What idle hardware actually pays: three machines, measured

There's a specific kind of afternoon I keep having: I look at a machine that is already running, already powered, already paid for, and I think — that thing is wasting money by doing nothing. So I went and checked three of mine. A rented VPS, my desktop with a graphics card, and an original Raspberry Pi that has been humming in a corner for years. I expected a ranking. What I got was one dead end, one wash, and one small boring win.

The box I never measured, on purpose

The VPS was the obvious candidate. Four vCPU, AMD EPYC, sitting at 0.2 load — it spends most of its life doing literally nothing. Before I ran anything on it I read the paperwork, and it turned out there was nothing to measure. Contabo's FAQ for the VPS line states that proof-of-work mining is not allowed. Oracle's free tier goes further and names crypto mining as a violation in its Acceptable Use Policy. Two providers, two different contracts, same answer: renting someone else's CPU does not include the right to point it at a chain.

That matters more than it sounds. A lot of the "free tier arbitrage" writing floating around quietly assumes you may do this. I did the arithmetic anyway, just to know how much money I was turning down. Monero on RandomX at four vCPU lands somewhere around 3.5 to 5 kH/s. Against a network hashrate of 6.3 GH/s and a daily emission of 432 XMR at a price of 536 EUR, that's 0.13 to 0.18 EUR per day — 4 to 5.50 EUR a month gross, before anything else. The VPS itself rents for 5 to 9 EUR a month.

So the ban cost me nothing. The rent alone would have eaten the gross yield, and then the pool would have taken its cut on top. I closed the tab and felt slightly better about having read the terms first for once.

The card that pays for its own electricity, roughly

My desktop has an RTX 5060 Ti with 16 GB, a power limit of 180 W, and an idle draw of 25 W. Mining the best realistic coin — minerstat pointed at XTM, Tari — comes out to about 0.99 USD per day gross. Some lists put Pearl, PRL, at the top; it's a new microcoin and the numbers on it are not usable for anything, so I ignored it.

The interesting part isn't the gross. It's what the socket does to it. German household electricity averaged 37 ct/kWh in 2026 per BDEW, and running the arithmetic with that: at 90 W the whole thing nets out around +2.50 EUR a month, at 150 W it's about -13.50 EUR. If the card actually sits at its 180 W limit while hashing, it's worse than that. The gap between "profitable hobby" and "quietly burning money" is entirely a question of which wattage you talk yourself into believing.

Renting the card out instead is the other option, and it's the one I expected to be a scam. Salad publishes a measured average of 0.045 USD per hour for this exact card, and that figure already includes the hours where nobody rents it. Multiply out: roughly 29 EUR a month gross. At 120 W the electricity comes to about 32 EUR. Net is approximately nothing. And the card still wants 180 W when it's actually working, not when it's idling between jobs.

I want to be honest about what that means, because "approximately nothing" is easy to dress up. Renting a consumer GPU in Germany doesn't beat the wall socket. It ties it. What you get for the tie is not being the guy refreshing a miner dashboard to watch 0.99 USD a day accumulate. For me that was worth it. It is not income.

The machine that actually paid

The Raspberry Pi 1 is the one that surprised me. armv6, 427 MB of RAM, permanently on because it does other jobs — a rule of thumb here is that memory and compute are irrelevant, the idle power was already being spent. I went looking at six bandwidth-sharing apps, and the field narrowed itself fast. Only Pawns.app ships an arm/v6 binary; I verified that by starting it and watching uname -m come back armv6l on the same board. Honeygain wants armv7 or newer. EarnApp wants a 64-bit system. Grass doesn't do ARM at all. Traffmonetizer targets arm32v7, Repocket goes amd64 or arm64. On this specific board, the choice was made for me.

The second constraint is the one people miss: datacenter IPs are blocked across all of them. You need a residential line. That's the actual asset being sold here — not your CPU, not your disk, your home connection with a real ISP address behind it.

The terms are 0.20 USD per GB, with a 5 USD minimum payout. An independent 30-day test logged 4.87 USD against roughly 18 GB of traffic, which is the number that made me stop looking at the marketing page and start trusting the mechanism. There's a referral bonus too, 1 USD to each side per payout, capped at 3 USD total. Realistic order of magnitude per connection: 3 to 5 USD a month.

What the three machines actually added up to

The VPS paid nothing because I wasn't allowed to use it for this, and the arithmetic says it wouldn't have paid anyway. The graphics card nets out around zero whether I mine on it or rent it out — the German electricity price and the gross yield are too close together for the difference to survive contact with a real watt meter. The Pi was the only one of the three where the arithmetic worked at all - the thing being sold is the IP address and the hardware was already running for free - and even there I never reached a payout, for a reason that has nothing to do with the money.

That's the whole pattern, and it's less exciting than the pitch decks imply. Hardware doesn't get paid for being idle. It gets paid for owning something scarce — a residential address, in this case — and the machines that already run around the clock are the only ones where the economics don't have to fight the power bill.

So did it actually run? No, and the wall is worth describing, because it is the part a profitability table never shows you. Pawns' server refuses the plain Linux login and points at its app clients: a correct password comes back as authenticate user failed (please use SDK), while a wrong one gets the shorter authenticate user failed. I ran both on purpose to be sure which one I was looking at, so the account was fine and the host application was not. Their arm/v6 build is the only client that board can run, which means the Pi went back to being a DNS server.

That leaves the honest version of this experiment: on Windows or Android the app is that client, so it works there - the same few dollars a month, on a machine that is already on. If you want to try it that way, this is my referral link and I get a small credit if you ever reach a payout: https://cleosnine.com/rp