8× B300
p6-b300.48xlarge · us-east-1
- Whole instance / hour:
- $142.42
- 24 billed hours:
- $3417.98
Catalog price · capacity not verified. Confirm stock and account quota before renting.
Price observed .
Compare B300 across providers →Compare AWS GPU catalog pricing and reported stock. 19 GPU models from $0.23/hr.
Hyperscaler with the broadest GPU catalog and region list; enterprise procurement, no public stock API. We track 19 GPU models across 910 configurations. AWS lists hourly on-demand rates; the provider publishes no stock feed, so availability shows as unknown. Egress is $0.09/GB after the first 100 GB free each month — budget for it on data-heavy jobs.
Estimate a job’s total cost on AWS (egress + billing increment) →
SourceGPU Finder does not currently receive a usable stock feed for AWS. You can compare listed prices, but we cannot verify current stock or publish a stock-history percentage. Confirm availability with the provider before renting.
Compare the lowest tracked on-demand whole-instance price for each GPU below, with a 24-hour compute budget. Match the instance size and region to your workload before comparing providers. AWS catalog prices do not confirm capacity or your account quota.
p6-b300.48xlarge · us-east-1
Catalog price · capacity not verified. Confirm stock and account quota before renting.
Price observed .
Compare B300 across providers →p6-b200.48xlarge · us-east-1
Catalog price · capacity not verified. Confirm stock and account quota before renting.
Price observed .
Compare B200 across providers →p5e.48xlarge · us-east-2
Catalog price · capacity not verified. Confirm stock and account quota before renting.
Price observed .
Compare H200 across providers →p5.4xlarge · us-east-1
Catalog price · capacity not verified. Confirm stock and account quota before renting.
Price observed .
Compare H100 across providers →p4d.24xlarge · us-east-2
Catalog price · capacity not verified. Confirm stock and account quota before renting.
Price observed .
Compare A100 across providers →g6e.xlarge · us-east-1
Catalog price · capacity not verified. Confirm stock and account quota before renting.
Price observed .
Compare L40S across providers →g6.xlarge · us-east-1
Catalog price · capacity not verified. Confirm stock and account quota before renting.
Price observed .
Compare L4 across providers →g4dn.xlarge · us-east-1
Catalog price · capacity not verified. Confirm stock and account quota before renting.
Price observed .
Compare T4 across providers →USD compute estimates use the unrounded whole-instance hourly rate × 24 billed hours, including setup and idle time. Displayed amounts are rounded to cents. Storage, egress, extra networking, software licences and taxes may cost extra. This is a catalog budget, not a reservation, a completion-time estimate or a guaranteed quote for your region. Confirm the final price, billing terms and capacity with the provider.
AWS on-demand pricing and quotasAWS G6 and fractional G6f specifications
The guide quotes the whole EC2 instance, including every GPU in that size. For example, an 8-GPU instance billed at $40/hour costs $960 for 24 billed hours, even though its normalized rate is $5/GPU-hour. Budget the entire instance; dividing by eight does not make one GPU separately rentable.
No. These are tracked catalog prices, not live stock reports. Check the exact EC2 size and region, your account's G/P instance-family vCPU quota, and capacity before launching. A price observation time is not a stock-check time. Unknown capacity does not mean sold out.
AWS G6f offers fractions of an NVIDIA L4, with only a share of its memory and compute. GPU Finder's whole-GPU comparison excludes these sizes instead of rounding a fraction to zero or one GPU. A fractional L4 price is not the cost of renting a full L4.
Start with GPU memory per card, software compatibility, region and the smallest instance that fits. L4, T4 and L40S can suit smaller inference or graphics jobs; H100, H200 and Blackwell instances suit larger memory or multi-GPU workloads. Check your actual model and benchmark rather than assuming a faster GPU finishes proportionally sooner. Inferentia and Trainium use AWS Neuron and are not drop-in CUDA GPU rentals.
In GPU Finder's current AWS data, single-GPU on-demand pricing starts at $0.23/hr and runs up to $10.99/hr; the full table lists 910 configurations, including larger multi-GPU nodes priced above that range. This catalog also includes AWS Inferentia and Trainium accelerators; its floor is not necessarily an NVIDIA GPU price. Use the model cards for NVIDIA-specific budgets.
Yes. AWS spot accelerator rows appear when the source catalog reports them; the lowest visible AWS spot accelerator price is currently $0.02/hr for 1x Inferentia. Spot prices vary and capacity can be interrupted; the 24-hour cards use on-demand rates instead.
Yes. This page includes AWS B300 pricing when it appears in the live catalog, along with any AWS H100, H200, and B200 rows currently available.
Page headline prices reflect the default offers shown before expanding these filters.
Is AWS the cheapest option? The cheapest cloud GPU prices per hour rank every GPU model by its lowest per-GPU-hour rate and the cheapest listing in stock, across all providers. For the most-compared training GPU, see H100 cloud pricing across every provider.