Compute is a two-sided GPU network — deploy VMs, containers and bare metal from a 47-GPU catalog, or run the open-source worker and earn credits from hardware you already own.
Compute connects two sides with no middleman. Hardware owners run an open-source worker and earn credits whenever their GPUs take a job. Renters spend credits to deploy instances on demand. Every site built on the Platform is a real job — so demand is guaranteed, not speculative.
VMs, containers, Kubernetes and bare metal — from a 47-GPU catalog, billed per second, no contracts.
The catalog spans consumer cards for light inference all the way to multi-GPU datacenter nodes. Tell Compute the workload and it filters to the GPUs that fit; pick one and it deploys with your template attached.
| GPU | VRAM | Best for | From |
|---|---|---|---|
| RTX 3070 | 8 GB | Small 7B LLMs, light inference | $0.08/hr |
| RTX 4090 | 24 GB | 13B LLMs, Stable Diffusion | $0.39/hr |
| RTX A6000 | 48 GB | 30B LLMs, 3D & video render | $0.50/hr |
| RTX 5090 | 32 GB | High-end image & video generation | $0.69/hr |
| GPU | VRAM | Best for | From |
|---|---|---|---|
| L40S | 48 GB | Image & video generation at scale | $0.90/hr |
| A100 | 80 GB | 70B-class LLM inference | $1.10/hr |
| H100 SXM | 80 GB | Large-model training & inference | $1.25/hr |
| H200 SXM | 141 GB | Frontier-scale models | contact |
A representative slice — the full catalog runs to 47 GPUs. Prices are starting rates; billing is per second.
Run AI inference jobs on hardware you already own. Workers earn credits redeemable for cash or build credits — no middleman, no markup, no auction games.
One pip command. Python 3.9+, NVIDIA CUDA or Apple Silicon. The worker is fully open source — read every line.
$ pip install dcs-worker
One-time login. The worker auto-detects your GPU, picks compatible models and sets a fair earning rate for your hardware tier.
$ dcs-worker login
Run the daemon and walk away. It heartbeats, polls for jobs, runs inference and submits results — earning while you sleep.
$ dcs-worker start
Auto-detected on registration. Rate scales with measured tokens/sec.
| Tier | Hardware | VRAM | Models | Tokens/sec | Credits/hr |
|---|---|---|---|---|---|
| Entry | RTX 3060-class | 12 GB | 7B | ~65 | 40 |
| Mid | RTX 4070 / M3 Pro | 12–16 GB | 7B · 13B | ~110 | 60 |
| High | RTX 4090 / M3 Max | 24–48 GB | up to 30B | ~180 | 120 |
| Datacenter | A100 / H100 | 40–80 GB | up to 70B | ~400+ | 240 |
1,000 credits ≈ $8.00 · redeem as build credits instantly, or queue a cash payout. A mid-tier GPU at 8 hours/day earns roughly 14,400 credits/month ≈ $115 — best-effort, varies with demand.
Five mechanics, already coded — a network built for operators, not gamblers.
Every worker gets a public 0–100 reputation badge — "98% reliable, 1,240 jobs" — so good workers stand out.
Workers that already have the right model loaded get prioritised — and paid a warm-pool credit bonus.
Reputation buckets workers into Probation → Verified → DCS Trusted. Trusted workers get a higher base rate.
No Docker, no Kubernetes, no YAML. One pip install and the worker
benches itself and starts earning.
We measure tokens-per-second on a benchmark — faster GPUs earn more, and no one can game a self-reported spec.
Convert credits to build credits instantly, or queue a cash payout — bank transfer via Stripe Connect.
Pick your hardware tier, drag the slider to how many hours per day your GPU is idle. Real per-tier credit rates from the live network.
1,000 credits ≈ $8.00. Real per-tier rates from the live network. Earnings vary with demand; redeem instantly as build credits or queue a cash payout (Stripe Connect, rolling out by region).
The Platform routes a share of its build inference through the worker network — so demand is guaranteed, not speculative.
"Front-of-chain inference" — the Platform routes its own builds through the worker network at traffic weight 0.5-0.7, so workers see real demand from day one, not synthetic load.
Transparent about what works now and what's on the way.
Workers auto-detect GPU vendor, VRAM and CUDA version. Consumer cards and Macs are first-class.
Pause on battery, set daily windows, whitelist models — the daemon honours your settings.
Credits and worker state stream to the dashboard — balance ticks up the moment a job submits.
Every line of the worker daemon is public on PyPI and GitHub — no telemetry, no hidden binaries.
The Platform routes a share of its build inference through the network — real demand, not synthetic load.
Convert credits to build credits instantly, or queue a cash payout via batch processing.
Direct bank transfer in 30+ countries, with per-region KYC and scheduled auto-conversion.
Per-member earning caps and concurrent-worker limits — for datacenter operators pooling hardware.
Each job's input and output pinned to Filecoin — workers can later prove what they produced.
Rent a GPU, or put your own hardware to work earning credits.
Compute is the power layer — these products run on top of it.
Real questions, answered straight.
dcs-worker, public on
PyPI and GitHub). It heartbeats, polls for jobs, runs inference, submits results —
no telemetry, no hidden binaries.Spin up an instance in seconds — or put idle hardware to work today.