 118GB): Intel® Optane™ SSD P1600X Series (118GB, M.2 80mm PCIe 3.0 x4, 3D XPoint™) Product Specifications

The other Optane drives have specs on Ark as well, but all the NAND SSDs have been scrubbed in the transfer to Solidigm.

 

[AlexGee]

So, I’m curious

IOPS are the criteria for sync write speed?

why are regular SSDs then so much slower, even though even higher IOPS (see example above with 990Pro)?


P.S. I recently ordered 4x P1600X 118GB, as I thought the M10 is too slow and 16GB are difficult to use in terms of size (as W11 boot drive, special vDev, etc…)

 

[mrpasc]

So, I’m curious

IOPS are the criteria for sync write speed?

why are regular SSDs then so much slower, even though even higher IOPS (see example above with 990Pro)?


P.S. I recently ordered 4x P1600X 118GB, as I thought the M10 is too slow and 16GB are difficult to use in terms of size (as W11 boot drive, special vDev, etc…)

IOPS for fsync at low queue depth for single thread at low latency and this sustained, not only for consumer benchmarks. An Optane has same specs regardless how much it’s filled or how long you write.
So do not trust CDI and co, test for yourself. Hammer a 990pro with looots of small writes, far beyond the SLC / DRAM cache. Compare this to the same workload to such a shiny P1600x.

more general: consumer SSDs specs are to read as „up to“ (under best conditions, like disk is empty and trimmed). Enterprise specs read as „at least guaranteed“.

 

[nexox]

the M10 is too slow and 16GB are difficult to use in terms of size (as W11 boot drive, special vDev, etc…)

I personally go for the 32GB M10, they're not as cheap as often but I have acquired a few for $10 or less over time, they make a fine boot drive for Linux systems.

 

[alecz]

Quick anecdotal update:
I created a Windows10 VM hosted on ZFS mirror (2 x 15K SAS drives) and ran my simple Python Fsync test:

• just the 2 x 15K SAS drives: 55 fsyncs / second (pretty good considering I was getting 33 on a laptop bare-metal Linux OS)
• With the 16G Optane SLOG on the ZFS mirror: 1000 fsyncs / second (huge acceleration, not quite like Optane alone, but more like SATA SSD)

So with a simple 16G Optane SLOG, you can have VMs hosted on HDD behave as if they were on SATA SSDs (when writing).
This way, I see the Optane drives as "VM Storage Accelerators" for VMs hosted on HDD.