Log in Become a host
RTX 4070 · 12 GB Ada with FP8 24 servers rented right now

Rent out your 4070
for up to ~$100/month.

An RTX 4070 on Clore.ai can earn up to ~$100/month per card, which is what one card brings in when it is rented every hour of the month at the upper end of today's RTX 4070 prices. Demand is solid: 24 RTX 4070 servers are rented right now, more than are waiting for a renter. Ada's FP8 tensor cores and 12 GB of GDDR6X suit quantised model serving and SDXL, and you set the price, with payouts in BTC, CLORE, USDT or USDC for every minute a renter keeps the card.

BTC USDT USDC CLORE
●FP8 tensor cores ●You set the rate ●Paid per rented minute
RTX 4070 · WHAT ONE CARD CAN EARN monthly ceiling
Order routePer-minute rentals
Hours rentedEvery hour
Price levelUpper end of today's prices
Paid forEvery rented minute
Up to, per card / month ~$100USD
MFP rewards when CLORE is locked +CLOREoptional
Tensor cores
FP8, Ada
Systems
Ubuntu, HiveOS
Rented now
24 servers
~$100
Earn up to per RTX 4070 per month
24
RTX 4070 servers rented right now, more than are free
FP8
Ada tensor cores for renters serving quantised models
12GB
GDDR6X with room for SDXL at 1024 and 8B at FP8

Three reasons renters
pick a 4070.

Ada's FP8 tensor cores, twelve gigabytes that hold the popular image and chat workloads, and a small fleet where more servers are rented than free. Here is what each means for your listing.

More rented than free

24 RTX 4070 servers are rented on Clore.ai right now, ahead of the ones waiting for a renter. A new listing joins a small, busy fleet, where a reliable server at a fair price is easy for renters to find.

Servers rented now24
Earn up to~$100 / card / mo

The card Ampere hosts cannot offer

The 4070 is Ada, so it carries FP8 tensor cores. Every Ampere card on this marketplace, from the 3070 to the 3090, stops at INT8. For renters serving quantised models that is a real capability difference, and it belongs in your listing.

ArchitectureAda Lovelace
FP8 tensor coresYes

Twelve gigabytes, clearly stated

Twelve gigabytes serves an 8B model at FP8 in roughly 8 GB with room left for the KV cache, and holds SDXL at 1024 by 1024. Put the memory size and those workloads in your listing so renters know exactly what they are booking.

VRAM12 GB GDDR6X
Memory bandwidth504 GB/s
★ OPTIONAL · REWARDS IN CLORE

Lock CLORE for
MFP rewards on top.

MFP Lock sits beside the rental rather than changing it. Lock CLORE against the server's Maximum Fair Price score and every minute the 4070 spends rented also pays out MFP rewards, in CLORE. The same lock lets you keep more of rentals settled in BTC, USDT or USDC. Nothing about it is required, and you can add it after the server is already earning.

RENTAL

Your rate, per minute

Your pricefor every rented minute

Credited while an order runs and withdrawn whenever you choose.

+ MFP LOCK

CLORE rewards on top

MFP × 9,000CLORE fills every tier

Each MFP point sizes 2,000, 4,000 and 3,000 CLORE of lock across Tiers 1, 2 and 3, and a partial lock still earns its share of rewards.

Optional at every step · Rewards are paid in CLORE · Read the MFP Lock page →

Two ways renters book your 4070.

Every order is either on-demand at the rate you set or spot, where renters bid for free time. Both bill by the minute and pay out as they run.

Spot orders
Bids

Bid-based, fills free hours

Whenever the 4070 has no order, renters can bid on it from your spot floor up; the best bid runs until an on-demand booking or a stronger bid replaces it.

On-demand orders
Fixed

Your rate, reserved

Your listed rate applies for the full length of the booking and nobody can outbid it, which suits renters who need the machine for hours at a time.

In BTC, CLORE, USDT or USDC, every payout is yours to withdraw to a personal wallet. Payouts and fees →

Getting a 4070
toward ~$100 a month.

Three things move a real listing toward the ceiling, and all three are in your hands.

Availability
24/7

Keep the server online

Uptime shows on every listing as a reliability score that renters can sort and filter by. A 4070 that stays online rises in that ranking and stays in front of them.

Pricing
Your call

Spot and on-demand, set separately

Each order type gets its own per-day figure for the machine, stated in USD if you prefer, and you can revise both at any time.

Bonus
+MFP

Rewards in CLORE

With CLORE locked, each rented minute also brings in a CLORE reward alongside the rental itself.

Driver checks, the install command and pairing are all walked through in the host guide. Full host guide →

Four steps from
install to income.

The hosting agent installs onto a machine you already own and already run.

01 / INSTALL

A single install command

Supported on Ubuntu 22.04 and 24.04 and on HiveOS through Hive Shell, x86_64 with NVIDIA cards. The driver has to be working beforehand because the installer does not install or upgrade it. Stable, beta and legacy channels exist.

02 / PAIR

Register the machine

Create the server in the dashboard and give the installer its init token, or run the agent's clore.sh with that token afterwards.

03 / PRICE

Name your rate

List spot, on-demand or both. Price Coach in My Servers marks your rate green when it sits at market against rented 4070 servers, so you can tune it before the first order.

04 / WATCH

Then leave it alone

Uptime is what gets a server chosen again. Reboots, driver churn and a flaky line cost more rentals than a few cents on the rate ever will.

What 4070 hosts ask.

What do I need before I install the agent on a 4070 machine?

A supported system and a working driver. Supported systems are Ubuntu 22.04, Ubuntu 24.04 and HiveOS on x86_64 with NVIDIA cards, and the installer leaves the driver alone, so nvidia-smi has to see the 4070 before you start. After that it is one command copied from the dashboard, plus the init token that pairs the machine with your account.

Can I earn MFP rewards on a 4070 as well as rental income?

Yes. MFP Lock is optional and runs alongside your rentals: with CLORE locked against the server's MFP score, the 4070 collects MFP rewards for each minute it is rented and keeps a bigger slice of BTC, USDT and USDC payments. Tier 1 takes MFP × 2,000, all three tiers together take MFP × 9,000, and a partial lock still counts.

Does FP8 support make my 4070 more attractive to renters than a 3080?

For renters serving quantised models, yes. The 4070 is Ada and carries FP8 tensor cores; the 3080 is Ampere and stops at INT8, so a model quantised to FP8 keeps more of its dynamic range on your card. For image work renters compare memory instead, and your 12 GB holds SDXL at 1024 by 1024 with room to spare, where a 10 GB card needs a tiled VAE.

Which renters pick a 12 GB card?

Renters with image, chat and vision jobs of that size. Twelve gigabytes covers SDXL at 1024 by 1024, SD 1.5 fine-tuning, quantised 8B serving and most computer-vision work, so those are the orders a 4070 attracts. Describe the card by those workloads and price it against other 12 GB listings, because that is the shortlist renters compare you with.

How busy is the RTX 4070 fleet on Clore.ai?

Busy for its size: 24 RTX 4070 servers are rented right now, and fewer than that are free. In a fleet this small each listing is easy for renters to find, and the ones that stay online with a clear description and a rate inside today's range are the ones that get booked.

Can I change my 4070's price after it is listed?

Yes, at any time from My Servers. If you state the price in USD per day, the platform recalculates the BTC and CLORE amounts as exchange rates move, so the dollar figure you chose holds. While you edit, Price Coach flags whether your per-GPU rate is at market, above it or well above it compared with rented RTX 4070 servers.

What people rent it for.

Three jobs that fit 12 GB, stated in gigabytes. When a renter asks whether their model will run, these are the answers to give them.

SDXL with one ControlNet
ComfyUI, fp16
~10 GB base, adapter on top

SDXL's own weights leave a couple of gigabytes free, which is enough for one adapter and rarely for two.

Read the guide →
Quantised 8B serving
llama.cpp server, GGUF
~8 GB at one byte each

At FP16 the same model needs about 16 GB and will not load. Quantised, it leaves roughly 4 GB for the cache.

Read the guide →
SD 1.5 subject training
Diffusers, 8-bit Adam
fits without offload

DreamBooth on SD 1.5 is the fine-tuning job a 12 GB board handles outright. SDXL wants a larger card.

Read the guide →

The 4070 beside two popular neighbours.

Monthly figures per card use the same method as the headline above, set against architecture and today's count of rented servers. Only the Ada card in this table carries FP8.

GPU
VRAM
Architecture
FP8 tensor cores
Rented now
Marketplace ceiling
Contract ceiling
Top figure / card
RTX 3070
8 GB GDDR6
Ampere
no
63
~$115/mo
n/a
~$115/mo
RTX 4070 / this page
12 GB GDDR6X
Ada
yes
24
~$100/mo
n/a
~$100/mo
RTX 3090
24 GB GDDR6X
Ampere
no
133
~$145/mo
~$330/mo
~$330/mo

Host docs for your 4070.

The installer guide, the MFP Lock breakdown and the payouts page, all on docs.clore.ai and kept current.

Host docs
Becoming a host
Host docs
MFP Lock, in full
Host docs
Payouts and fees
Host docs
All docs

The cards on either side.

RTX 3070
63 servers rented now · up to ~$115/mo per card
Host →
RTX 4070 Ti
17 servers rented now · up to ~$145/mo per card
Host →
RTX 4080
7 servers rented now · up to ~$230/mo per card
Host →

Twelve gigabytes of Ada,
earning every minute.

Install the agent on Ubuntu or HiveOS, pair it with an init token and set your rate. Each card can bring in up to ~$100/month, with payouts in BTC, CLORE, USDT or USDC.