GPU spot market — by class
Live offers from three open compute markets — Vast.ai, Akash, Nosana — bucketed by canonical GPU class. $/h is the per-hour USD ask. volatility is the coefficient of variation (stdev/mean) of the in-window prices, expressed as a percent — higher means the market is churning. The cheapest column names the provider, region, verified flag and host reliability of the floor offer for that class.
Data: shell/landing/dash-marketwatch-latest.json,
re-written by scripts/marketwatch-report.py on
every run. Polling cadence: every 15 minutes via
the on1-marketwatch.timer systemd unit. Phase 0:
watch-only, no bids placed.
GPU class — price floor & spread
| class | n | min $/h | median | max | volatility | cheapest (provider · region · verified · reliability) | trend |
|---|---|---|---|---|---|---|---|
| loading… | |||||||
Provider summary
| provider | n offers | min $/h | median |
|---|---|---|---|
| loading… | |||
Sources: vast.ai/api/v0/bundles (rentable hosts,
no auth), akash.network on-chain provider
attributes via REST, nosana.io marketplace
endpoint. All three are public read-only; the marketwatch
module normalises GPU labels and price units before insert.
Polling cadence: every 15 minutes (on1-marketwatch.timer).
Trend column shows hourly-bucket median $/h per
provider+class pair (last 30 points).