GridNexa
/ Compute Access — GPU Co-build

Turn idle GPUs into
steady host earnings

Platform Preview — Windows and Ubuntu node clients are Coming Soon. When ready, download the package, bind your Grid ID, pass security checks, and join the planned shared compute pool. Host listing share (proposed 80% during pilot) and settlement follow published pilot terms — not live GA inventory.

Compute Access is in Platform Preview. Windows and Ubuntu node clients are Coming Soon. When enabled, download, bind a provider token, and join scheduling under controlled rollout. Listing earn (proposed 80% host share) and settlement follow pilot terms — not live GA inventory.

How it works

Three steps to go online

Download client

Windows desktop zip or Ubuntu one-line install (Coming Soon). Auto-detect GPU, VRAM, and network to create a node profile.

Bind Grid ID

Paste your provider token from Account → Developer, complete hardware and security checks (~3 min).

Go online

Join the planned scheduling pool. Sandboxed jobs run on demand; GPU time meters per second toward listing earnings (Platform Preview).

Node client

Platform Preview · Coming Soon

Preparing
Windows client
Windows 10 / 11 · 64-bit · NVIDIA driver ≥ 550
  • GUI panel: node status, uptime, and pilot earnings
  • Do-not-disturb: pause scheduling when fullscreen apps are detected
  • Autostart and recovery after power loss
SHA-256 · size · version shown when ready
Preparing
Ubuntu / Linux client
Ubuntu 20.04+ / Debian · systemd daemon · x86_64
  • One-line curl install with systemd services
  • Headless daemon + optional AI worker (CUDA)
  • Provider token bind · auto-reconnect after reboot
One-line install
# Install script unavailable — download the Ubuntu package below, unzip, then run sudo ./join.sh

Run with sudo. The installer prompts for a provider token, or pass --token YOUR_GRID_TOKEN. Optional: --with-aigc on a prepared CUDA host.

SHA-256 · headless · Docker optional
Pilot requirements ≥ 100 Mbps symmetric bandwidth (recommended) NVIDIA driver ≥ 550 / CUDA ≥ 12.4 Spare PSU headroom ≥ 150 W Public IP or NAT traversal for inbound jobs
Scheduling

Jobs run only when your node is online and idle. Pause anytime from the client; unfinished work migrates to other nodes.

Host pricing

Reference GPU catalog · proposed metering

Reference market bands (USD) · pilot listing earn · not live inventory

Host share
80%
Of reference market rate
Billing
Per-second
Only while scheduled & running
Est. utilization
65%
For daily estimates · not a guarantee

Reference estimate only: third-party market rate × proposed 80% host share. The daily illustration assumes 65% utilization; neither demand, utilization, rate nor earnings is guaranteed. Metering and settlement are planned and subject to pilot terms.

0 / 0 models

Loading GPU catalog…

* Listing earn references public Community Cloud market bands for supported GPU types during Compute Access Pilot v1. Actual rates, share, and settlement follow pilot terms and vary with demand and uptime.

FAQ

Compute Access Pilot v1

Can I download the node client now?

Not yet for general download. Windows desktop (.zip) and Ubuntu install packages are marked Coming Soon on this page during Platform Preview. When a build is published, sign in, open Account → Developer to create a provider token, then bind it in the client.

How are earnings calculated?

Proposed pilot listing earn ≈ reference market rate × 80% host share (platform retains 20%). This is a planned pilot term, not a live production guarantee. Metering is per-second only while a job is scheduled and running. Settlement follows published pilot terms when Compute Access is enabled.

Does unstable network hurt earnings?

Uptime ≥ 95% keeps full scheduling priority. Below 90% may pause new jobs; accrued earnings for completed work are not reversed.

Can I leave anytime?

Yes. Go offline from the client. Incomplete jobs migrate; earnings for completed GPU time still accrue under pilot rules.