GPU server documentation: account to root shell

GPU Rent Hub rents dedicated bare-metal GPU servers by the week or the month, paid in Bitcoin, Ethereum, USDT, USDC, Monero, Solana or Litecoin, with no identity verification, and this documentation covers the whole path: account, deposit, deploy, SSH, storage, renewals and the v2 API. From nothing to a root shell on a dedicated GPU in about five minutes. The only things you will ever type into our site are a username, a password and an SSH public key.

1. Create an account: pick a username and a password. That's all.
2. Open Billing → Add funds, pick a coin, send at least $20 equivalent to the address shown.
3. After the network confirms, your balance shows the USD value. Open Deploy, pick GPU, region, term, template. Paste your SSH public key.
4. Wait for the status to turn green (usually under 90 s). ssh root@<ip>.

Account & recovery

Accounts are identified by username only. We store a PBKDF2-HMAC-SHA-256 with 600,000 iterations and a per-user salt hash of your password and nothing else about you. No identity check happens at sign-up or later, which is the whole model behind no-KYC GPU hosting.

Recovery options (optional)

• Recovery email — used only to send a reset link. Never used for anything else. Stored hashed; we can send to it but not read it back out.
• PGP public key — if set, resets are delivered as an encrypted message you decrypt yourself. Also used to sign quotes and invoices on request.
• TOTP two-factor — any standard authenticator app. Recovery codes are shown once.

Deposits & balance

Your balance is a USD ledger. Each deposit gets a unique address. The USD value is fixed at the spot rate (median of three exchanges) at the moment of the first network confirmation.

Withdrawals of unused balance go to an address you provide, in the coin you originally deposited, minus the network fee. Processed within 24 h. The settlement rules per coin are set out on paying for GPU rental with crypto, and the node and confirmation policy for XMR on Monero GPU hosting.

How do I deploy a GPU instance?

From the dashboard, or with a single API call. DFW-1 Dallas, IAD-1 Ashburn and PDX-1 Hillsboro are equivalent in price and features; pick the one nearest your users from the latency figures on the three US datacenters page. Stock shown is live; if the configuration is sold out you can join the waitlist without a deposit. The call below deploys one H100 SXM at $1,690 a month.

# create a 1× H100 SXM in DFW-1 for one month, PyTorch template
curl -s https://api.gpurenthub.com/v2/instances \
  -H "Authorization: Bearer $GPURENTHUB_TOKEN" \
  -d '{"gpu":"h100sxm","count":1,"region":"dfw","term":"month",
       "template":"pytorch-2.7-cuda12.8","ssh_key_id":"key_7Fq2","name":"train-01"}'

Templates

Templates are plain OS images with drivers and a framework preinstalled. Nothing phones home; there is no agent. You can also boot any image via a URL (qcow2 or raw) or netboot your own installer.

How do I connect to a GPU instance over SSH?

Password login is disabled on every template. Add one or more public keys under SSH keys; they're written to /root/.ssh/authorized_keys at provisioning. Keys added later can be pushed to running instances from the dashboard.

ssh -i ~/.ssh/gpurenthub [email protected]
nvidia-smi --query-gpu=name,memory.total,power.draw --format=csv

If you lock yourself out, the dashboard has a browser console (VNC over WebSocket, TLS) with a one-shot rescue boot that mounts your NVMe read-write.

Networking

• One public IPv4 and a /64 IPv6 per instance. Reverse DNS editable.
• 100 Gbps port, 10 Gbps guaranteed for single-GPU instances, dedicated 100G pair on 8-way nodes.
• Inbound unmetered and scrubbed for volumetric DDoS. Outbound 20 TB/month included, then $4/TB.
• All ports open by default. Firewall is yours to configure — nftables ships with a permissive ruleset.
• Private VLANs between your instances in a region: $25/mo, included on 8-way nodes.

Storage

Local NVMe is encrypted with LUKS using a per-instance key held in the host's TPM. Releasing the instance destroys the key; the drive is additionally blkdiscarded before reuse. Network volumes are replicated three times inside the region and survive instance release. Snapshots are stored encrypted with the same per-account key.

Renewals & grace period

1. T−24 h: renewal amount is drawn from your balance. If the balance is insufficient, you get a dashboard banner (and an email or PGP message if you configured one).
2. T+0: if unpaid, the instance is stopped. Nothing is deleted.
3. T+72 h: end of grace. Instance is released to inventory but the NVMe stays intact.
4. T+7 d: NVMe key destroyed, drive wiped. Snapshots and network volumes are unaffected.

GPU rental API reference

REST, JSON, bearer tokens created under API in the dashboard. Base URL https://api.gpurenthub.com/v2. Every dashboard action is available here; the dashboard is itself an API client.

Rate limit 600 requests/minute per token. Errors are RFC 9457 problem documents. Webhooks (instance state, deposit confirmed, renewal failed) are signed with HMAC-SHA256.

CLI

curl -sL https://get.gpurenthub.com | sh          # single static binary, signed
ib login                                       # paste an API token
ib catalogue --region dfw
ib deploy --gpu rtx5090 --count 2 --term week --template comfyui --key key_7Fq2
ib ssh comfy-box

Tor mirror

The dashboard and API are reachable as an onion service; the address and its signature are published under Settings → Tor once you are logged in, and on our PGP-signed contact page. Deposits and deployments work identically over Tor. We don't treat Tor traffic differently, and we don't log it differently — we don't log it at all.

Limits

• New accounts: 8 GPUs across all instances until the first renewal has been paid, then 64. Beyond that, open a ticket — it's usually approved same day.
• Balances: no maximum. Withdrawals over $50,000 equivalent are processed in tranches over 72 h.
• Instances per region: 32 by default.
• Outbound port 25 is closed by default and opened on request.