Cheapest observed hourly instance price by GPU model and provider over time. Prices are not normalized per GPU