Login
You're viewing the mstdn.social public feed.
  • Alyxalyx@infosec.exchange
    Jul 4, 2026, 3:11 PM

    Edit: I got kinda overwhelmed with the amount of requests :neocat_laugh_sweat:
    Try to go through all of them, but please don't be sad if I miss your message :neocat_cry:

    But by now all cards got a new home.

    ---

    Anyone has a use for older GPUs (Nvidia 1050/1060/1070 AMD RX480/580)?

    We have a pile of them at work for free but nobody really knows what to do with them. Too old for your new gaming rig, but not old enough for e-waste :neocat_think:

    If anyone has a use for them, I would give some of them away for free, you would only need to cover shipping or pick them up in Zuid-Limburg, NL.
    Can also ship from DE if needed.

    Just send me a DM :neocat:

    💬 10🔄 46⭐ 16

Replies

  • Dave Wilburn :donor:DaveMWilburn@infosec.exchange
    Jul 4, 2026, 4:00 PM

    @alyx

    FWIW, I've heard of folks making decent use of these kinds of cards for running local inference using smaller GenAI models using quantization and/or layer swapping. At least if that's the sort of thing you're into. Local execution of small models addresses some, but certainly not all, of the ethical problems with this tech.

    Those cards are also perfectly good for local training and execution of more conventional deep learning for medium-sized data sets and models, like the kinds of ML that got shit done in the era before the AI grifters took over. I think the challenge there is you might not be able to use modern versions of CUDA or ROCm because of planned obsolescence from NVidia and AMD, but generic Vulkan support might be fine for a lot of work.

    💬 1🔄 0⭐ 0
  • kopper :colon_three:kopper@not-brain.d.on-t.work
    Jul 8, 2026, 8:37 PM
    @DaveMWilburn @alyx with an rx580 at least you're effectively gonna be stuck with vulkan if you want to do compute, which rules out like 80% of software unfortunately.

    the latest rocm that "works" is 5.7.1 which is bitrotting quite fast, and the only competent opencl you have is rusticl which is still unfinished (won't help with machine learning or blender cycles, but does the job with, idk, video editing if you're content with never upgrading past davinci resolve 19 and can tolerate flickering during effect previews)

    for local llm performance: expect ~128k context with ~90-130 tok/s pp, ~15-20 tok/s tg, on something like qwen3.6 35b-a3b at ~Q3 (with quite a bit offloaded to regular ram) (via llama.cpp, none of the other engines support vulkan afaik). which is probably not fun if you use language models seriously
    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Alyxalyx@infosec.exchange
    Jul 8, 2026, 8:24 PM

    @ppc64be Yea, had someone else from the US already.
    Even ignoring the shipping cost, there is also the issue that many of them are also still exceeding the "gift limit" the US customs have at the moment.

    I assume after shipping, import tax and customs handling fees, we would end up cheaper just buying one used on craigslist :neocat_sad:

    I'm really sorry :neocat_cry:

    💬 1🔄 0⭐ 2
  • 💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Amber :neodog_box:puppygirlhornypost2@transfem.social
    Jul 8, 2026, 8:18 PM

    @alyx@infosec.exchange my best advice would be to see if you have any schools around that offer classes for 3d modeling, game design or graphic design. A 1000 series would beat the piss out of the older cards (think GTX 710 and related). Call them up and ask if they are in need of a donation or know someone who is. Typically more funded academia passes hand me downs to other institutions. (at least, that was the case when I worked in K12 as a sysadmin.)

    💬 1🔄 0⭐ 0
  • kopper :colon_three:kopper@not-brain.d.on-t.work
    Jul 8, 2026, 8:34 PM
    @puppygirlhornypost2 @alyx the nvidia ones would prolly work assuming cuda hasn't bitrotted for them yet, but the amd ones won't do shit for cycles. they'd help for eevee but that's pretty much the same as regular 3d rendering with a kick so gpu compute doesn't really play into it
    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 1
  • 💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 1
  • Shrig 🐌Shrigglepuss@godforsaken.website
    Jul 8, 2026, 9:17 PM

    @alyx these would go great in an e-waste game console emulation machine, if anyone wanted some inspiration 👀

    💬 0🔄 0⭐ 1
  • Amyamy@cathode.church
    Jul 9, 2026, 8:39 PM

    @alyx ohhhhhh, I would love a 580, I am currently running a 1060 3GB, and wayland really does not like it..... (also, I am in nl, but wouldn't be able to pick it up, but can definitely cover shipping)

    💬 0🔄 0⭐ 0
  • Simon Zerafasimonzerafa@infosec.exchange
    Jul 9, 2026, 8:54 PM

    @alyx

    The 1050's often don't need external power so can be useful in older OEM systems which no upgradable PSU's

    The 580's would be useful in a Linux build for basic desktop graphics and very light gaming.

    Where are you based? 🙂

    💬 0🔄 0⭐ 0
  • Blass Rose :unverified:Blass_Rose@furry.engineer
    Jul 9, 2026, 9:42 PM

    @alyx
    *Sees these cards being called old and basically useless*
    *Looks over at my PC running a gtx 980ti still*
    (The large amount of VRAM for the time certainly has helped keep it relevant, but I know 6GB isn't impressive nowadays)

    💬 0🔄 0⭐ 0