Intel Optane - Reliability

Hi,

This post is part discussion and part telling my experience with Intel Optane.

In the past, I opened this discussion here: Using a Raspberry PI CM4 with Intel OPtane M.2 SSD. It was not possible to boot an Raspberry Pi from an Optane drive. Today, it is possible.
I tested it on RPI CM4 and CM5 with M10 (16GB) and the 118GB (Intel P1600X) versions of Optane. I would like to use them for small docker setups with Pi Hole, FreshRSS and other tools with a small storage footprint.

In the past, I lost an M10 16GB drive due to overheating a component (at least I believe, that this was the issue, because the color of the component did change…).
The 64GB M10 ones seem to be quite reliable.
The 118GB P1600X is quite sensitive to heat, but none died so far.
I have one pre-M10 Optane drive, which I used for several years without any issues. These drives seem to be quite reliable, too.

The 900P 280GB and 905P 960GB U.2 versions are the best. They deliver great performance and so far none died (as expected, if a drive can survive multiple PB of rewriting data). These ones are used in PCs.

What is your experience with Optane? The technology might be dead, but you can still get them for quite cheap (especially in comparison to the current NAND flash prices).

2 Likes

I use m10’s as small linux boot drives all over the place. I’m actually moving my primary boot from 1X32GB optane to a 2x32 Optanes in a small spanned array for some added capacity. I haven’t had any issues with any of them. I also have a 16GB in my gaming PC accelerating a 4TB HDD using primocache and its working much.

2 Likes

I’ve never used optane and my environment doesn’t ask for storage expansion (I’m barely using what I already have). I’d like to have some, but I’m not going to try find a use for them.

If I had the hardware to push it, I’d use them for ZFS metadata special device (mirror vdev) and write-cache (another mirror vdev) for a spinning rust pool. But I don’t (most of my hardware is low-power and little connectivity / expansion).

You see, that’s where I diverged from the norm and kinda went schizo. I’m using a combination of net-boot root-on-nfs and local boot off of microsd with root-on-tmpfs (to basically stop all writes to the sd - everything happens in RAM).

Call me crazy, but this way you can just use NFS for your os (and data) and have cheap redundancy and backup (I have a central NAS with SATA SSD mirror and I can back it up with zfs-send over to a spinning rust pool and to my other spinning rust backup server). Well, back when I built this setup, sata was still cheaper than nvme and optane drives weren’t nearly as affordable.

But the thing is, given the small sizes, I don’t see optane being much of a use, other than accelerating zfs hdd pools. I think that’s the ideal place for it. If I want to have a bunch of OS rootfs, I’d go for a larger (and slower) sata ssd pool (like I already have - if I were to build something today, I’d go full nvme if I had the money lmao, RIP whoever’s building PCs or buying SBCs in 2026).

A 2 or 4TB mirror vdev can handle the OS rootfs for a huge number of hosts and containers and even store some data that needs faster access times (idk, some small DB workloads - we’re talking homelab / homeprod here). All the additional data can go on the larger spinning rust (or in my case, my 8TB SSD stripped-mirror pool on my other NAS-hypervisor combo, before it goes to my HDD mirror NAS).

Justifying my own stuff.

My idea behind the small 2TB pool was to have a “bare-minimum” infrastructure running on it (VPN router VMs, DNS, DHCP, network ad-blocker, monitoring, home portal etc.) and have the beefier PCs do heavy duty stuff (heavy virtualization, like lab setup and testing, more intensive tasks like in-home vid streaming etc.).

That was for power constraints (I don’t like paying huge electric bills) and I took it way too far (I admit, but I’m not going back now). And after that, I kinda delayed and delayed, then I moved and I got to a pretty dusty place and I haven’t build a filtered environment for my PCs, so I avoided powering them on for quite a while now (to not have them ingest the nasty kind of dust from here).

With the weather getting warmer in the northern hemisphere, I can resume working on that and build an enclosure with some air-conditioner vent filters (the HEPA stuff), so I know my PCs won’t need tons of maintenance when running them constantly. I’ll basically be building a “mini-rack” (for lack of a better term) with a custom size (since I can’t use the network racks with the 16 x 20" filters and have enough space for everything else). And I need to attach my fused switch boxes on the outside and route the power cables in through some grommets. Yeah, as you might’ve guessed, I’m taking things too far… again.

1 Like

Optane should theoretically be more reliable than other SSD technologies due to how it stores data in a more “permanent” way than NAND or even HDDs to a lesser degree; I think it would be the controller that would be the weak part of an Optane drive.

I’d argue against both those points. SOM is coming along very well as a direct evolution of 3D XPoint/Optane memory; performance and ease of manufacture have been greatly improved.
Optane memory is still extremely expensive compared to NAND, with PCIe 4.0 connected Optane going for >1.00USD per gigabyte on the used market while PCIe 4.0 connected NAND is going for ~0.15USD per gigabyte on the new market.

1 Like

I split an Intel 905P Optane 480gb PCIe card for each ZFS pool I have in two machines (two 905P cards).

40Gbps x 5 seconds = 200Gb = ~22-ish GiB

First partition I made 60GB, and assigned as ZFS Cache (ZIL) for the pool. Extra space for tuning 40Gbps data packets and increasing the 5s flush timeout if need be.

The other 420 GiB assigned as L2ARC cache because ZFS 2.0+ defaulted to persisting L2ARC-assigned drives between reboots.

One pool of 14x HDD mirror pairs and another pool of 10x HDD split into two vdevs.

I have to say, the performance is just insane on the pools. I peg the 48Gbps SAS2 backplane in both machines.

I’ve actually questioned why I have VNAND PCIe drives for VMs/apps when the pool is this fast (the answer is HDD spin down at 98% idle).

I have one 960GB PCIe 905P card I wanted to swap in for a larger L2ARC but it sits in a box because I’ve never noticed a slow down over the years and years.

Anyhoot, we love our Optanes.

2 Likes

@Sudo_Apt: When I still used Windows, I used primocache in combination with an Optane drive, too. The performance gain was insane. I used it to accelerate my 2TB game HDD.

@ThatGuyB: Your setup is definitely the better one. Currently, I do not have a NAS, so I can only install drives directly in the Raspberry Pis.

The PCIe 4.0 devices are still insanely expensive. My focus is on the PCIe 3.0 ones, because they are cheaper and my PCs do still only have PCIe 3.0 (the hardware is a bit older). I was able to get the M10 16GB 2242 ones for ~4$ a piece. I bought a few of them to use them in the Pis.

When I bought the none M10 devices, they were a bit cheaper than today. Amazon had them on sale. Today, you can find cheap ones on eBay, but you are correct, the price per GB is still higher than the one of normal SSDs. On the other hand, the Optane drives might be more reliable than some of the cheaper SSDs.

I have an Optane 900P 280GB model in my TrueNAS as an L2 ARC, for what it’s worth it’s got 100% available spare on it and has had 402TB written to it, so it’s still in great condition.

This is a PCIe model not a U.2 model.

69,418 power on hours, nice. Might need to screen grab that in 2 hours lol.

5 Likes

It’s been four hours. Where is that screen grab?

1 Like

My 118GB P1600X M.2 SSD does definitely need a heat sink. Even with direct airflow, it did not work properly without one. I used one with screws, because I do nit trust rubber bands… If the heat sink would fall, it could take some other components with it.

I’ve got a m.2 P4801X and with the bespoke heatsink, it’ll still run in the mid-40s idle.

1 Like