TerraMaster F4-423 review – A low-power 4-bay NAS tested with Unraid

30 Replies to “TerraMaster F4-423 review – A low-power 4-bay NAS tested with Unraid”

1. I’ve been looking for a box to run Proxmox Backup Server and this could be it since the main storage is going to be 3.5″ HDDs

   Karl, would you be willing to flash PBS to a USB drive and try booting it?

   Thanks

   Reply

   1. But you understand that PBS is just an added repo and a bunch of packages added to an otherwise normal Debian or Proxmox installation?

      Reply

   2. I will test and get back to you

      Reply

2. I’ve bought the TerraMaster F4-423 in May 2022 and tested it with 32GB RAM and TrueNAS Scale (I believe installed on the NVMe SSD), which ran nicely, but I returned it anyway.

   The main reason I didn’t find it suitable: The setting in the BIOS for which device to boot from was reset after every boot cycle. This meant I could not reboot the device remotely, because I had to use a connected keyboard/monitor to enter the BIOS boot menu to actualy boot TrueNAS Scale every time.

   I contacted TerraMaster support and they informed me the resetting of the BIOS boot device every restart was intended behaviour, because the F4-423 was not meant to run other operating systems.

   I decided to return the TerraMaster and buy a QNAP TS-364. It runs TrueNAS Scale just as nicely as the TerraMaster (with 1 disk less), but doesn’t reset its boot device setting every reboot.

   Did you also encounter this behaviour? Or has TerraMaster changed it with an update anyway?

   Reply

   1. I will test and reply

      Reply

      1. To clarify. Mine boots fine from usb without interaction and doesnt reset the bios every boot. I did find it aggressively boots from the Terramaster installation usb until removed.

         Reply

   2. I had the same problem with my Terramaster F2-423 and OpenMediaVault. For me it helped to reinstall grub with

      grub-install /dev/sda –force-extra-removable

      I tried TrueNas also but I didn’t have your issue. I’ve only had issues with the date/time settings.

      Reply

3. Works with 2x32GB RAM (64 total RAM)

   Reply

4. Bios version 2.21.1278

   Bio settings do stick with one exception.

   Booting is not as straight forward. Booting order does appear to be unchangeable. The order seems to be USB > NVME > HD. I was unable to boot from a HD with a NVME that has a bootable partition installed until I disabled the NVME. This was able to be overcome but like Job said it is an attended boot every time. Boot into the bios set the order and it will work for only that boot. It only one partition is present this should not be an issue.

   Proxmox installs fine.

   Reply

5. Conclusion :
   To avoid if you care about you data.

   Maybe good for a cheap secondary backup of the main Nas where to put replication snapshot.

   Reply

6. After reading the CPU data sheet I don’t how even is possible to immagine having 4 X 2.5 Gbe port and 2 nvme ,4 sata ,usb3.1 gen2
   🤷🏿‍♀️🤷🏿‍♀️🙄🙄
   Pcie 3.0 Max 8 of PCI Express Lanes
   No Lan integration
   2 SATA integrated

   And ow is possible that that say max supported ram is 32GB when CPU spec says max supported 16GB????

   This kind of products are basically legal scams, they give specs that are really misleading for people that have low tech understanding.
   if you aren’t afraid of putting 1000$ of hdd on a cheap case, go ahead.

   Reply

   1. > I don’t how even is possible to immagine having 4 X 2.5 Gbe port and 2 nvme ,4 sata ,usb3.1 gen2

      Just read the review above since Karl explained everything. The NVMe SSDs are attached with a single lane only and the two additional SATA ports come from an ancient ASM1061 (making also use of just a single PCIe lane) which is the same chip millions of motherboards with those Intel Atom thingies used already in the last decade two provide two more SATA ports.

      Reply

   2. > And ow is possible that that say max supported ram is 32GB when CPU spec says max supported 16GB????

      That’s because Karl missed 64GB are possible. At least since Gemini Lake the usual formula for those entry-level Intel SoCs is: ‘max capacity by specs’ * 4

      But you need to be careful since in the past DIMMs that were too fast (CAS latency too low) resulted in the CPU not booting: Hardkernel wiki. Maybe that’s why Intel plays safe here?

      Reply

   3. The 16 Gb on spec sheet threw me off two, but checked and it isn’t uncommon for Intel to underspec actual ram access.

      Reply

7. The review could not come at a better time. I was looking for new low power NAS for home usage and if you want more than 2 drives (and you will want more than 2), you will have to make sacrifices because these Intel based Celeron SoCs (no matter the version) can only support 2 native SATA ports. I know that these days a low power NAS is the one you want, especially with the price increase for electricity. I ended up repurposing my old desktop that has an ATX motherboard and kind of doubled my consumption, but I am at least happy it can be a long term solution.

   I highly doubt any newer chipsets will support more SATA drives natively, but I will switch to them when and if they will be available. Also I prefer to not spend 500 USD/EUR when I can reuse existing hardware that is anyway a lot more powerful than any of these NAS devices. Anything I would be able to find locally and used in mini-ITX form factor, but with enough power would cost me more or around the same as my old desktop (which is using X99 chipset and has 6 cores with 12 threads). I preferred to keep it at the expense of higher power bills, which were not that big to start with.

   Reply

   1. > these Intel based Celeron SoCs (no matter the version) can only support 2 native SATA ports

      Yep and the simple and obvious solution for more SATA ports is to sacrifice a single PCIe lane for an ASM1061 (as it’s done on the F4-423) or better an JMB582 in the meantime (Gen3 x1) to provide two more SATA ports. Or use an JMB585 or ASM1166 for 5/6 ports.

      BTW: the Intel SoC’s native SATA ports support SATA PMs so with a cheap JMB575 you can attach 5 SATA devices to each port.

      Reply

      1. On my oDroid H2+ it means I have to sacrifice nVME and I really love the fast boot times and the overall snappiness of the system when booting from nVME. I recently learned that there are still ITX motherboards for recent Intel chipsets that have 4 SATA ports, but sacrifice the PCI-E extensions, so adding more SATA drives would be difficult.

         Reply

         1. As written directly above: two (or even 5 or 6) additional SATA ports require one single PCIe lane if a SATA HBA should be added or just an inexpensive SATA PM connected to a SATA port to get 4 additional SATA ports. On a data dumpster performance never is an issue.

            Reply

         2. > snappiness of the system when booting from

            …SSD. It’s the storage media’s technology that makes SSDs ‘snappy’ since they’re great with random I/O unlike HDDs. The protocol (NVMe, SAS, SATA, USB3, even USB2) doesn’t matter that much.

            Unfortunately consumers are trained to look at the wrong metric (MB/s and not IOPS) so they make the wrong buying decisions all the time.

            Reply

   2. > more than 2 drives (and you will want more than 2)

      Nope, why would I? Care to elaborate?

      Reply

      1. I started with a single SATA 8 TB drive and it filled up sooner than I expected. I added a second drive with a bit higher capacity and that was also filled up. Getting higher capacity drives is not a solution because they are insanely expensive in my country the higher the capacity. The best bang for the money is around 12 TB (sometimes 16 TB drives prices are low as 14 TB, but very, very rarely). If I ever want to use RAID 1, then the number of drives doubles. Sometimes the performance on the low power NAS dropped without any heavy disk activity (a lot more often than on my repurposed desktop) that has 6 usable SATA ports as we speak (using nVME cuts 2 SATA ports). This was my experience and qith 3 drives I reached a compromise.

         Using several low power NAS devices is not really a solution, it makes sharing and management difficult.

         Reply

         1. > If I ever want to use RAID 1

            Wow, it’s 2023 and ‘RAID is not backup’ is still a thing… and especially RAID-1 IMO ist the most horrible waste of disks possible. Do you back up your data? If not your NAS is just a data dumpster since obviously the data has no value and you’re OK loosing it any time.

            With valuable data you need backup and that’s at least one other copy at another location with 120%-130% the original’s capacity to store versions.

            Reply

            1. That’s a point where I think we’ll always disagree 🙂

               I’m replacing my home RAID-5 file server with a RAID-1 because it’s sufficient and less burden to maintain (i.e. 2 SSD instead of 3, connected directly to H2+ without needing any adapter). There are even a few new mobos with two M2 adapters that allow you to connect two NVME devices, but they appeared just after I bought my H2+ board otherwise I would have gone for this.

               What I value here is indeed availability, since this machine is backed up every day. But when you say “how needs that at home?”, I would say “everyone”. Without RAID, you lose your whole data for as long as it takes to fully restore them, which can take from hours to days depending on the volume. That’s the type of thing that can ruin your experience or even a whole week-end. With RAID you just have to go to the local store, buy a replacement for the failed device, take one hour to replace it if a screwdriver is needed, and let it reconstruct in background while the service is operational again. This is why I *do* value RAID1 and RAID5. And at work it has saved us something like 5 times in 20 years, which is not bad at all, considering the alternatives that consist in telling employees “go back home till the end of the week, a disk died, data are currently being restored”!

               Where I agree with you is when some people consider RAID as backup, and sadly it’s indeed quite common. But I wouldn’t tell them “RAID is bad”, just “RAID is not backup”.

               On the other hand I’ve seen people choose and abuse RAID1 in DCs to speed up server deployments by replicating them, and while I was first horrified by the principle, when they asked me “I dare you to indicate me a faster method to deploy a whole rack”, I noted that sadly, they were right! because the deployment speed grows in 2^N without even needing network transfers, so you can deploy 32 servers with one install + 5 replication steps. This is a bit less interesting with large disks nowadays, but when it takes 10 min to replicate a 60GB partition at 100MB/s, in less than one hour you’ve indeed done 32 servers. And nowadays with SSDs it’s even better, with 500MB/s replication, it only takes 10 min to replicate everything, so most of the time is only spent in hardware manipulations.

            2. > Without RAID, you lose your whole data for as long as it takes to fully restore them

               Nope since I’m neither using anachronistic filesystems like ext4/UFS nor anachronistic tasks like ‘full restore’. It’s 2023 after all 🙂

               Already 25 years ago the amount of data stored in the graphics industry where I work(ed) in exceeded any limit that could be restored in a reasonable time window. Back then we used scripted sync attempts or commercial alternatives like PresStore Synchronize to keep another copy of data on another machine (primitive versioning included) to switch over to this other server in case disaster strikes. Of course combined with traditional RAID since spinning rust fails unpredictably.

               With filesystems developed in this century we can now have all of this as part of the filesystem’s inherent features. Both ZFS and btrfs allow for as many snapshots as you would like so you already have versioning as part of the filesystem as one integral building block of any backup. ‘rm -rf’ on a primitive RAID-1 and your data is gone, on ZFS/btrfs it’s a short laugh and reverting back to last snapshot. Same with ‘ransomware’, you simply laugh and don’t care.

               To fulfil the other essential requirement of real backup (at least one data copy physically separated from the original data) both filesystems have a send|receive functionality that unlike primitve rsync doesn’t need to scan source and destination all the time but simply sends the contents of last snapshot to 1-n other locations in the most efficient way imagineable.

               The fine-graned versioning is automagically inherited at the backup location so all snapshots exist there too. We usually consolidate snapshots on the productive storage to not go back further than 3 months but keep a lot more snapshots on the backup systems so even user requirements like ‘can we compare the versions of this doc in 2022 and 2021?’ can be fulfilled. Those versions need minimal additional disk space since snapshots only contain changed blocks so they’re the same as ‘differential backup’.

               All of the above are integral parts of the filesystems and all of this works flawlessly.

               Maybe time to overthink anachronistic backup strategies and rely on tech made in this century?

               Almost forgot: unlike mdraid-1 with a zMirror or btrfs’ raid-1 you get ‘data integrity’ detection/correction for free combined with a lot more sources for ‘data loss’ eliminated than primitive/anachronistic RAID-1 that is designed to only take care of a hard fail of a single HDD. A HDD slowly dying while corrupting data en masse will not even be detected!

            3. > Wow, it’s 2023 and ‘RAID is not backup’ is still a thing
               Yes, I know about the controversy and want to have some backup (but I don’t have lots of money to throw towards that). I have looked it up online and I will at least backup everything to a secondary drive that will only be used for this purpose only and will be spun down when not used (which will be most of the time).

            4. FWIW I’ve replaced my old backup machines with an Odroid-M1 which makes an excellent local backup machine. Mainline support, powerful but cool CPU (RK3568), gigabit, nvme, SATA, USB3, RTC, eMMC support etc and not very expensive. What’s nice about using such a machine with any form of storage is that it doesn’t inflict a performance constraint on your main NAS. You just nightly rsync to it, and even if it runs off an old spinning rust, it takes the time it takes but you have your backup to restart from in case the NAS dies, is stolen, or the disks die together. Also since the CPU supports crypto you can have inexpensive on-disk crypto storage.

            5. > FWIW I’ve replaced my old backup machines with an Odroid-M1
               I have an oDroid H2+ that I replaced because of the lack of native SATA ports. It could be used as a real and separate backup for the main NAS, but for now it will stay unused for now for lack of funding.
               I read more today and I will use SnapRAID for the current setup instead of RAID 1. oDroid H2+ will have to wait a little longer.

            6. > I have an oDroid H2+ that I replaced because of the lack of native SATA ports.

               Huh? N2+?

               H2+ is Gemini Lake refresh with the usual 2 native SATA ports that can be combined with SATA port multipliers to turn two SATA ports into ten.

            7. > Huh? N2+?
               My wording was wrong. I really meant H2+ and I meant that H2+ only has 2 native SATA ports and I needed more than 2, while still using nVME to boot OS. I think I will use it in the future as a secondary backup system. 2 large HDDs should be enough for the data I have. I have to draw a line somewhere, otherwise backup costs will increase exponentially with every storage increase. Money is and will always be an issue.

               I also think that the smaller case made ventilation a bit of a challenge during the hot season (I was using a Fractal Node 304).

8. Question:
   I’m running Google Coral on one of my NVME slot on my current NAS. If I buy this NAS would the google coral work on the NVME slot? In other words can you install anything aside from NVMES in the (2) NVME slots?

   I run Frigate and I need the Coral to do the AI for me.

   Thoughts?

   Reply

Leave a Reply

Connect with:

Your email address will not be published. Required fields are marked *

Comment *

Name *

Email *

Website