AWS Inferentia instance prices: AWS vs Azure
4 instance types have AWS Inferentia GPUs, all on AWS EC2. Per GPU, the cheapest is inf1.xlarge on AWS at $0.228 per GPU-hour on demand (1 GPU, $0.228 per hour).
Every AWS Inferentia option, cheapest first
| Cloud | Type | vCPUs | Memory (GiB) | GPUs | Per GPU /hr | On-demand /hr | 1-yr reserved /hr | Spot /hr |
|---|---|---|---|---|---|---|---|---|
| AWS | inf1.xlarge | 4 | 8 | 1 | $0.228 | $0.228 | $0.144 | $0.039 |
| AWS | inf1.2xlarge | 8 | 16 | 1 | $0.362 | $0.362 | $0.228 | $0.0862 |
| AWS | inf1.6xlarge | 24 | 48 | 4 | $0.295 | $1.18 | $0.744 | $0.1807 |
| AWS | inf1.24xlarge | 96 | 192 | 16 | $0.2951 | $4.721 | $2.974 | $0.616 |
Other GPUs
AMD alveo U250 fpga 64 gbAMD instinct MI25 16 gbAMD radeon pro V520AMD radeon pro V620 32 gbAMD radeon pro V710aws INFERENTIA2aws trainiumNVIDIA A10NVIDIA A10 24 gbNVIDIA A10GNVIDIA A100NVIDIA H100NVIDIA H200NVIDIA L4NVIDIA L40SNVIDIA pcie A100 80 gbNVIDIA pcie H100 94 gbNVIDIA rtx pro 4500NVIDIA rtx pro server 6000NVIDIA T4NVIDIA T4GNVIDIA tesla K80NVIDIA tesla M60NVIDIA tesla M60 16 gbNVIDIA tesla T4 16 gbNVIDIA tesla V100NVIDIA tesla V100 16 gb
On-demand Linux prices in USD per hour, in each cloud's home region: US East (N. Virginia) for AWS, East US for Azure. Previous-generation types are left out.