Guide · Rented GPU

Rent a cloud GPU and mine

4 steps · about 15 minutes

  1. Read this before you rent

    Renting a GPU purely to mine Pearl did not cover its own rent in our tests. We think that is the single most useful thing this page can tell you, so it is step one rather than a footnote.

    On 2026-10-31 we rented an H100, an RTX 4090 and an RTX 3090 and mined each to this pool for real. Measured against the network yield and PRL price on the day, every one of them cost more in rent than it earned — the cheapest gap was the 4090 and it was still badly underwater. We destroyed all three. We have not re-run the test since, and nothing about the market has changed in a direction that would reverse it.

    So rent for the reasons that still make sense: to measure a card you are thinking of buying, to test your setup before committing hardware, or to see your own worker appear on the dashboard. Not as a way to make money.

  2. Rent an hourly GPU

    Any provider that gives you root on a CUDA box will do. Vast.ai is the one we use for this kind of short test — single cards, by the hour, no commitment.

    What to pick: a recent NVIDIA card, an image with CUDA already on it, and a host with good reliability. Check the live price yourself at the moment you rent — it moves hourly and by region, and any figure printed on a page like this one is stale by the time you read it.

    Watch the disk line. The headline hourly rate on most providers excludes storage. Budget for the real total, not the number in the search results.

  3. Install the miner and connect

    Once you have an SSH session on the instance, this is the whole job.

    curl -fsSL -O https://github.com/doktor83/SRBMiner-Multi/releases/download/3.8.3/SRBMiner-Multi-3-8-3-Linux.tar.gz
tar xzf SRBMiner-Multi-3-8-3-Linux.tar.gz
cd SRBMiner-Multi-3-8-3
./SRBMiner-MULTI --algorithm pearlhash \
  --pool stratum.linkrra.com:5567 \
  --wallet prl1your-address-here --worker rented1

    Give it ten minutes before you believe the hashrate — the first few minutes read high. Check your worker on the dashboard to confirm the pool is accepting shares, not just that the miner says so.

  4. Destroy the instance when you are done

    This is the step people skip and it is the one that costs money. A rented GPU bills until it is destroyed — stopping the miner, closing the terminal and closing the browser tab all leave the meter running.

    Destroy it in the provider's console and then reload the page to confirm it is gone. We have had teardown commands report success while the instance kept billing. Verify, do not assume.
Notes

Why no profit figures on this page. Pearl's yield per TH/s moves with the network hashrate, the block time and the PRL price, all of which change daily — and when a faster miner build rolls out network-wide, everyone's hashrate rises together and everyone's share per TH falls. A number printed here would be wrong within days. Do the arithmetic against live figures at the moment you rent: rent per hour against your card's measured hashrate, today's network hashrate and today's PRL price.

Do not use our on-site mining calculator for this. It has a known block-time bug that makes its output too optimistic. We would rather tell you that than have you rent on the strength of it.

Measured hashrates for the cards worth renting are on the NVIDIA guide — those are our own runs against this pool, with dates.

The pattern that does work is a card that serves AI inference and mines only in the gaps, so paid work covers the box and mining fills the idle seconds. That is a different guide, and today it is a private beta.

Pool details: stratum.linkrra.com:5567, no extra fee over the upstream pool's own (currently 0%), PPLNS, 1 PRL minimum.

The rent-versus-revenue finding is our own: three single-GPU rentals on 2026-10-31, mined to this pool, verified by pool-side accepted shares, costed against the hourly rate actually billed and the PRL price that day. Miner download from the SRBMiner-Multi releases page, checked 2026-11-03. We name Vast.ai because you would have to choose a provider yourself; we have no arrangement with them.