"I've got a bunch of GPU mining rigs, surely I can do something useful with them like train my own AI or provide hashing power to astronomy or genome folding operations?"


Having acquired a large collection of GPUs that were previously mining ETH and ETC, as well as a nice stack of opterons with server RAM that were mining Monero I now find myself with a pretty big stack of silicon that isn't being used.

The GPUs seem to me to provide excellent support for graphics algorithms and the opterons I believe might be viable for CPU intensive tasks.

simply put: "I had a good run mining crypto", but I didn't sell my hardware...

in this nearly decade long process I learned quite a bit about Linux (since that's what every rig is running on)

My setup was pretty straightforward, I just used Ubuntu and manually installed each mining app, and then manually tuned each GPU. I simply started off with an eclectic mix of mining software, and eventually I pointed everything at NiceHash

once you tap into a serious silicon collection the tendency is to push it as far as you possibly can (it inevitably took over my entire garage)

that served me quite well until 2021. of course 2022 was rough, and in January of 2023 I have shut down the whole operation. (it's just not profitable to mine on GPUs, ETC and ETH aren't viable choices anymore. half of my setup is 4GB cards so those stopped being useful years ago, but I do still have 8 GB and a couple 12 GB cards)

I have a healthy background in full stack development (JavaScript, PHP, SQL). remoting into the machines constantly to babysit them for a decade gave me a healthy crash course in command line

at this point I'm very good at stripping down the hardware and giving "GPU baths" to the gear, rebuilding it, flashing the cards, cloning Linux drives, setting up auto run scripts, replacing fans, ect.

I sold off my ASICs, (let me know if that was a mistake, but I don't see how antminer s9s are anything more than paper weights at this point)

I suspect there's probably a pretty large segment of the mining community (just like me) that is just sitting on a ton of silicon that could be put to better use...

I don't know a whole lot about algorithms but I suspect GPUs are a good choice for training AI? (correct me if I'm wrong about that)

I'm also open to getting different GPUs, maybe a stack of Teslas? (those were horrible for mining, but I see them flooding into eBay on the cheap)

I'm thinking "gee, maybe I'd like to have my own AI? or perhaps these could be put to good use genome folding?"

at this point I've consumed a ton of information from Google (countless bookmarks closely related to the question I'm asking here)

but I find myself overwhelmed with the information, can anyone help me with the following questions:

first consider this eclectic mix of hardware: Nvidia GPUs, AMD GPUs (with DDR4 RAM) 4GB, 8GB and 12GB cards, couple Ryzen thread rippers, Ryzen 3800s, 3950s, a couple dual CPU Epyc's and a range of old Opteron processors (with ECC server ram), all of which were previously modded to maximize hash rates for cryptocurrency mining.

The CPU rigs were devoted to Monero, GPUs of course were various altcoins, ETH and ETC

  • 1: can I stick with Linux and train an AI?

  • 2: if so, what would be a good AI platform to start with?

I would be interested in training my own generative art AI or simply crunching data. I am not so much interested in language models (I don't want a chatGPT)

I'm not too concerned about the electricity cost, I have an excellent network and the correct electrical infrastructure. (it's not on an industrial scale, but it is a reasonably serious 'hobbyist' home mining operation)

I feel it's also critically important to point out that I've donated hash power to Folding@Home, and I thought that was a very interesting experience.

I'm also open to gearing these rigs to do genome folding (but I'm not sure where to start)

I'm very interested in training an AI for pattern recognition and image recognition. for example, feeding the AI 10,000 images and having it correctly identify patterns or similarities in those images.

if you were to give a checklist of suggestions to get started in training an AI (in my garage on old mining equipment) what would you suggest?

if this is an illogical pipe dream feel free to point out why (I've considered it might be better to just sell the hardware)

if I were to simplify this into an elevator pitch it would simply be: "I've got a colossal amount of mining equipment that's doing nothing, I think I could put it to good use in the AI sector what would you suggest if I wanted to train my own AI puppy?"

essentially I'm hoping for a series of links and a sobering reality check on electricity cost or hashpower (is it even feasible)

turning a profit is only of minor importance to me, I am more interested in the exploration and the experience of training an open source AI from the ground up.

  • I don't want a botnet and have no interest in any social media AI related shenanigans
  • I do think it would be nice if I could just point these rigs at an AI platform (maybe something in astronomy or genetics) and have them turn a small profit on electricity costs, or even just the opportunity to donate the hash power and contribute something meaningful with my 'seemingly' worthless silicon stack
  • I'm moderately interested in training the AI on stock market data (but I think this is a bit of a fantasy, but one I'm honest about)
  • I'm extremely interested in training the AI on pattern recognition and generative art
  • video rendering and 3D render farm opportunities are also of moderate interest (especially if AI is involved)
  • I would be very interested to see if it's possible to train my own AI to defend websites against bot spam, click fraud and RPA attacks (several of my web clients have indeed been attacked by bad actors using AI and RPA in SEM click fraud attacks)

side note: if you want to provide me with your own "TLDR", I want to point out I will be totally happy to read your entire dissertation. (I don't believe in TLDR because TLDR should be based on character limit)

  • $\begingroup$ My comment: TL;DR $\endgroup$
    – lpounng
    May 22 at 7:20
  • 1
    $\begingroup$ PCIe lanes matter a lot more. Crypto mining rigs usually run 1 PCIe lane to each card because that's the minimum. AI wants probably 8 $\endgroup$
    – user253751
    May 22 at 21:34

1 Answer 1


The answer is "it depends, but generally no".

See, AI operations is all about memory. That is useless for GPU mining. GPU mining cards are generally cost/performance optimized and that leads not to the top end with a lot of RAM, which makes them seriously suboptimal for anything other than the smallest models.

On top, TRAINING AI is a LOT more computational intensive than running it - hence companies sing thousands of GPU for the training part.

  • $\begingroup$ "sing" -> "using"? I think worth adding an important additional reason for "no" which is that NVidia refuse to license the CUDA driver software for gaming GPU cards in racks (apart from for crypto specifically) . . . the Tesla cards would be OK though. $\endgroup$ May 22 at 7:00

Not the answer you're looking for? Browse other questions tagged .