Overview
Strengths & Best For
AWS EC2 P5 instances are powered by NVIDIA H100 Tensor Core GPUs (up to 8× H100, 640 GB HBM3, 3,200 Gbps EFA networking). P5e and P5en instances use NVIDIA H200 GPUs (up to 8× H200, 1,128 GB HBM3e). P5en adds Gen5 PCIe and third-generation EFA for 35% lower latency. All three families deploy in EC2 UltraClusters scaling to 20,000 GPUs with petabit-scale nonblocking networking delivering 20 exaflops aggregate. SageMaker HyperPod manages resilient training clusters at any scale. Amazon Bedrock provides serverless access to DeepSeek v3.2, MiniMax, Kimi 3, Z.AI GLM 5, and Writer Palmyra. 30+ global regions with enterprise SLAs.
- P5/P5e/P5en: H100 and H200 in UltraClusters up to 20,000 GPUs
- 20 exaflops aggregate compute in EC2 UltraClusters
- 3,200 Gbps EFA networking with GPUDirect RDMA
- SageMaker HyperPod for resilient distributed training
- Amazon Bedrock: serverless access to DeepSeek, MiniMax, Kimi, GLM, Palmyra
- 30+ global regions with enterprise SLAs and compliance certifications
- Highest on-demand GPU rates among all providers
- Complex pricing — on-demand vs reserved vs spot vs savings plans
- GPU quota increases require support tickets and can take days
No reviews yet — add yours ↗
OpenRelay offers on-demand cloud GPU instances and an OpenAI-compatible inference API, billed hourly with no commitment, quota, or reservation required.
No reviews yet — add yours ↗
Live GPU Pricing
Region Coverage
Popular Comparisons
Provider pages
More AWS comparisons
Key differences
AWS
P5 (H100), P5e/P5en (H200), UltraClusters up to 20,000 GPUs — hyperscale AI infrastructure
OpenRelay
On-demand cloud GPUs and OpenAI-compatible inference, billed hourly with no commitment, quota, or reservation.