How much does an idle NVIDIA A10G cost in the cloud?
The lowest per-GPU rate for the NVIDIA A10G is on g5.xlarge on AWS (us-east-2), at $1.01/GPU-hr on-demand. That whole box left running around the clock costs about $734/mo. Scheduling it off outside business hours saves up to $516/mo (est.).
Cheapest per GPU
$1.01/GPU-hr
g5.xlarge · us-east-2
Whole instance 24/7
$734/mo
730-hour month
Saved with a schedule (up to, est.)
$516/mo
up to 70% off the 24/7 bill (est.)
Instances carrying the NVIDIA A10G
| Instance | Provider | GPUs | vCPU / Memory | $/hr | $/GPU-hr | Spot $/GPU-hr |
|---|---|---|---|---|---|---|
| g5.xlarge | AWS | 1× GPU | 4 vCPU · 16 GB | $1.006/hr us-east-2 | $1.01 | $0.42 |
| g5.2xlarge | AWS | 1× GPU | 8 vCPU · 32 GB | $1.212/hr us-east-2 | $1.21 | $0.40 |
| g5.12xlarge | AWS | 4× GPU | 48 vCPU · 192 GB | $5.672/hr us-east-2 | $1.42 | $0.55 |
| g5.4xlarge | AWS | 1× GPU | 16 vCPU · 64 GB | $1.624/hr us-west-2 | $1.62 | $0.54 |
| g5.8xlarge | AWS | 1× GPU | 32 vCPU · 128 GB | $2.448/hr us-east-2 | $2.45 | $0.76 |
| g5.16xlarge | AWS | 1× GPU | 64 vCPU · 256 GB | $4.096/hr us-east-2 | $4.10 | $0.90 |
About the NVIDIA A10G
The A10G is AWS's mid-tier workhorse in the g5 family: 24 GB of VRAM aimed at inference, graphics, and small-scale training that outgrows a T4. It became the default "GPU dev box" for a generation of ML engineers — capable enough to prototype almost anything, cheap enough that nobody escalates its cost.
That default status is exactly why it leaks money: g5 instances are personal in a way training clusters are not. One engineer, one box, one forgotten Friday — multiplied across a team. The per-box waste looks modest next to a training node, but dev boxes outnumber training nodes ten to one, and they idle in the same pattern: nights, weekends, and the weeks after a project wraps.
Common questions
A10G or L4 for inference work?
They sit at the same 24 GB of VRAM; the L4 is the newer, more efficient part, while the A10G carries more raw throughput on some workloads. On AWS the two families offer the same instance ladder, from single-GPU xlarge through the four-GPU 12xlarge to the single-GPU 16xlarge, so the honest tie-breaker is your own benchmark against the current per-hour rates — plus the fact that the L4 is also sold on Google Cloud as the g2 series, if that is where your data already lives.
What is the cheapest way to run a personal GPU dev box?
Stop paying for the hours you sleep. A dev box that runs only during working sessions costs roughly a third of an always-on one (est., schedule-dependent) on the identical hardware. Personal boxes are the hardest to police because each has exactly one owner and no shared schedule — which is why the stop has to belong to the box, not to the person.
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.