A GPU you already own
- 01Any NVIDIA card with 16 GB of VRAM or more, on Linux with Docker
- 02100 Mbit/s symmetric and a public address is enough
- 03We install the agent, benchmark the card and add it to the pool
Turn idle GPUs into hourly income
Connect a GPU you already own and get paid every hour for the capacity you are not using. No deposits, no lock-in, disconnect whenever you like.
Just need compute? Rent capacity by the hour
Paid every hour80% goes to operatorsIsolated containers
GPUs online
3,180+142 this week
Regions
41+3 new
Paid to operators
$1.24M+14.2% MoM
Built on the stack you already run
NVIDIA
CUDA12.x
PyTorch
Docker
Kubernetes
USDCBase
USDCPolygon
Prometheus
01 / How it works
Idle silicon in.
Cash out.
You already bought the card. We fill the hours it spends idle with paid inference and training jobs, and settle your cut every hour.
Settled every single hour
- 01Operators share 80% of every winning bid, split by GPU-hours actually delivered
- 02Balance updates hourly — no invoices, no net-30 terms
- 03Withdraw from $10 in USDC, or by bank transfer
AI teams bidding for capacity
- 01Teams bid for GPU-hours in sealed auctions, hour by hour
- 02Second-price auctions mean the pool clears at a fair rate
- 03More demand in your VRAM class is simply a higher rate for your node
02 / Platform
The machine runs itself,
you just collect.
Hourly settlement, second-price auctions, continuously probed nodes and instant failover — income that does not need babysitting.
01
Real CUDA workloads
Containerized training and inference jobs against your actual card — not synthetic benchmarks.
02
Hourly payouts
Earnings settle every hour in USDC on Base, Polygon or Arbitrum, or by bank transfer. Withdraw from $10.
03
VRAM-class scheduling
Every job lands on the nearest healthy node that fits its memory profile. Failover is instant and invisible.
04
Fair hourly auctions
Sealed second-price auctions run every UTC hour. Outbid? Your held balance releases immediately.
05
Live performance data
GPU-hours, utilization and earnings per card, streaming into one dashboard.
06
Isolation by default
Every job runs in a locked-down container. Workload contents are never inspected or retained.
07
Reserved capacity
Need guaranteed throughput? Reserve a VRAM class ahead of the spot market at a fixed hourly rate.
08
CLI + REST API
Everything the panel can do, scriptable. Manage nodes, bids and payouts from code.
03 / Control room
One panel. Every node.
Real-time everything.
Watch each card earn as it happens — live auctions, per-node utilization, health and hourly payouts in a single dashboard.
live
last 24h71.6
GPU-hours / day
92.4%
Utilization
$8.42
Earned today
Live node analytics
GPU-hours, utilization and earnings per card — refreshed every minute.
Wallet · this hour
$8.42
19:00 UTC+$0.84Base
18:00 UTC+$0.91Base
17:00 UTC+$0.79Polygon
Hourly payouts
Settled every hour, on the rail you choose.
Slot · 20:00 UTC · 24GB classcloses 12:44
northwind-ai$14.20
lumen-research$13.85
parabola-labs$12.40
Winner pays the second price: $13.86
Second-price auctions
Win the hour, pay the second price. Holds release the moment you are outbid.
Nearest node lookup
FRA · Frankfurt12ms
AMS · Amsterdam18ms
SIN · Singapore41ms
GRU · São Paulo58ms
VRAM-class scheduling
Memory-aware allocation with instant failover to the next-best node.
de-fra-03 · checked 40s ago
✓CUDA probe99.98%
✓Thermal headroom99.95%
✓Throughput probe99.92%
Always-on health checks
Every node is probed continuously and pulled from the pool the moment it degrades.
Auto · nearestFrankfurtSingaporeSão Paulo
docker run --gpus all voltmesh/agent --token vm_8f4a9e…One-command connect
A single docker run brings a card online. No dashboard clicking, no manual registration.
04 / Global reach
3,180 GPUs,
41 regions, one pool.
Operators run real hardware where demand is highest and get paid wherever they are. Memory-aware routing sends every job to the closest healthy node that fits it.
live · refreshed every 60s
<30ms
Median scheduling decision
99.94%
Pool uptime
41
Regions online
2.4M
GPU-hours / month
05 / Leaderboard
Top operators
this week.
Real hardware, real payouts. The pool splits 80% of every winning bid across operators, in proportion to the GPU-hours each one delivered.
NO
#01@nordiccompute
FI · Helsinki
$412.18
earned this week
38 GPUs99.98% uptime
GU
#02@gulfcompute
AE · Dubai
$366.02
earned this week
27 GPUs99.95% uptime
AN
#03@andeangrid
CL · Santiago
$298.77
earned this week
19 GPUs99.91% uptime
06 / Proof
People already earning,
in their own words.
Marek K.
Operator · 38 GPUs · Helsinki
I connected three 4090 rigs on a Sunday afternoon. The first payout landed before dinner — $0.91 for doing absolutely nothing. It has settled every hour since.
07 / Insights
From the network,
with real numbers.
Earnings breakdowns, auction mechanics and scheduler internals — written for people who run the pool, not around it.
Case studyJul 14, 2026 · 6 min read
How @nordiccompute turned three idle 4090 rigs into $412/month
No trading, no tenants, no maintenance window — just scheduled GPU-hours settling hourly to a wallet.
GuideJul 7, 2026 · 4 min read
Second-price auctions, explained — why you never overpay for an hour
Bid your true max. Pay the second price. Get your hold back the instant you are outbid. The math, walked through.
EngineeringJun 30, 2026 · 8 min read
41 regions in: how the scheduler picks your node in under 30ms
Continuous probes, latency histograms and instant failover — inside the scheduling layer that keeps the pool honest.
08 / FAQ
Questions,
answered.
Everything operators usually ask before connecting their first card. Something else? Support answers in under a minute.
Each hour the winning bid is charged to the customer. The platform keeps a 20% commission; what is left forms the operator pool for that hour. The pool is then divided between operators in proportion to the verified GPU-hours each one delivered — so what grows your share is capacity actually used, not the number of cards you connect.
No. You do not buy anything from us and we never take deposits. You connect a card you already own and earn a share of the 80% operator pool, sized by the GPU-hours your node delivers. You keep your hardware and can disconnect at any time.
Any time the wallet shows $10 or more. Withdrawals are paid in USDC on Base, Polygon or Arbitrum, or by bank transfer — typically within a few hours of the request.
Any NVIDIA GPU with 16 GB of VRAM or more, on Linux with Docker, plus 100 Mbit/s symmetric. Larger cards reach higher-paying VRAM classes, so a 24 GB card clears more per hour than a 16 GB one.
No. Every job runs in a locked-down container with no access to the host filesystem or your other processes. We retain scheduling and billing metadata only — workload contents are never inspected or stored.
The job is rescheduled onto the next-best node automatically and the customer is not charged for the lost time. Your health score takes the hit; drop below 70 and you stop receiving new jobs until it recovers.
Each bid is capped by your maximum and funds equal to your current bid are held from your balance — you can never bid more than you have free. If you are outbid or cancel, the hold is released instantly.
No. Voltmesh is a fictional company built as a portfolio piece to demonstrate landing-page design, content architecture and front-end engineering. Every figure on this page is illustrative.
09 / LaunchPut your idle card
Put your idle card
to work tonight.
Connect in about five minutes, share in the 80% operator pool, and withdraw whenever you like. The first payout lands within the hour.
Hourly settlement · isolated containers · no lock-in