Those earnings come through the bare-metal contract route. Clore.ai sells A10 capacity to renters on bare-metal terms, 8 GPUs and 30 days at minimum, and cards supplied into those contracts can earn up to ~$420/month per card. The A10 family is what the large clouds put into their inference instance types, so its renters are easy to picture: teams porting or sizing a workload against one of those instances. Each NVIDIA A10 brings 24 GB of GDDR6 at 600 GB/s, and a single server can also be listed by the minute at a price you choose.
Renters who already know the card, a contract route with a clear minimum, and hardware that suits dense servers. Three reasons in detail.
The large clouds standardised on the A10 family for inference instances, so teams who benchmark there arrive knowing what they want. Clore.ai sells A10 capacity on contract, which means supplied cards go straight into capacity the platform already offers. Set your rate and name the part exactly.
Six hundred gigabytes per second of bandwidth is double an L4's, and 125 dense FP16 tensor TFLOPS is roughly double its throughput too. For renters whose bottleneck is token generation, that is the difference that matters, and it is what your listing is selling.
An A10 rents as a whole card, and quantized serving on it runs in INT8 and INT4. Say both in your listing so the renters who want exactly that find your server first and stay with it.
Locking CLORE against your server's MFP score is optional. It opens a stream of MFP rewards in CLORE on per-minute rentals and increases your take on those settled in BTC, USDT or USDC. The lock is sized by the score, so it grows with the servers you list.
Tier 1 holds MFP × 2,000, Tier 2 × 4,000 and Tier 3 × 3,000, so a higher score opens a bigger lock.
Each tier's daily reward is capped at your daily rental price, which keeps rewards in step with rental income. Fill all three tiers and a larger share of every BTC, USDT or USDC rental is yours.
Per-minute orders bill against the price you set and pay out in BTC, CLORE, USDT or USDC; contract income is paid in USD stablecoin or BTC.
Renters pay for the minutes they use, against the daily USD price you enter, whichever order type they choose.
Blocks of eight or more A10s on terms of 30 days or longer are arranged with the Clore.ai team, at up to ~$420/month per card.
Per-minute earnings can be withdrawn in Bitcoin, CLORE, USDT or USDC, and rewards from an MFP lock are credited in CLORE.
Payout rules for hosts, in one place. Payouts and fees →
The figure assumes an A10 supplied on a bare-metal contract and working every hour of the month at the best rate on the A10 contract schedule. Blocks start at 8 GPUs on 30-day terms, and supplying them is something you arrange with our team rather than a setting you flip in the console.
For operators with eight or more A10s in servers built for passive cards. Tell us what you can supply and for how long.
List A10 servers on the marketplace at a price you choose and revise it as often as you want. The documented ceiling is $1,000 per day, and every rented minute is billed.
Renters who benchmarked on an A10-family cloud instance come looking for the same class of card, and your listing is where they find it.
Supplying eight or more A10s? Talk to us about supplying bare-metal capacity → Pricing a single server? How pricing works →
The hosting agent installs onto a machine that is already running Ubuntu, or attaches to a HiveOS rig through a flight sheet.
The A10 has no fan. It belongs in a server that forces air across it. While you are in there: at least 8 GB of RAM, 16 recommended, and 32 GB of disk with 30 GB free.
Install the NVIDIA driver before the agent. Then My Servers, Add Server, and run the generated one-line install command as root on the machine.
Given how easily this part is confused with others, spell the model out. A renter who wanted something else and got your server is a refund conversation nobody needs.
Prices go in as USD per day and can change at any time. With eight or more cards, the contract route pays up to ~$420/month per card, so ask about it before pricing cards one at a time.
Up to ~$420/month per card on a bare-metal contract. Renters buy A10 capacity from Clore.ai in blocks of eight or more GPUs held for 30 days and longer, and those blocks are filled with cards from hosts like you. A single A10 server can also go on the per-minute marketplace, with a daily price you choose and payment for every rented minute.
Not on a single card. The mechanism for that is MIG, and NVIDIA's supported-GPU list, checked on 21 September 2026, puts it on the A100 and A30, the H100 and H200, the B200 and GB200, and the RTX PRO Blackwell cards rather than on the A10. A server with several A10s can still serve several renters at once, because an on-demand renter can book just part of the machine, a few whole cards at a time. If slicing one GPU between tenants is the business you want, an A100 does it with up to seven instances.
A chassis that moves air. The card has no fan of its own, so it depends on the server pushing air across its heatsink, which is what rack servers are built to do. In return it is a single-slot card, so a server with the right airflow can carry several. CLORE's documented host minimums are 8 GB of system RAM, 16 recommended, and 32 GB of disk with 30 GB free.
On the contract route the A10 does: up to ~$420/month per card against up to ~$350/month for an L4. What renters pay for is throughput, 600 GB/s of memory bandwidth against the L4's 300 and 125 dense FP16 tensor TFLOPS against 60.5, and that matters most for token generation. The L4's own strength is FP8, which the Ampere A10 does not have, so describe each card for the work it does best.
Often, yes, and they are the clearest source of demand for an A10 listing. The family is what the large clouds put into their inference instances, so the people who want this specific card are usually porting a workload off one of those or sizing one against it. Name the part exactly, because the A10, the A10G that those cloud instances carry and the A100 are three different products, and a renter who gets the card they searched for is a renter who comes back.
Set an on-demand rate in USD per day, plus a lower spot rate if you want interruptible work too. Both are turned into BTC and CLORE at the going rate, orders bill per minute, and the documented ceiling is $1,000 per day. You can change prices whenever you like, and the pricing guide in the host docs walks through each setting.
Useful context for your listing description: each of these arrives looking for exactly the strengths this card has.
Token generation follows the memory bus, so renters serving chat models pick the card with the faster one.
Payouts and fees →Renters who outgrow an L4's throughput move to an A10. If you own both cards, list both and let each find its workload.
L4 hosting →Renters whose models no longer fit a T4's 16 GB move to 24 GB cards, and the A10 puts 600 GB/s behind the extra memory.
T4 hosting →Monthly earning ceiling per card first, then the specs renters compare. Each of these cards is contracted in blocks from 8 GPUs.
All four earn their top figure on bare-metal contracts; per-minute listings are priced by their hosts.
Setup, the MFP lock, payouts and the full index. Everything a new A10 host needs sits in one of these.
The renter page covers the same card from the other direction: the naming trap, what fits in 24 GB, and how renters order an A10 block.
Eight or more A10s in rack servers can earn up to ~$420/month per card on a bare-metal contract. A single server lists per minute at a price you set.