What keeps billing when the GPU stops?
Check storage and transfer alongside compute. A cheap hourly GPU rate does not tell you the full bill.
A stopped disk still has a bill.
RunPod Volume Disk · one month at the published state rate.
100 GB held for one month: $10.00 running, or $20.00 idle. Volume disk only.
GB × $0.10 running, or GB × $0.20 idle. Excludes GPU time, other storage and taxes. Not a quote.
Source ↗H100 SXM5 resource price. CPU and memory billed separately; not a dedicated-instance total.
Source ↗Physical core = 2 vCPU · minimum 0.125 cores per container. Resource rates, not total request cost.
Source ↗Vast marketplace. No exact live offer captured. Check GPU, host, location and both bandwidth rates.
Source ↗Instances. Storage rates vary by host; stopping is not deletion.
Source ↗Instance host rates. Check both rates in the selected offer.
Source ↗Paperspace Machines in legacy console. GPU form factor and memory not specified on price page. Do not equate with GPU Droplets.
Source ↗Paperspace Machines. Hourly add-on charges continue until resources are destroyed. Non-GPU storage and public IP resources have maximum monthly charges.
Source ↗Paperspace Machines. Do not transfer this claim to DigitalOcean Droplets.
Source ↗H100 SXM · 80 GB VRAM · up to 24 pCPU and 240 GB RAM per GPU. Billed per minute. Maximum CPU/RAM, not a guaranteed chosen flavor.
Source ↗Pricing page other-services traffic. Read alongside selected service and contract; storage and public IP separate.
Source ↗HGX H100 · 16 vCPU · 200 GB RAM in the listing. Region-specific live stock and total quote unknown.
Source ↗Shared Filesystem by Nebius. WEKA, object and block storage have separate schedules.
Source ↗Object storage Standard egress. Networking ingress/egress is separately listed free. Do not apply that to this storage tier.
Source ↗Persistent disks and shared disks respectively. Separate from compute; no complete total or tax estimate.
Source ↗