Each RTX A6000 can earn up to ~$630/month per card on Clore.ai. That top figure is the bare-metal contract route, for operators with 8 or more cards on 30-day terms. List per minute instead and a card earns up to ~$300/month, at a price you set and paid for every rented minute in BTC, CLORE, USDT or USDC. Either way, 48 GB of ECC memory puts you in front of renters who filter by VRAM before anything else.
Capacity brings renters to your listing, per-minute settlement pays for every hour they stay, and one machine is enough to begin. Operators with eight cards or more get a second route on top.
The 48 GB threshold is a hard line in how renters search. A 70B model at four-bit precision needs roughly 40 GB resident, so whoever is running one filters on VRAM before anything else and lands on cards like yours. That is demand a 24 GB listing never sees.
Every minute an order runs is a minute you are paid for, in BTC, CLORE, USDT or USDC. A renter who books for an afternoon pays for exactly that afternoon, and one who stays for a week pays for the whole week.
There is no minimum fleet and no eligibility tier to qualify for. A single tower with one A6000 in it goes through the same install command and the same dashboard as a rack. With eight cards or more, bare-metal contracts open up as well.
MFP Lock is a bonus you can switch on once the A6000 is earning. CLORE locked behind the server raises what you keep on marketplace rentals paid in BTC, USDT or USDC, and it pays MFP rewards in CLORE while the machine is rented per minute. Leave it off and your listing works exactly the same.
Tier 1 and Tier 2 share in the network's CLORE emission for each minute your A6000 carries an order, with each tier sized against the server's own daily price.
Tiers fill in order, starting with Tier 1. The fuller the lock, the more of each BTC, USDT and USDC payment stays in your balance.
List per minute on the marketplace, or supply a bare-metal contract through Clore. Here is what each route can pay for a single card.
That is an A6000 rented around the clock at the top of what renters currently pay for this card on Clore.ai. You choose the price and the currencies, and every rented minute is paid.
The top bare-metal contract rate for the A6000, from 8 GPUs on 30-day terms, around the clock. These contracts are arranged with Clore rather than switched on in the console, so talk to us about supplying bare-metal capacity.
Currencies and payout timing are explained in the host docs. Payouts and fees →
Both route figures describe a card that stays busy. These are the settings that move your own A6000 listing toward them.
Clore.ai never prices your machine for you. Publish an on-demand price and a spot floor for each currency, or track USD so the crypto amount follows the exchange rate. The pricing form hints at the market while you type.
Renters can sort the marketplace by reliability as well as price. Telemetry and an uptime chart for every server help you keep the A6000 near the top of that sort.
Lock CLORE behind the server to keep more of every Bitcoin and stablecoin marketplace order and to collect MFP rewards while it is rented per minute.
Setup, pricing and payouts are all in the host guide. Full host guide →
There is no CLORE operating system, no bootable image and no PXE step. The agent installs onto an existing Ubuntu or HiveOS box over an NVIDIA GPU, and the token from your dashboard registers it to your account as it goes.
On Ubuntu 22.04 or 24.04, x86_64 (18.04 and 20.04 are no longer supported), or on HiveOS with the same command. Your dashboard fills in the token when you start the add-server flow.
The token passed to the installer makes the box self-register. If you installed without one, run /opt/clore-hosting/clore.sh --init-token <token> afterwards. AMD cards are not supported.
Decide whether you want spot orders, on-demand orders or both, and set the daily price yourself. You can change it whenever you like from My Servers.
MFP Lock works on a server that is already running, so there is nothing to set up on day one. Lock CLORE once the orders are coming in.
Yes. Someone who needs Llama 3 70B at INT4, around 40 GB of weights, sorts by capacity first and never sees a 24 GB listing at all. Your A6000 sits on the right side of that filter, and a single card can earn up to ~$300 a month on the per-minute marketplace or up to ~$630 on a bare-metal contract.
The split is NVLink against FP8. Your A6000 has an NVLink connector and the RTX 6000 Ada does not, so two of your boards can run tensor-parallel over a 112.5 GB/s peer link. The Ada card reads memory at 960 GB/s against your 768 and adds FP8 tensor cores. If you already own the A6000, list it and lead with the pairing: it is a feature the newer board lacks, and both cards sit in the 48 GB bracket renters filter for.
By the minute. Each marketplace order draws against the daily price you publish, and the income collects in your balance in the currency the renter paid, from the set you enabled: Bitcoin, CLORE, USDT or USDC. My Servers has a revenue view per server, so you can see which machines earn the most.
You list the machine, not the bridge. CLORE lists a server with its GPU count and the renter picks the server, so anyone who wants tensor parallelism across two boards is shopping for a 2-GPU machine. Be precise about what you are advertising: an NVLink pair is two 48 GB address spaces with a fast link between them, not one 96 GB pool. A renter who arrives expecting 96 GB opens a support ticket.
Contracts start at 8 A6000 cards on a minimum 30-day term, and they are set up with Clore directly rather than created from the dashboard. If you run eight or more cards, contact support to offer them. With one or two cards, the per-minute marketplace is open to you today.
Yes, on on-demand orders. When every GPU in a rig is the same model, a renter can lease just some of its cards, so one A6000 server can host several renters side by side. The feature stays on unless you switch it off for that server.
Operational patterns a 48 GB board attracts, and what each one asks of the box it sits in. Throughput depends on the renter's stack, so these are shapes rather than benchmarks.
A four-bit 70B model claims most of the board and keeps it for the life of the order. Steady draw rather than spiky, and almost no VRAM free for anything else while it runs.
Read the guide →Orders measured in hours or days, with the board busy throughout. A machine that stays online collects every one of those hours, billed by the minute.
Read the guide →Renters sharding a model across cards search on GPU count, so a single-card machine never surfaces for them. If you own two A6000s, listing them in one server opens a filter a pair of separate listings does not.
Read the guide →Specs from NVIDIA datasheets. The A6000 pairs 48 GB of ECC memory with an NVLink connector, which the newer Ada board leaves out. Follow a row for that card's host page.
Install steps, payouts and the MFP lock tiers referenced above all come from these. If something here disagrees with the docs, the docs are right and we want to hear about it.
The renter page works through the 48 GB memory budget in detail: what a 70B model at INT4 leaves behind, and how 768 GB/s sets the pace of generation.
Get an account, generate a token and run the installer on the Ubuntu or HiveOS machine the card already lives in. From there each A6000 can earn up to ~$630 a month, and renters are already booking this card.