Log in Become a host
RTX 4080 · per-minute hosting paid for every rented minute

Your 4080 can earn
up to ~$230/mo.

An RTX 4080 listed on Clore.ai can earn up to ~$230/month per card on the per-minute marketplace. The 4080 pool is small and busy: most of its servers had a renter at the latest count, so a new listing is easy for renters to find. Your price, your listing: renters pay for each minute they use, and payouts arrive in BTC, USDT, USDC or CLORE.

BTC USDT USDC CLORE
●You set the price ●Paid per rented minute ●BTC, USDT, USDC or CLORE
RTX 4080 · EARNING CEILING per-minute marketplace
RoutePer-minute listing
Order typesSpot and on-demand
Payout currenciesBTC · USDT · USDC · CLORE
SettlementEvery minute
Earn up to, per card per month $230USD
Memory renters get 16GB
Bandwidth
716.8 GB/s
Tensor path
FP8
Minimum to list
One GPU
$230
Up to, per card per month, on the per-minute marketplace
1
GPU is enough for a listing, even in a desktop
16GB
GDDR6X per card, room for 8B serving at FP8
4
Payout currencies: BTC, CLORE, USDT and USDC

A small pool
that stays busy.

Most GPUs here are listed in the hundreds. The plain RTX 4080 is a far smaller pool, and at the latest count the majority of its servers were rented. That makes every new listing visible, and it puts the price in your hands.

Every new listing gets seen

In a pool this size, one more server is a visible share of what renters can choose from, with no wall of look-alike listings to stand out against. A search for RTX 4080 also returns 4080 SUPER servers, so your listing reaches both groups of shoppers.

Servers rented, latest count7
RoutePer-minute marketplace

Paid per rented minute

Earnings credit to your Clore balance while an order runs. Withdraw to BTC, USDT, USDC or CLORE whenever you want.

Billing cadencePer minute
Payout currencies4

One card or twenty-five

A single 4080 in a desktop is a valid listing. An account carries up to 25 servers by default and support raises that on request.

Minimum servers1
Default account cap25
★ OPTIONAL · PAID IN CLORE

Lock CLORE,
collect a daily emission.

MFP Lock is optional. You lock CLORE against a server's Maximum Fair Price score and the network pays a daily emission in CLORE on top of the rental, with each of the first two tiers capped at the server's own daily rental price. The lock also raises what you keep when a renter pays in BTC, USDT or USDC.

RENTAL

What renters pay you

Your rateset by you, billed per minute

Credited per minute, withdrawn in BTC, USDT, USDC or CLORE.

+ MFP LOCK (OPTIONAL)

Daily emission in CLORE

Tiers 1 + 2each capped at the daily rental price

Lock size follows the MFP score, starting at 2,000 CLORE per point in Tier 1, and every tier raises what you keep on non-CLORE rentals.

Optional, and nothing is lost by skipping it · Emission paid daily in CLORE · How MFP Lock works →

Two order types, both paid by the minute.

Your 4080 can take spot bids, on-demand orders at your fixed rate, or both. Either way the order settles into your balance minute by minute while it runs.

Spot orders
Bids

The auction side

Renters bid for the server and the highest bid holds it, so when the small 4080 pool is busy, spot demand can lift your rate.

On-demand orders
Fixed

The reserved side

Renters pay the rate you publish and the order cannot be outbid, which suits serving, demos and interactive work that needs the card held.

Payouts arrive in BTC, USDT, USDC or CLORE. Payouts and fees →

Where up to ~$230/mo per card comes from.

One card, one month, on the per-minute marketplace. The cards below show the ceiling, what moves you toward it, and what MFP adds on top.

The ceiling
$230

Up to ~$230/month per card

A 4080 rented day and night at the top of today's RTX 4080 prices on Clore.ai. That is the per-minute route, open to a single card in a desktop.

Your lever
Price

You set spot and on-demand

Choose spot, on-demand or both, and reprice at will. A competitive rate and steady uptime are what keep the card rented toward that ceiling.

On top
MFP

Optional CLORE emission

Lock CLORE against the server and Tiers 1 and 2 pay an emission in CLORE on top of the rental, each capped at the server's daily rental price.

The ceiling is per card, per month, before any MFP emission. Full host guide →

From an idle card
to a listing.

The agent installs onto a machine you already have. There is no Clore operating system and nothing to boot from a USB stick.

01 / INSTALL

Run the installer

On Ubuntu 22.04 or 24.04, or on HiveOS through Hive Shell, run the one-line installer from the public hosting-agent-installer repository. NVIDIA cards only, and a working driver has to be in place first because the installer does not touch drivers.

02 / PAIR

Attach it to your account

Create the server in your dashboard and pass its init token to the installer, or register afterwards with the agent's own clore.sh script.

03 / PRICE

Set spot and on-demand

List one order type or both, and change the rate whenever you like. The marketplace shows what other 4080 and 4080 SUPER hosts are charging.

04 / LOCK

Consider MFP Lock

Optional. Lock size follows the server's MFP score, from 2,000 CLORE per point in Tier 1 up to 9,000 CLORE per point across three tiers, and Tiers 1 and 2 pay a daily emission in CLORE.

Pricing and earning with a 4080.

How much demand is there for RTX 4080s on Clore.ai?

More than the size of the pool suggests. At the latest count most of the RTX 4080 servers on Clore.ai had a renter on them. The marketplace GPU filter is a substring match, so a renter searching RTX 4080 also sees 4080 SUPER servers, which puts your listing in front of both groups of shoppers.

Does 16 GB put me in the same search results as 24 GB cards, or a different bucket?

A different one in practice. Renters filter on VRAM before anything else, so a 16 GB card is weighed against other 16 GB cards rather than against the 24 GB tier, and the RTX 5080 is the other 16 GB card they compare you with. Closer still is the RTX 4080 SUPER: the marketplace GPU filter is a substring match, so a renter searching RTX 4080 sees both products in one list, and 16 GB of VRAM does not separate them either. Those are the listings you are actually competing with, not the 4090.

How and when does a 4080 host get paid?

Minute by minute. While an order runs, settlement credits your Clore.ai balance for every minute of use, in BTC, USDT, USDC or CLORE. Those balances can be withdrawn on-chain whenever you choose, and if you lock CLORE in MFP, its emission arrives in CLORE as well.

Which renters choose a 4080?

The ones whose work fits in 16 GB, which covers most diffusion pipelines, 8B models served at FP8 and QLoRA on 7B and 8B bases. For those jobs a 4080 does the work at a lower hourly rate than a 24 GB card, and that is exactly why they pick it. Knowing them helps you describe the listing in the terms they search for.

Should I price against the 4070 Ti below me or the 4090 above me?

Against the card above. The 4070 Ti sits in a different VRAM bucket, so the only renters comparing it with you are the ones still deciding how much memory they need. The 4090 is the real alternative for anyone who could use either. Sitting a little under the 24 GB tier is a stronger position than sitting a little over the 12 GB one.

Do I need a datacenter, or will a desktop do?

A desktop is fine. The hosting agent installs onto an ordinary Ubuntu 22.04 or 24.04 machine, or onto HiveOS through Hive Shell. It needs NVIDIA hardware with a working driver already in place, around 32 GB of disk with 30 GB free on the system partition, and 8 GB of RAM as a floor. What matters more to renters than the chassis is uptime and a stable line, because a server that vanishes mid-order stops getting picked.

What renters come here for.

A renter's first question is whether their model fits, so these three are sized in gigabytes rather than in throughput. Knowing them helps you answer support questions and price with a straight face.

Quantised 8B serving
vLLM or TensorRT-LLM
~8 GB of weights at FP8

FP16 weights for an 8B model come to about 16 GB, the whole board, so renters quantise and spend the remainder on cache. Ada does FP8 in hardware.

Read the guide →
SDXL production batches
ComfyUI, fp16
batch 4 stays resident

Sixteen gigabytes holds the UNet, the VAE and both text encoders at once. That headroom is the difference renters pay for over a 12 GB board.

Read the guide →
QLoRA on 7B and 8B bases
PEFT, 4-bit NF4
~4 GB of weights

Four-bit leaves twelve gigabytes for optimizer state and batch size, which is why fine-tuning renters shop at 16 GB rather than 12.

Read the guide →

The card below you and the card above.

Per card, per month, calculated the same way for each GPU from what renters currently pay on Clore.ai. The bare-metal column applies where Clore sells contracts for that card.

GPU
VRAM
Memory bandwidth
Servers rented
Earn up to
Per-minute route
Bare-metal route
Contract entry
RTX 4070 Ti
12 GB GDDR6X
504 GB/s
17
$145
$145
n/a
n/a
RTX 4080 / this page
16 GB GDDR6X
716.8 GB/s
7
$230
$230
n/a
n/a
RTX 4090
24 GB GDDR6X
1,008 GB/s
112
$620
$490
$620
5 GPUs, 14 days

The three documents behind this page.

Installation, the MFP Lock mechanics and payouts. Everything above is derived from these plus a read of the marketplace.

Host docs
Becoming a host
Host docs
MFP, the Maximum Fair Price lock
Host docs
Payouts and fees
Host docs
All docs

Other cards you can list.

RTX 4070 Ti
12 GB Ada on the per-minute route
Host →
RTX 4090
24 GB Ada, contracts from 5 GPUs
Host →
RTX 5080
16 GB Blackwell with an FP4 path
Host →

Turn a spare 4080
into income.

Install the agent on an Ubuntu or HiveOS machine, pair it with an init token, set your own rate. Payouts land per rented minute in BTC, USDT, USDC or CLORE.