Renters keep this fleet busy: on 21 September 2026, 185 of the 186 V100 servers listed on Clore.ai were rented, across 1,389 cards, the largest datacenter fleet on the platform. The headline number comes from the bare-metal contract route: supply V100s in blocks of 8 or more for 30 days at a time and the contract pays up to ~$320/month per card. You can also list V100s per minute at your own price, and an SXM2 baseboard with eight boards is already the size a contract starts at.
Renters book nearly every listed V100, and Clore.ai sells V100 capacity on bare-metal terms as well. Three reasons to bring yours.
185 of 186 listed servers were rented on 21 September 2026, across 1,389 cards. Whatever renters run on these boards, they keep running it, and a fleet that size staying booked is a strong signal for anyone adding capacity.
On a fleet this tightly booked, renters get little extra for accepting interruption, so many reach for on-demand orders that keep the machine theirs until they stop. Offer both order types and let renters choose.
An SXM2 baseboard carries eight V100s joined by NVLink 2, which is exactly the size a bare-metal contract starts at. Supplied on contract, each card can earn up to ~$320/month; listed whole on the marketplace, the node sells the fabric renters cannot build from separate machines.
Every server carries an internal quality score, its Maximum Fair Price. Locking CLORE against that score brings in block-reward emissions and, separately, leaves you a bigger cut of every marketplace order paid in BTC, USDT or USDC. Both are optional, and the lock can grow server by server.
Maximum total is the score times 9,000 CLORE. Unlock periods are 10, 20 and 30 days by tier.
Tiers 1 and 2 share the emission allocation. Tier 3 adds no emissions; its job is to finish the lock, and a finished lock means more of each BTC, USDT or USDC rental stays with you.
The daily price you enter is what every V100 order bills against, minute by minute. Earnings land in your Clore.ai balance and leave it in the currency you choose.
The order renters favour on a tight fleet: the machine stays theirs and bills per minute until they release it.
Withdraw in whichever of the four you prefer. Locked CLORE, if you add an MFP lock, earns its rewards in CLORE.
For withdrawal details, see the host docs. Payouts and fees →
How the figure is set: ~$320 assumes a V100 working a bare-metal contract day and night at the best contract rate on offer for this GPU, with 8 GPUs and 30 days as the minimum. A per-minute listing is paid at the daily rate you pick.
For operators with eight V100s or more, which one HGX baseboard already is. A contract is set up in conversation with the Clore.ai team, not from the console.
For any V100 server, at a price you set per day in USD. Renters are already looking for V100s, so a reliable listing meets demand that is already there.
The strongest demand signal among datacenter cards on Clore.ai: nearly every listed V100 server was under an order at the last snapshot.
Have eight or more V100s? Talk to us about supplying bare-metal capacity →
Volta takes the same one-command install as every other card on Clore.ai; the care goes into the driver and the OS underneath it.
Volta is old enough that the driver is the part your renters will feel, so install a working NVIDIA driver before the agent goes anywhere near the box. The base underneath it is Ubuntu 22.04 LTS on x86_64 as the recommendation, with 24.04 supported and 18.04 and 20.04 not, plus 8 GB of RAM (16 recommended) and 32 GB of disk. HiveOS rigs follow their own path.
Choose Add Server and copy the one-line installer it generates with your init token. Run it as root. Docker is installed during setup.
After a reboot the machine appears online under My Servers. Confirm the reported memory: 16 GB and 32 GB V100s both exist, and renters filter on that.
With 186 servers listed, you can see what renters book before you set your own daily price, and change it whenever you like. MFP lock, if you want it, can wait.
Value for renters, mostly. 32 GB of HBM2 at 900 GB/s is more memory bandwidth for the money than anything else on Clore.ai, and for work bound by how fast weights move rather than by which data types the tensor cores support, that is hard to beat. Add the pipelines that were written against this generation and never migrated, and you get a fleet that clears.
Sell the fabric. An SXM2 baseboard carries NVLink 2 between the boards at roughly 300 GB/s and 256 GB of HBM2 across the eight of them, and no renter can reproduce that by booking eight unrelated machines. List the node whole for renters who want that fabric; on-demand renters can also take some of its GPUs rather than all eight, since the boards are identical. Eight boards is also exactly where bare-metal contracts start, so the same node can earn up to ~$320/month per card on contract.
Some will, and it is not something you can fix from the host side. Anything gated on Ampere or newer will refuse to run: FlashAttention, FP8 code paths, BF16 code paths, and an increasing number of wheels that ship only sm_80 and above kernels. What you can do is say the card generation plainly in your listing so that renters who need those things self-select out before they book, which is better for your reliability score than a refund conversation.
Orders on a V100 bill per minute, and what they earn is credited to your balance as it accrues. Cash out as Bitcoin, CLORE, USDT or USDC, and any MFP rewards on those orders land in CLORE. Contract income for bare-metal V100 blocks is paid in USD stablecoin or BTC.
NVIDIA has not published an end date we can cite, so this page will not put one on it. What is observable is the direction: Volta predates BF16, FP8 and MIG, and libraries are steadily narrowing which architectures they compile for. In practice that shows up as renters hitting unsupported-architecture errors rather than as a driver disappearing overnight, so state the generation in your listing and keep a driver branch your renters' images support. The pipelines pinned to Volta are exactly the ones that keep this fleet booked.
Blocks begin at 8 GPUs and 30 days. There is no contract button in the dashboard: supply is agreed directly with the Clore.ai team, so write to Support with how many V100s you have and how long they can stay committed. Each contracted card can earn up to ~$320/month, which makes a full HGX baseboard worth that conversation before you list it per minute.
Three categories, all bound by memory traffic rather than by tensor-core features, which is exactly where this board shines. Useful context for how you describe your listing.
Decode speed follows the memory bus, and this bus is fast for the money. Mention the 32 GB and the bandwidth in your listing.
Host docs →Code that was never migrated and does not need to be. This demand is stable precisely because it is not chasing new features.
Payouts and fees →1,389 cards across 186 servers averages well above one card per machine, and multi-GPU listings are what that fleet is made of.
MFP mechanics →Per-card monthly figures on Clore.ai with the specs renters filter on. Top route shows which way each card earns its highest figure.
Four links for V100 operators: getting the agent installed, the MFP lock in depth, how payouts work, and the full docs index.
Renters see a fleet that is hard to get onto. The rent page covers what Volta runs, which libraries need care, and how renters get a V100 when the fleet is booked.
185 of 186 listed servers were rented on 21 September 2026. Install the agent and list per minute, or bring eight or more cards to a bare-metal contract at up to ~$320/month per card.