Earn up to ~$400/month per card by supplying RTX 5080s through Clore.ai. The headline comes from bare-metal contracts, which for the 5080 open at just two GPUs on 30-day terms; one card on its own, rented by the minute, can earn up to ~$300/month. Renters are already there: most RTX 5080 servers on Clore.ai were rented at the latest count. Set your own price and collect for every rented minute, in BTC, CLORE, USDT or USDC.
A 5080 is the lightest way into the current generation on both routes: a single card can be listed, and two cards meet the bare-metal minimum. Renters come for GDDR7 bandwidth and the FP4 tensor path.
81 RTX 5080 servers had a renter at the latest count, which is most of the pool. They choose it for 960 GB/s of GDDR7 and a Blackwell FP4 path that fits a quantized 8B model in about 5 GB, leaving the rest of the card for cache and concurrency.
You are paid for every rented minute in BTC, CLORE, USDT or USDC. Lock CLORE in MFP and you keep more of the BTC, USDT and USDC rentals, with MFP rewards in CLORE on top.
The agent installs onto a machine you already run: Ubuntu 22.04 or 24.04 LTS on x86_64, or a HiveOS rig, with an NVIDIA card whose driver already works. One GPU is a valid listing.
MFP is a score Clore assigns each server from its hardware and connection, capped at 200. That score sets how much CLORE you may lock: 2,000 per point in Tier 1, and MFP times 9,000 across all three tiers together. Locking raises what you keep on BTC, USDT and USDC marketplace payouts, and filling Tiers 1 and 2 adds CLORE rewards for as long as the server is rented by the minute.
These two tiers take MFP times 2,000 and MFP times 4,000 CLORE, and pay out in CLORE on top of the rental.
Tier 1 unlocks in 10 days, Tier 2 in 20 and Tier 3 in 30. Partial fills give a proportional boost.
A 5080 listing can take spot bids, on-demand orders at the rate you publish, or both. Whichever arrives, the order pays out for each minute a renter is on the card.
Spot renters bid for the card and a higher bid can take it over mid-rental, which suits the restartable batch jobs a 16 GB card often carries.
On-demand renters pay your published rate for as long as the order runs, and on a multi-GPU 5080 rig they can take one card or several.
Earnings settle into your balance in BTC, CLORE, USDT or USDC as each order runs. Payouts and fees →
Start with a single card on the per-minute marketplace, or bring two or more to a bare-metal contract. Each figure is a monthly ceiling for one card.
A single 5080 rented every hour of the month at the highest rates renters are paying for the card on Clore.ai today. One card and the installer is all it takes.
The top bare-metal contract rate for an RTX 5080, counted for every hour of the month, on blocks from 2 GPUs and 30-day terms. Contracts are set up with Clore directly rather than from a console toggle, so talk to us about supplying bare-metal capacity.
Fifth-generation tensor cores run FP4, so an 8B model fits in about 5 GB and most of the 16 GB is left for cache. That is the work that keeps 5080 listings rented.
Both routes are ceilings per card, per month, with MFP rewards available on top of the per-minute one. Full host guide →
There is no Clore operating system and nothing to boot from. The steps below follow Clore's installation guide, and the first one is the one to get right.
The installer will not install or upgrade a GPU driver. If nvidia-smi does not already return your 5080, fix that before going further.
Ubuntu 22.04 or 24.04 LTS on x86_64, or HiveOS. Stable is the default channel; beta and legacy exist if you need them.
Either add --init-token <token> to the install command, or run /opt/clore-hosting/clore.sh --init-token <token> once it is in place.
Spot, on-demand, or both, and you can revise either rate whenever demand shifts.
RTX 5080 bare-metal blocks start at two GPUs on 30-day terms, the lowest entry point of the Blackwell parts on Clore.ai, and this is the route behind the up to ~$400/month per card headline. Nothing in the console turns it on: contracts are agreed with Clore, so raise it with the team through support. Meanwhile both cards can sit on the per-minute marketplace, each earning up to ~$300/month per card.
They filter on VRAM first, and yours is 16 GB whatever the tensor format. FP4 matters after that filter, because it changes how much model fits inside those 16 GB: an 8B model is roughly 16 GB at FP16, about 8 GB at FP8 and about 5 GB at FP4. So the format widens what your listing can usefully serve without changing the number renters sort by. Nothing needs enabling on your side, since the capability belongs to the silicon rather than to a setting.
As close to continuous as you can make it. Renters can sort and filter the marketplace by reliability, so a server that stays online climbs the list and one that drops out mid-order slides down it. Clore’s hardware guidance asks hosts to disable overclocking and stress-test for stability before listing, and a current driver plus a steady connection do the rest.
On capacity it sits with the 4080 at 16 GB, so a renter filtering on memory sees your card among other 16 GB listings rather than among 24 GB ones. What separates you inside that group is bandwidth, 960 GB/s against a 4080’s 716, and an FP4 path that Ada does not have. Price against the 16 GB group and let the Blackwell features win the comparison.
The failure worth guarding against is a renter’s image refusing to start on hardware that looks perfectly healthy to you. GB203 is recent silicon: a driver predating it will not enumerate the card at all, and one that merely lags will decline CUDA runtimes newer than it supports, which is exactly what a current container ships. The installer is no help here, deliberately, since it neither installs nor upgrades drivers and the guide simply requires that nvidia-smi already returns your card. The guide names no minimum version, though your server’s page in the dashboard flags any driver below the 535 branch as outdated, so the working rule is practical: put a driver new enough for Blackwell on the box before anything else, and treat keeping it current as part of keeping the listing rentable.
Four: BTC, CLORE, USDT and USDC. Your balance is credited for every minute a rental runs and can be withdrawn on-chain, with the stablecoins leaving on Ethereum or BNB Smart Chain. MFP rewards are paid in CLORE, and a CLORE lock in MFP raises what you keep on the BTC, USDT and USDC rentals.
Sixteen gigabytes decides which jobs arrive at your listing. These three are the common ones, described by memory rather than by benchmark, since throughput belongs to the renter's runtime and not to your server.
The bread-and-butter image workload for a 16 GB card, and the one that arrives most often at listings in this bracket.
Read the guide →Full precision would take all 16 GB and leave no cache, so serving on this card means quantizing. FP4 is what the generation adds.
Read the guide →FP4 brings Flux-class image generation inside 16 GB, which is exactly the renter a 5080 listing is built to win.
Read the guide →Per card, per month. The per-minute column is a single listing's ceiling and the bare-metal column the top contract rate, with each card's contract entry alongside.
Prerequisites and the install command, how the MFP lock is sized, and how payouts work. If this page contradicts them, they win.
Rent one by the minute for FP4 serving or SDXL batches, and pay in BTC, CLORE, USDT or USDC.
Up to ~$300/month per card on the per-minute marketplace and up to ~$400/month per card once two or more go on contract. The install is one command on a machine you already run.