How much does an idle NVIDIA V100 cost in the cloud?
The lowest per-GPU rate for the NVIDIA V100 is on p3.2xlarge on AWS (us-east-1), at $3.06/GPU-hr on-demand. That whole box left running around the clock costs about $2,234/mo. Scheduling it off outside business hours saves up to $1,569/mo (est.).
Cheapest per GPU
$3.06/GPU-hr
p3.2xlarge · us-east-1
Whole instance 24/7
$2,234/mo
730-hour month
Saved with a schedule (up to, est.)
$1,569/mo
up to 70% off the 24/7 bill (est.)
Instances carrying the NVIDIA V100
| Instance | Provider | GPUs | vCPU / Memory | $/hr | $/GPU-hr | Spot $/GPU-hr |
|---|---|---|---|---|---|---|
| p3.2xlarge | AWS | 1× GPU | 8 vCPU · 61 GB | $3.060/hr us-east-1 | $3.06 | $0.43 |
| p3.8xlarge | AWS | 4× GPU | 32 vCPU · 244 GB | $12.240/hr us-east-1 | $3.06 | $0.46 |
| p3.16xlarge | AWS | 8× GPU | 64 vCPU · 488 GB | $24.480/hr us-east-1 | $3.06 | $0.40 |
About the NVIDIA V100
The V100 powered the deep-learning boom's first act, and AWS still lists it in the p3 family. New work has largely moved to newer tiers with better economics, but V100 instances persist in real accounts — pinned by old CUDA stacks, reserved-instance leftovers, and pipelines nobody wants to touch.
A legacy p3 box idling between occasional runs is the most expensive kind of nostalgia: modern hardware would finish the same job faster, and the idle hours bill at full rate either way. If a p3 is still in your fleet, the first question is how many hours a week it actually runs a job; the second is whether the workload justifies any always-on GPU at all.
Common questions
Is it worth migrating a working V100 pipeline?
If it runs rarely and works, a schedule that stops it between runs often captures most of the available savings (est.) with zero migration risk. If it runs weekly or more, newer tiers usually finish faster and cost less per completed job — benchmark one run before deciding.
What keeps V100 instances alive in 2026?
Pinned CUDA and framework versions, prepaid capacity, and the reasonable fear of touching a pipeline that works. All are legitimate; none require the box to run while idle. Stopping preserves the EBS volumes and the environment on them, so the pipeline comes back as you left it — just move anything you keep on instance-store scratch first, since that does not survive a stop.
Other GPU tiers
Stop paying for idle cloud GPU servers
Idlefy runs a free audit of your fleet, then flips dev servers to stopped-by-default: lease one when you need it, and
when the lease ends it shuts itself off — stop, not terminate, persistent disks intact. Timers are set by your team, never
guessed, and Idlefy only manages instances tagged idlefy=enabled — on AWS that boundary is enforced by the IAM policy itself.
Prices are on-demand list price for Linux (AWS) / standard (GCP) instances and exclude storage, network, data transfer, and any negotiated or volume discounts. Actual bills vary by usage and account-level pricing. Spot figures are indicative — AWS values are derived from AWS's published typical savings vs on-demand for the region; GCP values are the listed Spot price. Spot capacity can be reclaimed by the provider. Stopping preserves EBS and persistent-disk volumes; data on local instance-store or local-SSD drives does not survive a stop. Prices as of 2026-08-31.