GuideMarch 13, 2026•12 min read
AWS EC2 GPU Instances: P4, P3, G4 Complete Pricing Guide
AWS EC2 offers the most mature GPU cloud infrastructure, but navigating the P4, P3, G5, G4dn instance families can be confusing. This guide covers everything you need to know.
EC2 GPU Instance Pricing (us-east-1, 2026)
| Instance | GPU | On-Demand | Spot |
|---|---|---|---|
| p4d.24xlarge (÷8) | 1× A100 40GB | ~$4.10/hr | ~$1.23/hr |
| g5.xlarge | 1× A10G 24GB | $1.006/hr | ~$0.30/hr |
| g5.12xlarge (÷4) | 1× A10G 24GB | ~$1.42/hr | ~$0.43/hr |
| g4dn.xlarge | 1× T4 16GB | $0.526/hr | ~$0.16/hr |
| p3.2xlarge | 1× V100 16GB | $3.06/hr | ~$0.90/hr |
Savings Strategies
- EC2 Spot Instances: 70%+ savings for fault-tolerant workloads
- Compute Savings Plans: 30–60% discount, flexible across instance families
- EC2 Instance Savings Plans: Up to 72% discount for a specific instance family
- Reserved Instances: 1-year or 3-year commitment for predictable workloads
AWS vs Dedicated GPU Clouds
AWS adds a significant premium over dedicated GPU clouds:
- G5 (A10G): $1.01/hr on AWS vs $0.69/hr on RunPod (46% more expensive)
- P4d (A100): $4.10/hr on AWS vs $1.89–$2.50/hr on Lambda/Vast.ai (65–117% more)
AWS wins on: enterprise security compliance, SageMaker managed training, S3/Bedrock integration, and availability guarantees. Choose AWS when ecosystem integration justifies the cost premium.
Compare AWS vs Other GPU Clouds
See how AWS EC2 GPU pricing compares to dedicated GPU cloud providers.
Compare GPU Prices →Share this article: