• 0 Posts
  • 53 Comments
Joined 15 days ago
cake
Cake day: December 20th, 2025

help-circle





  • No that’s probably a bad idea.

    But if you were to get a gpu, the most recent and highest end ones are the ones that stay relevant the longest. People could still run shit on titans until nvidia dropped support from the latest driver package.

    You could try doing without a gpu, in which case you save a lot of money you would have spent on one but now you have to spend more money on a motherboard/cpu/memory in order to squeeze maximum performance out of the onboard gpu and when one of them starts to be too slow you’re fucked.

    If you do buy a gpu then you can reasonably expect only a 5% drop in frame rate when you use a very cheap 6-7 year old ddr4/pcie3/old cpu combo, which aren’t being hit quite as bad with ram pricing.

    Gpus have grown significantly in the last two decades from components that often shipped with a half height bracket in case you wanted to stick it in a sff pc to the main geometric limitation of case choice. Use the “length” filter field in the pc part picker website to not end up with an unfortunate situation.

    Amd or nvidia? People will say there are serious differences on linux, I’m not seeing it. The nvidia stuff tends to be performant longer but ymmv. The top end current generation amd cards can be had for under a grand, you’ll be lucky to get a 5090 for under two grand.

    People will disagree with me, and they’re wrong, but the best gaming experience overall is nvidia. It may not be worth 1k to you, but it’s reality.

    If I were feeling antsy and needed to pull the cost effective trigger on some parts I’d look at the area around me or eBay for a used professional workstation targeting pcie3/ddr4 and gpu mounting length and get a 9070. Once it’s up and running, mission accomplished, I’d still sock cash back and keep an eye out for an nvidia deal.









  • Uhh it stands for host bus adapter I think. It’s the word for a sas add on card. They make about a million different kinds that are all secretly the same card and just have different numbers of plugs, location of plugs, card dimensions or interfaces and support different protocol versions.

    An hba will let you plug up to any sas thingy and talk to it. If that thingy is an enclosure then all the drives in the enclosure will be directly addressable as if they were directly plugged into your hba inside your computer instead of connected to your computer by a wire.

    Sata is physically and electrically one way backwards compatible with sas, so a Sata drive can plug into a sas enclosure but the reverse isn’t true.




  • Some linux or windows and mergerfs plus snapraid running on some old computer lets you make use of many mismatched drives.

    Some linux or windows will give you a platform from which to enable file sharing. If you want a web management interface on top of it there’s a lot of those but it’s not strictly necessary.

    Mergerfs merges several filesystems into one big filesystem. It will blob directory matches together, so if two or more of your disks are windows c drives for example it may be worthwhile to make a unique root folder on each drive that contains everything else.

    It sounds complicated but actually it’s the simplest thing in the world.

    Snapraid makes parity snapshots when you tell it to. It needs at least one drive to use for parity that’s as big as the biggest data drive. It’s different than real raid or zfs because the parity you can use to recover from isn’t real time, it’s as old as the most recent snapshot.

    There are many benefits to that arrangement instead of zfs or real raid. If you want to know the trade offs I can elaborate.

    The benefit of what’s described in all of the above is that you can use anything to run it instead of needing a nas appliance, which in my experience are hot nasty dogshit until you spend as much money on it as you’d need to get a used 1u server and drive shelf and at that point just get the more reliable, capable device with very broad documentation and widely available parts and service.

    If you choose to use an old computer and just hook up all the drives, that’s great and old computers are easy to find and will work fine. The power use is truly negligible but if you were to get a smaller, ostensibly more efficient pc like a crappy little sff dell, you could slap an hba with external ports in it and attach that to some sas enclosure and use all your drives that way with (maybe?) less power draw.

    E: I made the same post twice. Age is a harsh lash under which to suffer.


  • You use mergerfs/snapraid for this. You need one extra drive as big as your biggest drive to do snapraid.

    Mergerfs mashes all the drives into one big filesystem, so if you don’t want file name collisions then put a unique root folder on each drive. It’s a pain if you’re serving up drives yanked directly from old pcs but it’s a blessing when you want to make maximum use out of each drives free space.

    Snapraid makes a parity snapshot when you tell it to. It needs as much space as the biggest device on your mergerfs. Its perfect when you don’t care if you lose a days work and don’t need bulletproof 100% uptime. If you’re like me and use secondhand drives exclusively, it offers the ability to do n parity which will let you recover from errors that span n disks.

    These two systems function independently of each other.

    Set all this up on some computer with the drives in it. I think both packages support windows but I’ve only used em under linux. There are a million tutorials on this.

    If you don’t have a case/psu/cables for the number of drives you need, it’s better and cheaper to find an upgrading gamer with an old one they can sell you than to get a good, functional usb enclosure. If you plan on making a hoarding habit out of this, a drive shelf and hba with external ports is an affordable solution.



  • Thanks for the informative and detailed answer! I’ve only ever installed and used arch for fun so the finer points of how pacman handles manually installed packages never came up.

    You said mostly safe, what kinds of issues can doing what you just described cause? You said pinning it through pacman would be an unsupported partial upgrade, even though that would give the package manager visibility on what you’re trying to do it would result in types of dependency resolution that aren’t supported or tested for I imagine?