Calor Docs.
Everything you need to rent NVIDIA GPUs on Calor, pay in SOL or Calor, and run training, inference and rendering workloads — settled on Solana.
Introduction
Calor is an on-demand GPU compute network settled on Solana. You rent bare-metal NVIDIA GPUs — H100, H200, A100 and RTX — by the hour, pay in SOL or the Calor token, and connect over SSH or Jupyter with full root access.
Every rental is a transparent, on-chain transaction. There are no cloud accounts, no opaque monthly invoices and no reserved-instance lock-in. You pay only for the GPU-hours you rent, and each job that settles in Calor triggers an on-chain SPL burn.
- Bare-metal GPUs across five regions, provisioned in seconds.
- Hourly billing, fully on-chain — $5 minimum per rental.
- Pay in SOL, or redeem
Calorfor protocol-subsidized rates up to 85% below market. - Full root over SSH or Jupyter, with a persistent
/workspacevolume.
Quickstart
From wallet to a running GPU in four steps.
- Connect your wallet. Link Phantom, Solflare or Backpack. Your rentals, balances and Calor rewards are tied to your Solana wallet address.
- Pick a GPU. Choose a model and region from the live catalog. You see the market rate and the Calor rate side by side, billed per hour ($5 minimum).
- Pay & provision. Settle in SOL on Solana (devnet during beta; mainnet at launch). Your instance spins up automatically in seconds.
- Connect & run. Add your SSH key or open the web terminal and start training, inference or rendering. Stop any time — billing finalizes on-chain.
Core concepts
GPU node
A single bare-metal NVIDIA GPU instance with dedicated VRAM, CPU and NVMe storage. Nodes are not shared or virtualized — you get the whole card.
Region & routing
Nodes live in five regions. Jobs route to the nearest idle cluster to minimize latency; you can also pin a specific region.
Hourly billing
You choose a rental duration up front and pay for it in a single transaction (a $5 minimum applies per rental). The instance runs for that window; when it expires it is automatically shut down.
Calor, burn & staking
Calor is the network's utility token. Redeeming it for compute applies a protocol subsidy and burns a portion on every job (deflation). Staking Calor grants priority routing to high-demand clusters.
Wallets & authentication
Calor authenticates with your Solana wallet — there are no passwords or email logins for compute. Supported wallets:
- Phantom — browser extension & mobile
- Solflare — browser extension, mobile & hardware
- Backpack — browser extension & mobile
You sign a one-time message to prove ownership of your address; Calor never has custody of your keys or funds. Payments are explicit transactions you approve in your wallet.
Renting a GPU
Open the catalog, filter by model or region, and review the live availability and per-hour rates. When you confirm:
- You approve a SOL (or Calor) transaction in your wallet for the estimated duration.
- The node provisions in seconds and enters the
runningstate. - You connect, run your workload, and terminate when done.
- Final cost settles on-chain; any unused prepaid time is reconciled.
A job moves through these states: queued → provisioning → running → completed (or failed).
Supported GPUs
Live catalog with per-hour rates. The Calor rate reflects the protocol subsidy available at token launch.
| GPU | Architecture | VRAM | FP16 TFLOPS | Avail | Cloud avg $/hr | With Calor |
|---|---|---|---|---|---|---|
| H100 PCIe | Hopper | 80 GB HBM3 | 3,958 | 8 | $2.76 | $0.41 |
| H100 SXM5 | Hopper | 80 GB HBM3e | 3,958 FP8 | 4 | $1.84 | $0.28 |
| H200 SXM5 | Hopper+ | 141 GB HBM3e | 4,915 | 2 | $3.26 | $0.49 |
| A100 SXM4 | Ampere | 80 GB HBM2e | 2,496 | 12 | $0.39 | $0.06 |
| A100 PCIe | Ampere | 40 GB HBM2 | 1,248 | 8 | $0.62 | $0.09 |
| RTX 4090 | Ada Lovelace | 24 GB GDDR6X | 1,321 | 24 | $0.19 | $0.03 |
Regions
Five regions, with jobs routed to the nearest idle NVIDIA cluster.
| Region | Location | Nodes | Status |
|---|---|---|---|
| US East | Ohio | 12 | • online |
| US West | N. California | 8 | • online |
| EU West | Paris | 9 | • online |
| AP South | Mumbai | 5 | • online |
| US Central | Iowa | 4 | • online |
Pricing & billing
Compute is billed per hour for the duration you select, with a $5 minimum per rental. No reserved commitments and no charges after your window ends.
Example: an H100 PCIe at a live market rate of $1.84/hr for 4 hours costs about $7.36 in SOL — settled and visible on-chain. Short rentals are subject to the $5 order minimum. At token launch the same card runs far cheaper at the subsidized Calor rate.
What you pay for
- GPU-hours at the live model rate
- Outbound bandwidth above the included allowance
- Persistent storage beyond the instance lifetime (optional)
Paying: SOL & Calor
Two ways to pay, both settled on Solana (devnet during beta, mainnet at launch):
Pay in SOL
Approve a SOL transaction per rental. Fast (~400ms) finality, no account top-ups required.
Redeem Calor
Redeeming Calor applies the protocol subsidy (up to 85% below market) and triggers an SPL burn of a portion of the tokens spent — supply shrinks as the network does real work. Stakers additionally get priority routing.
SSH & web terminal
Every instance gives you full root access. Add your public SSH key at creation, or open the in-browser terminal — no key required.
# Add your public key when creating the instance, then connect:
ssh root@h100-eu-west-3.calor.run -p 22
# Or open the in-browser web terminal from the dashboard — no key needed.
# Every instance has full root access and a persistent /workspace volume.SDK reference
The @calor/sdk abstracts the entire protocol — submit a job, pay in Calor, and the burn happens automatically on completion.
Install
# npm
npm install @calor/sdk
# or with pnpm / yarn
pnpm add @calor/sdk
yarn add @calor/sdkSubmit a job
import { Calor } from "@calor/sdk";
const calor = new Calor({
wallet: process.env.SOLANA_WALLET_ADDRESS,
cluster: "devnet", // mainnet-beta at launch
});
// Submit a GPU compute job
const job = await calor.compute({
gpu: "nvidia-h100-pcie-80gb",
script: "./train_model.py",
duration: 3600, // seconds
});
// Calor settles & burns on completion — automatic
console.log(`Job: ${job.id}`);
console.log(`Cost: ${job.cost} Calor`);
console.log(`Burned: ${job.burned} Calor`);
console.log(`TX: ${job.burnTx}`);Track status
// Poll a running job and stream its status
const job = await icpx.jobs.get("job_8f2Kw...");
console.log(job.status); // "provisioning" | "running" | "completed" | "failed"
console.log(job.gpu); // "nvidia-h100-pcie-80gb"
console.log(job.region); // "eu-west-3"
console.log(job.costSol); // 0.0142
console.log(job.icpxBurned); // 0.0034
// Subscribe to live status changes
icpx.jobs.watch(job.id, (e) => {
console.log(`[${e.ts}] ${e.status} — ${e.message}`);
});REST API
Prefer raw HTTP? Every SDK call maps to a REST endpoint. Authenticate with a wallet-signed bearer token.
| Method | Endpoint | Description |
|---|---|---|
| POST | /api/jobs | Create a GPU rental job (gpu, region, sshKey, duration). |
| GET | /api/jobs/:id | Fetch a job, its status, cost and burn data. |
| GET | /api/jobs | List your jobs, filterable by status and region. |
| DELETE | /api/jobs/:id | Stop a running instance and finalize billing. |
| GET | /api/gpus | Live GPU catalog: models, regions, availability, rates. |
| GET | /api/network | Live network stats: nodes, utilization, PFLOPS, burn. |
| GET | /api/rewards/stakes | Read your staked Calor and reward accrual. |
| GET | /api/marketplace/listings | Browse open compute listings on the network. |
Calor token
Calor is engineered for utility and scarcity:
- Compute credits — redeem for H100/A100 time at protocol-subsidized rates.
- Burn-on-use — every GPU job permanently burns Calor via the SPL burn instruction; deflation accelerates with usage.
- Staking priority — staked Calor grants priority routing to high-demand clusters.
- DAO governance — holders vote on upgrades, fees and provider onboarding on-chain.
Staking & priority
Stake Calor to earn a share of network rewards and to move to the front of the queue for scarce GPUs. During high demand, staked wallets are routed to available H100/A100 clusters before non-stakers. Unstaking is permissionless; rewards accrue per epoch.
Security
- Key custody — Calor never holds your private keys or seed phrase. You approve every payment in your own wallet.
- Instance isolation — nodes are dedicated bare metal, wiped between tenants.
- SSH — access is key-based; you control which keys are authorized.
- On-chain audit — every rental and burn is a public, verifiable Solana transaction.
Rate limits & quotas
API requests are rate-limited per wallet. Concurrent-instance and per-region quotas scale with your usage history and staked balance. Hitting a limit returns 429 Too Many Requests with a Retry-After header. Need higher limits? Reach out to our team for enterprise onboarding.
Troubleshooting
My job is stuck in provisioning
Capacity in the chosen region may be momentarily exhausted. Remove the region pin to let the job route to the nearest idle cluster.
SSH connection refused
Confirm the instance is running and that the public key you added matches your local private key. Use the web terminal as a fallback.
Transaction failed
Ensure your wallet has enough SOL for the rental plus network fees, then retry. Failed transactions are never charged.
FAQ
What can I run on an Calor GPU?
Anything that runs on NVIDIA hardware — LLM training and fine-tuning, real-time inference, rendering, simulation and batch compute.
Do I need an account?
No. Your Solana wallet is your identity. Connect, pay, and run.
When does Calor launch?
Token generation follows the community beta and DEX listing. Browse GPUs today to rent at market rate, with discounted Calor rates activating at launch.
Is my data secure?
Instances are dedicated bare metal, wiped between tenants, with key-based SSH and on-chain billing.
Support
Questions, higher quotas, or enterprise onboarding? Browse the live GPU catalog to get started, and our team is available for onboarding details and a direct support channel.