I put together a mining farm, mostly made up of used cards from Ebay. This is the story of how that worked out for me. Buying used cards has some major ups and downs, so prepare for a bumpy ride.
In January of 2017, I set about on a project to set up a small crypo farm with the goal of creating the fastest possible return on income- 10 weeks was the target. ZCash was big news, and farming it was going to be ultra profitable. The golden age of GPU mining had just begun, and I intended to ride that wave for as long as it was there to ride.
In truth, I was pretty broke. I was looking for a supplement to my income. I needed cards that didn’t require a huge outlay. I needed to know that the resale value of my cards would be preserved, so that if the ZCash bubble burst, I’d be able to recuperate most of my costs. I also needed volumes of mined coins that would actually make a difference to my bottom line when it came to paying my bills. That meant I needed £1,000/month (about $1300 at the time). Anything less than that wouldn’t scratch the surface.
What I decided I needed, was Ebay. At the time an RX480 would set you back around £240, and would hash at about 18 H/s (These were early Claymore miner days, the newer, faster mining software had yet to be released). But an R9 280X could be picked up on ebay for around £75, and would produce a near identical hash rate. Okay the 280X used 225w/card, where the RX480 was down at around 120w. Those are the power numbers I measured at the wall, not official TDPs. However the way I saw it an extra 100w was only about 2.4KWH/day, or in real terms at the rate of power I was paying it would take well over 3 years for the cost of electricity to even out the profit of the cards, and that suited me just fine. Of course, I was making one major error, but we’ll talk about that later.
So, what did i buy?
I ended up buying this rather extensive list of parts:
R9 280x x 14 (Paid on average: £75)
R9 290x x 2 (Paid on average £120)
7990 x 1 (Paid £130)
R9 295X2 x 3 (Paid on average £230)
RX480 8GB x 2 (Paid on average £240)
RX480 4GB x 3 (Paid on average £165)
Okay, so that was quite a lot of graphics cards- but if you look at the pricing, you’ll see that I worked hard to find competitive prices, and that most of the cards have gone up in value significantly at the time of writing. Those 295X2 cards are going on ebay at around £700 today.
The total spend on GPUs here is about £3100. The production capability goes up and down, but at the time it was around £2000/month. I spent at least another £2000 on all the other hardware needed, so I did hit my ROI target of 10 weeks.
For the sake of simplicity, I won’t talk too much about the other supporting hardware I bought. It was a mixture of old a new parts, but I will summarise one major point I learnt in this project- Don’t buy used high end PSUs from ebay. I bought 6 respectable high end PSUs from well known brands- and we’re talking Corsair, Antec etc, PSUs in excess of 1000w, and not a single one of them could support even 75% of the advertised wattage, and most of them died trying. New brand name PSUs work and keep working. Also don’t buy used hard drives, I had a 100% failure rate on the 6 used mechanical drives I bought, from various sources. Motherboards and CPUs seem fine used, so that’s the good news.
I bought metal shelves from Homebase to put them on, these set me about around £12 each, for shelves that could support up to 8 cards. It’s not pretty, but it works. Just don’t forget to put the motherboards on standoffs so they don’t short circuit.
Okay, so what happened, did the cards work?
3 of the 280X cards have failed so far. There have been a few fan failures, but mostly due to me sticking my fingers in the fans causing blades to snap. Of the failed cards, really only 1 of them worked and later ceased to work. The others showed up dead, and I was too busy to lazy to return them. By and large I’ve found that used cards actually work fine, as long as they show up working they should continue to work. Keep in mind also that GPUs going on ebay which are advertised as artifacting or having other visual defects generally work fine for mining, at least in my experience. Cards behaving badly normally just need a good clean.
There was however, one major fault in my plan, which I touched upon earlier. The higher power draw of the cards was not really an issue in terms of profitability. However, it really does limit the number of cards you can have running before you hit the power limit the wiring in a standard home or office can happily carry. Sharp readers will notice that I was pulling well over 6KW with the listed farm (and i didn’t mention the 12 additional RX480 cards I added later). I ended up closer to 8KW, which is over 30amps. The wiring in the room I was using was simply not designed for that kind of power draw. I ended up getting shut down by the landlord, who felt it posed a fire risk. I had to split the farm up over 3 different buildings to dilute the power draw.
Heat management was also an issue. In the winter it was not a problem, but once summer hit that room was edging towards 39*C, even with some major fans extracting hot air. 8KW is rather a lot of heat to be dumping into a small room. I don’t think mining with air conditioning is a very green thing to do, and certainly not good for your profitability. So make sure you budget for air management in a serious way, with in-line fans, ducting, and suitable access to outside to pull air in and dump it out to.
All in all, I think it would have made life easier if I’d just stuck to the newer low wattage cards. The actual hash rate and reliability of the used cards really never turned out to be the issue. However power draw and heat management really are the issues that stop a farm scaling. If you’re happy to stay small then great, go with the old cards, save some money and enjoy the speed with which you make your money back and turn a profit very quickly. If you want a farm that really scales up, modern low wattage cards really are the way.