Can a pi achieve the same iops? I'd be highly suspecious of any such claims.

I don't know why people love Pi-s so much. They filled a niche, once, years ago, and were quickly outcompeted not even a year and a half after.

Get some old i7 or Ryzen, get a big case, put 12-18 HDDs, spend a little extra on quality cooling solution if you have the server in your bedroom / living room, install modern Linux, tinker to your heart's content.

For homelabs, yer you can get something much better for much less.

For use cases where consistency and future support is key (education and industry) you really can't beat a Raspberry Pi. Their hardware and software support is top class. The first Raspberry Pi is still supported by the latest version of their OS over a decade later and it's even still being manufactured.

For all their products they commit to long term availability. For example, the Pi 5 will be in active production until at least January 2036 (assuming the company itself exists of course).

For anyone with a fleet of these, that's an amazing commitment. It means that when a piece of hardware breaks you can buy a band new but identical piece of hardware to replace it.

For most other companies you'd need to buy a different piece of hardware. Yes, the specs would be better, but now you have a fleet with mixed hardware which _you_ need to support and maintain going forwards.

Oh, I see. It's about fleets of easy-to-manage / predictable-to-support machines. That's valid, thanks for making me aware.

And indeed I was wondering about homelabs. RPis were never good there, not even when they got out for the first time. The form factor is what won over people back then. Feature- and speed-wise they were always mostly substandard. Not to mention Linux kernel support and driver issues (that might have been fixed since the last time I looked, admittedly).

And I agree on the fleet thing. Best if you can flash an SD card, drive to the spot in meatspace, pluck away the broken RPi, plug the new one in, wait for boot, test, drive away. Heard people doing that with RPis and others.

Because for $40 I have a system that runs at a decent speed.

For $300 I could get an ITX to run.

So for the cost of an ITX, I could run a dozen RPIs. Who wants to have a server running in their bedroom? Have you heard the noise those things make? Sorry, no.

I'm uncertain of why $40 vs $300 is even a point of debate on HN. The latter is a one-time investment and you likely can expand it a bit i.e. add a 2.5" or M.2 drive later.

What's the gain of running 12 RPi, exactly? Do you do research work requiring distributed low-cost computing?

I do distributed computing, and doing it at home for low costs without cloud spend helps…

Are virtual machines not an option for your use case? From the outside looking in they appear like they would be easier to manage and far less costly.

They are if the GPU can be attached. I avoid virtual machines in favor of container workloads from containerd for this reason. It’s easier to attach Mali GPU and do my work than it is to find cash in this economy for a dozen RTX’s.

A “server” doesn’t need to mean a pizza box with 15k rpm jet engine fans.

My server is repurposed desktop hardware in a desktop tower case and is nearly silent except for the subtle hard drive noises. The hardware cost next to nothing and is far faster and more capable than any pi (except the pio of course which wouldn’t be used anyway).

An ITX isn’t the competitor for a Pi. I’d suggest a USFF prebuilt. I use an HP Elitedesk and Dell and Lenovo each have similar tiny PCs. They’re nearly silent or completely silent, and half the size of a Mac Mini. Cost is about $150 for hardware that is more than enough for me, plus they can have 1-2 SSDs and a hard drive inside the case.

Clarification: They're about half the height of the OLD Mac Mini. Better comparison: They're the size of a typical hardcover book if you chopped it to be square.

You’re running the pi and drives in a plastic take away container off usb power for that price.

At the very least you want the case and psu. At which point the question is which cpu+motherboard+ram combo do you want in that case. The rpi is one of many such options and is actually quite expensive for the amount of cpu+ram you get for the price.

Does that $40 include everything to make the Pi work?

After looking at lots of small board options, I got a NUC for $110 to be the brains of my NAS.

> I don't know why people love Pi-s so much. They filled a niche, once, years ago, and were quickly outcompeted not even a year and a half after.

They still fill a niche for me, just not a server niche. The easy-to-access GPIO in a close-to-vanilla Linux system really doesn't have a competitor at its price point. For a fourth grade science project last winter, I had a pi 4 already (but it'd have been about $40 at my local microcenter if I hadn't). We were able to source a few $2 sensors off Amazon. I showed her how to look up the pinouts, figure out which GPIO pin to connect the dupont connectors to, and helped her write a python program to log the data from the sensors to a spreadsheet. She had fun with it, learned some stuff, and it really sparked her interest.

I don't think anyone has outcompeted them in accessibility for that kind of tinkering and learning. Or, if they have, they haven't caught my attention yet, and I've usually got my eyes open for that kind of thing.

Ah, education, right. I never had interest in the whole GPIO thing but I'll admit life has been pulling me in very different directions, hence this dropped off my radar. Thanks for the reminder.

Thing is, I was aiming at servers. I've read many HN comments where people adore a Pi for some reason that I just can't see; they have to install custom kernels, get Pi hats, do some extra cabling, 3D-print cases, mount small (or big) fans, and all that.

And don't get me wrong, I love tinkering myself but after reading people's experiences for a while I just thought to myself "Why all this trouble? Get a $250 - $400 mini PC off of Amazon / eBay / AliExpress and put a 2-4 TB NVMe SSD and you have something 20x more powerful and with 100x the storage space".

Again, I love me some tinkering. But nowadays I want to get something out of it in the end. Like the mini PC I bought that I want to dedicate only to a PiHole even if it's a 50x overkill for it. Might add some firewalling / VLAN management capabilities to it down the line.

So yep, for education RPi and Arduino (+ its derivatives) seem mostly unbeaten.

On a RPi I can control more aspects than I can a mini pc ITX board. I can boot straight to my program. I can write directly to frame buffers. I don’t need Linux. I don’t really need a kernel…

Here are some examples of where an RPi outshines a mini-PC (though one can still achieve the same results, just putting the box outside the box):

Coffee table Digital Touch map.

Weather Station powered by a solar panel and a LiPo battery.

ADSB receiver also powered by solar and a battery.

Arcade Cabinet that sits on a bar top with a bill reader.

Mini JukeBox at the local hacker space.

Sailing autopilot using NMEA2000 connectors.

Wearables.

Playing with high density distributed computing. (More than 5 machines)

Where the mini pc really shines is:

Storage. (NAS included)

Media PC (TV sold separately)

Gaming Console

Personal Cloud (docker + nfs + caddy + <insert personalized preferences>)

General Autopilot (sensors that need GPU support).

You have left over old PCs and don’t want to open your wallet…

Pretty cool, thank you. Those things have been not on my mind for a while, thanks for the reminder.

I was commenting in the context of why people choose them for servers but I recognize that I did not make that clear.

It's just an illusion. You're still running under https://en.wikipedia.org/wiki/ThreadX

Only for initial boot into Linux. But yes, technically it’s step0.

As I understand it, that is incorrect. It's more like a hypervisor still running in the background. That may have rolled back more and more with recent versions, but in principle the hypervisor can show shit like temperature, voltages and frequency in wrong ways, or delayed to the 'guest', whatever that may be. Actually that was the case, for some time, as some low-level tinkerers and/or overclockers discovered.

Similar to anything running in SMM (System Management Mode) on X86/AMD64 since the times of the 386SL. Be it BIOS, APM, ACPI, UEFI, or whatever.

I repeat: "It's just an illoooshn..." (There is now raw iron/silicon for consumers)

If you don’t have use for GPIO or some ISC^2 sensors and want to use it as a server then yes you should get something else.

Userspace-accessible GPIOs, I2C, SPI, PCM, and UART on a system that runs Linux. My employer uses them for a bunch of our hardware-in-the-loop test automation, with the GPIOs used for CAN, relays for switching various signals, vibration table control, etc. The USB gets used for SCPI device control (power supply, multimeter, etc.) and DuT connection. It's a lot cheaper to use a Pi for this than it is to use a small form factor x86 machine with a bunch of USB-<protocol> dongles.

Power consumption is a major draw (pun intended) to keep Pis and other SBCs of that kind of form factor employed.

Valid, thanks. But to what degree? The light bulb that runs 18h a day in the kitchen likely draws the same power that my mini form factor Optiplex 3060 does.

To me arguments like "2W vs 10W" are fairly meaningless.

I am much more concerned about data center power usages, especially in the age of LLMs.

Like that ancient German teacher I had that kept preaching we should stop using electric kettles because it's bad for the planet. While the 3 plants in her hometown amounted to ~83% of all power usage and ~92% or all pollution. Boy, was she unhappy when I did that research and pointed it out to her.

Pi-s / SBCs are I suppose very good for computing out there in the meatspace, where you might need a battery because sometimes power stops for 6 hours? Could be that.

Wait how did she suggest people heat water for tea/coffee instead? I've never heard an environmentalist attack electric kettles before.

She did not offer any alternatives. That was also a very funny element to her preaching. She saw a class of students and thought she can signal her virtues.

She was, shall we say, disappointed with the response.

Also this was some 15 years ago.

> Get some old i7 or Ryzen

power draw. running 24/7 it makes a huge difference in overall power usage, and by switching from a repurposed desktop mobo & cpu, to a dedicated low power saved me thousands in electricity costs every year.

that fact that in EU power isn't cheap... is also a main reason for keeping total power draw as low as possible.

Around 200-270MiB/s is what has been publicly benched. I’m sure there’s someone squeezing 300 out of one.

The PCIe bus in an RPI is Gen 2 so it’s not that fast. The point isn’t whether an RPI is a Synology device. The point is there are other ways of having a cheap NAS other than Synology.

Hell, a Beelink with an external USB 3.0 HDD rack would also do just fine.

500MB/s NVMe via the M.2 hat.

It doesn't even have to be a Pi though, just look at competing NAS solutions that have hit the market since Synology peaked in popularity.

Why am I spending more on a Synology versus something like a UGREEN NAS and just flashing a wide selection of NAS/home cloud operating systems on it? Synology's customer base certainly has the technical know-how to accomplish that.

Oh wow...I'm surprised at 500 MB/s NVMe.

I've got an RPi 4 with a Samsung 990 EVO Plus 1 TB NVME SSD in an external USB-C enclosure connected to one of the Pi's USB 3.0 ports, and get 280 MB/s.

I would have expected going to an RPi 4 with an NVME SSD not going through USB to do a lot more than just boost storage speed by 80%. I had been thinking of getting an RPi 5 and moving my RPi 4 stuff to the 5, freeing the 4 to replace the 3 that is current running Home Assistant, but for what I'm doing on the 4 I'm no longer sure the 5 would actually give much noticeable performance improvement. It may be better to simply get another 4 to replace the 3.

I guess this is a side note personally don’t think any of the Raspberry Pi hardware is worth it unless you are using the GPIO pins or any of those not-NAS not-PC type of functionality the Pi offers. I think for general compute it’s hard to make it make sense.

I think there are a whole lot of mini PC type of solutions that just make more overall sense.

Do you need it to?

Which SoC boards have ECC ram? ECC ram is essential for any reliable data storage system. Disks have built-in error correcting codes, and RAID can detect errors, but none of this helps if the data is corrupted in RAM before it ever reaches the disk.

The RPI CM5…

The specs claim "ECC" [0], but give no further details. ejolson on the Raspberry Pi forums [1] thinks it is on-die ECC, not traditional ECC, which would mean transfers between the RAM and the memory controller are not protected and there is no means of monitoring errors. Some discussion on Reddit [2] also suggests it's on-die ECC. If this is true, it's better than nothing but still not a replacement for a NAS with traditional ECC RAM.

[0] https://www.raspberrypi.com/products/compute-module-5/?varia...

[1] https://forums.raspberrypi.com/viewtopic.php?p=2296449#p2296...

[2] https://old.reddit.com/r/raspberry_pi/comments/1irryax/raspb...

The chip is the memory controller…

Yes it’s on-die. Yes it has error reporting. Don’t spread fud. There isn’t a dedicated chip because there doesn’t need to be.

Broadcom BCM2712

In that case I incorrectly thought (like the other forum posters) it was like DDR5 on-die ECC. What you describe is better than DD5 on-die ECC. Is this error reporting supported by Linux? Is there some way I can do fault injection (e.g. undervolting the RAM) to check it's working?

ECC is very helpful.

Having used both, I can't help but notice how NAS' routinely run just fine without it.

>NAS' routinely run just fine without it.

How do you verify your data to confirm that?

ZFS helps, and many people are okay with the risk of a cosmic ray causing a bit flip while data is in flight once in a blue moon.

I currently manage four NASes (two primary, two backup replicas). Only one has ECC RAM. And I'm okay with my setup.

ECC is great to have, but it is oversold by some as being absolutely required for all storage devices, IMO.

For truly important files (photos), I’ll take the slight added expense of ECC for a little more peace of mind that old photos aren’t being gradually degraded with every resilver or scrub.

Good point about ZFS. Having more than one copy helps too. ECC is great when possible.

Multiple backups.

How many files have you personally seen gone corrupt on non-ecc?

ECC originated first out of server grade servers. Self-hosting rarely hits that level of demand.

My first thought is the same way everyone's laptops and desktops and cellphones without ECC data do?

I'll share any more that come to mind.

Use a RAID5 and hope the write hole doesn't eat it all =(

Ubiquiti smelled the blood in the water and released a whole new NAS product line. They don’t run arbitrary apps but for basic storage on the network they look pretty solid.

I have a DS923+ and it's been great as a combo storage device and low-powered Docker host for homelab stuff. But if I had to replace it, I would break it apart into pure storage (like the Ubiquiti device) and a mini PC to run as a server.

Another way to put this is that Synology misjudged their customers' appetite for alternatives.

The ease of use of the Synology solution was always a plus of the product, but Synology misjudged the values and abilities of its core customer. They also misjudged the rapidly maturing market of competitors (e.g., why am I buying a Synology instead of UGREEN?)

Their core customer always had the ability to set up their own NAS in a more manual way, they just didn't really want to have to do that when an easier solution was available.

This isn't a situation like iCloud where the whole purpose of the product is to provide a service that 99% of the customer base doesn't know how to do on their own.

For a typical Synology customer, setting up their own TrueNAS box is something that probably only takes an hour including watching a YouTube setup tutorial. The person who is considering a Synology solution in the first place tends to be highly technical to begin with.

I can confirm that I bought a Synology NAS because I didn’t want to tinker with the backup system for my family’s data. And when I read about the drive requirements for a new Synology NAS I decided that tinkering might not be such a bad thing. They really screwed up.

Same. I like my Synology unit well enough but I see a trend toward less openness, toward greed (including removing capabilities from units they've already sold) and toward a decline of their business as a result of tanking customer goodwill. So they no longer seem like a reliable bet for the long term, which is what I'm looking for in a NAS.

That is exactly where I am. The value prop on synology has fallen off. Esp since they have let their kernel rot. There are tons of perf they are leaving on the table. The default external ports are usually 1g and most others have moved to at least 2.5g.

I just wanted something I just didnt have to mess with a lot. And could pop in an external USB drive here and there. Other solutions will fill that need just fine too. Just didnt really want to fiddle with DIY.

This is just another version of "why Dropbox when rsync" and equally silly.

What is silly about building your own? Explain? If rsync works for you, why would you buy Dropbox? “Why Lamborghini when Honda” is equally as silly yet I’ve seen them race head to head. Honda won.

I don't agree with the grandparent comment... I don't think it's silly.

But building your own doesn't scale to all the things. For everybody who wants to build their own X, the same person doesn't also want to build their Y and Z.

They will eventually need to buy some products. So there will generally always be a market for pre-packaged solutions.

For example: someone building an app may need network storage. They may not also want to block the building of the app on the building of the network storage.

In which case enjoy your Synology DRM and don’t complain that an RPi or ITX DIY build isn’t comparable…

There's nothing silly about building your own. What's silly is declaring a convenient, user-friendly product to be pointless because it's possible for a skilled person with a lot of free time to build their own.

If you want to build your own Dropbox with rsync, go wild, have fun, we'd all love to see what you come up with. But I don't have time for that. My family doesn't have the skills for that. Dropbox is great for us, and building our own is not a realistic alternative.

It really isn't, though.

1) there exist viable commercial competitors providing approximately equivalent functionality

2) the roll your own solution with, e.g., TrueNAS, also provides equivalent functionality and is about 90% as easy.

I say this as someone who owns and manages three Synology boxes and one more recent TrueNAS box. There was a time when Synology offered something quite better than the alternatives, but that time is no longer.

My newest one (192TB) I bought the hardware pre-assembled and tested from a VAR, installed TrueNAS, and was off to the races. It cost more than buying the individual components would have, but it had zero headache and was cheaper than buying the equivalent amount of storage from Synology.

I looked at all of those and they came nowhere near the convenience and software that Synology provides.

It's literally the "Why would you buy Dropbox when I can glue it together with rsync" level of ignorant comment, completely ignoring how behind most of those solutions like TrueNAS are in time cost.

I find that odd given that I literally run 3 Synologies and one TrueNAS - I've had the Synologies for over a decade and the TrueNAS for about three years now. They've all been great and do everything I want in a nas. Perhaps you are looking for more bells and whistles for something like DVR; I just do NFS + SMB and snapshots and remote backup.

For one rPIs are severely i/o limited still. May be fine with one ssd.

For two, if you like power adapters going into boxes out of which usb cables to go more external hard drives, a Pi may be fine. If you want one neat box to tuck somewhere and forget about it, they aren't.

But then people buy Intel "NUCs" where the power adapter is larger than the computer box...

And three, the latest Pis have started to require active cooling. Might as well go low power x86 then.

Exactly the route I took. I had an aging tower machine full of spinning disks running on an old LSI adapter that was doing hardware raid. They were out of space and I began to get nervous the LSI adapter could die and I would trouble replacing it. Decided JBOD for the future.

External drives were on sale, I bought several and setup with a RPI. Lots of headaches. It took effort to iron out all the USB and external disk issues. Had to work out alternative boot. Had power adapters fail for the RPI. Had to enhance cooling. etc. Kept running into popular Docker containers still not having aarch64 variants.

I finally replaced the RPI with a used Dell SFF. Kept the USB drives and it's been solid with similar power draw and just easier to deal with all around.

Though I am considering going back to a tower, shucking the drives (they're out of warranty) and going back to SATA.

I think most LSI adapters you can get a battery backup for. I've got one on mine, plus a spare battery sitting on a shelf somewhere. I admit when I put the system together for the first time I was a little hesitant to go with hardware RAID but it's worked out fine so far.

I reckon the issue is more in replacement than transient data loss: what are you going to do when you can't find a replacement controller card, or it only available at ludicrous prices?

With a proprietary on-disk format you can't exactly hook them up to any random controller and expect it to work: either you find a new one from the same controller family, or your data is gone.

Replacing your RAID controller is already major maintenance, so there's going to be downtime. I wouldn't be opposed to just wiping the drives and restoring from the latest backup. I routinely do this anyway, just to have assurance that my backups are working.

And a risk! I've had this on a premium machine put together specifically for that purpose and when the raid controller died something got upset to the point that even with a new raid controller we could not recover the array. No big deal, it was one of several backups, but still, I did not expect that to happen.

> But then people buy Intel "NUCs" where the power adapter is larger than the computer box...

You say that like it’s a mystery why people by then but NUCs are fantastic little PCs.

The power adapter is just hidden under the desk whereas the NUC is sat on the desk (or behind the monitor/TV).

It’s the same as with Mac Minis and Apple TV. And other devices of that ilk.

I've had mixed experiences with my NUC. It has what I think is a firmware bug that causes display output to fail if you connect a monitor after boot. Very annoying if it ever drops off the network for some reason.

There seems to be a Windows-only update tool available that might fix it, but that's rather inconvenient when it's used as a server running Linux! No update available as a standalone boot disk or via LVFS. So I haven't gotten it fixed yet because doing so involves getting a second SSD, taking my server offline to install Windows on it, just to run a firmware update.

Both the Mac Mini and the Apple TV use internal power supplies.

Ah yes, of course they do. Doh! Thanks for the correction

That’s the point. The tiny nuc sits on the desk while you hurt your feet on your power adapter beneath.

The Mac Mini sits on your desk and there’s nothing under it.

If you use a couple of magnetic disks, the pi is fast enough. The disks will be the bottleneck. There are sata cards that allow up to four magnetic disks, and where you power that card which in turn powers the pi. It's very doable.

It's of course more work to set up than synology, and if you want a neat box, you have to figure that out yourself

You'd be surprised. A single spinning rust drive can hit 200MBps for sequential access, so that's plenty to saturate its 1Gbps NIC.

However, in my experience with a Pi 4, the issue is encryption. The CPU simply isn't fast enough for 1Gbps of AES! Want to use HTTPS or SSH? You're capped at ~50Mbps by default, and can get it up to a few hundred Mbps by forcing the use of chacha20-poly1305. Want to add full-disk encryption to that? Forget it.

The Pi 5 is supposed to have hardware AES acceleration so it should be better, but I'm still finding forum posts of people seeing absolutely horrible performance. Probably fine to store the occasional holiday photo, but falls apart when you intend to regularly copy tens of gigabytes to/from it at once.

The Pi 5 is working well for me with encryption. I tried dding a cold file to /dev/null now and got

1293685061 bytes (1.3 GB, 1.2 GiB) copied, 5.14336 s, 252 MB/s

which is good enough for me on magnetic disks

It apparently hit 387MBps for a few hours while running the montly raid scrub. I run luks on top of mdraid though so the raid scrub doesn't have to decrypt anything.

scp to write to the encrypted disk seems to get me something in the 60 - 100MB/s range.

So long as the storage system is capable of serving a video stream without stuttering, that covers the 99% performance case for me. Anything beyond that is bulk transfers which are not time sensitive.

My point is there are alternatives, like you said.

The alternative to a Synology NAS isn't RPi. There are plenty of alternatives - QNAP, UGreen, a tower running TrueNAS - but a messy pile of overpriced unreliable SoCs attached to SATA hats isn't an alternative for a single device with multiple hard drive bays, consistent power and cooling, and easy management.

The alternative is anything not Synology that can do NAS with SATA SSD or NVMe storage. That’s it. Anything more than that is in a class of enterprise servers that deserve its own discussion over a simple DS1522+

This is nonsense. Both a horse and a pickup truck can be used to pull a wagon, but no one seriously considers one an alternative to the other.

What you are describing is a hobby item for enthusiasts who want to enjoy tinkering, setting something crazy up and constantly debugging it. That is very much NOT the market of people who buy Synology products.

Synology is not a JBOD RAID. It is an appliance that does many things (storage, services, web access, etc.) and can automatically keep itself up to date with no additional contact necessary. It can be hooked up to a UPS for resilience. If you have a problem there are support forums and online articles. It can be synced to other Synology devices at other sites. Etc., etc.

A pile of RPi is none of that.

you dont even Pi and sata shields, just buy a SOC that has direct M2 ports...

Storage should be a home appliance, not critical stuff to maintain and manage.

The ability to hot swap a drive when it needs replacement without a disruption to one's life is what a NAS is for.

I feel like hot swap is great if you work in a datacenter, but in order to be a useful benefit in a home setting, you have to have new, replacement hard drives sitting around on a shelf somewhere. My RAID alarm went off about a year ago warning me that a drive was failing, and I had to place an order and wait a week. Plus, the amount of time it took for the HW RAID controller to rebuild the new drive, I probably could have restored from a full backup.

I don't need my data offline when it doesn't have to be.

You don't need extra drives sitting around. When one fails, you buy one, Amazon can have it over in a day, or local shops. If it's not realistic for that, having one spare isn't a bad thing.

If you replace with a larger capacity drive, the existing raid only uses the same size to keep the raid.

Depending on the drives you are using, SMR technology can take much much longer to rebuild a raid than CMR.

Self-storage should be like a cloud - people need to rely on it like a cloud provider. Hot swap is a negligible cost over the 5-10 years you keep a NAS.

Hot swap chassis whether it's one you buy or a Synology/QNAP, etc is the way to go. Hot swap used to cost a ton, it's considerably come down market.

Storage is like a home appliance for me, just because I could build a stove doesn't mean I should. I've spent enough time swapping hard drives manually and powering off gear to know that I don't care for it if I don't have to anymore.

Ill assure you the amount of Linux bros that bought it was probably already small. Most buyers of preconfigured solutions are buying it because it's a preconfigured solutions with no need for a computer science degree.

RPIs have no ECC RAM. Without ECC RAM you can get bitrot in your RAID/ZFS much more easily.

This article is only saying that ZFS can mitigate disk data corruption caused by bad RAM, mainly through using checksums, not that it can completely prevent disk data corruption.

Also, it does not talk about the scenario where the in-RAM data being corrupted does not come with checksum. For example, data received from the network by the NFS/SMB server to be written to a file, before it gets passed to ZFS. This data is stored somewhere in RAM by the NFS/SMB server without any checksum before it gets passed on to ZFS. ZFS does not do any work here to detect or repair the corruption.

So, ZFS does not prevent on-disk data corruption caused by bad RAM, and only mitigates it. Using ECC RAM results in a huge relative reduction of such corruption, even though some people may consider the non-ECC probability to be already low enough.

Don't take my word, here's Matt Ahrens, a, ZFS developer. It's not required but a good idea.

"There's nothing special about ZFS that requires/encourages the use of ECC RAM more so than any other filesystem. If you use UFS, EXT, NTFS, btrfs, etc without ECC RAM, you are just as much at risk as if you used ZFS without ECC RAM. Actually, ZFS can mitigate this risk to some degree if you enable the unsupported ZFS_DEBUG_MODIFY flag(zfs_flags=0x10). This will checksum the data while at rest in memory, and verify it before writing to disk, thus reducing the window of vulnerability from a memory error.

    I would simply say: if you love your data, use ECC RAM. Additionally, use a filesystem that checksums your data, such as ZFS."

Pis are actually pretty terrible at running a NAS. Sure there are people who do it and create content about it (Jeff Geerling) and that's kind of the schtick - it's quirky and weird and has some sharp edges. Great for making content or going down rabbit holes, not so great for actually running a high availability system that just works with minimal fussing.

There are a ton of very capable x86 systems that are small and accomplish the task at great power and noise levels.