Instance comparison
Compare hardware specs and price across accelerator instance types. Search and add any number of types; the table starts with the xlarge of each family (or the smallest size a family offers). Specs come from DescribeInstanceTypes; on-demand list price is for us-east-1; the cheapest spot is the lowest price seen across regions in the last 24h; and the 24h averages are the mean spot price and mean spot placement score (1–10) over that window across all regions.
dl1.24xlargedl2q.24xlargeg4ad.xlargeg4dn.xlargeg5.xlargeg5g.xlargeg6.xlargeg6e.xlargeg6f.xlargeg7.2xlargeg7e.2xlargegr6.4xlargegr6f.4xlargeinf1.xlargeinf2.xlargep3dn.24xlargep4d.24xlargep4de.24xlargep5.4xlargep5e.48xlargep5en.48xlargep6-b200.48xlargep6-b300.48xlargetrn1.2xlargetrn1n.32xlargetrn2.48xlargevt1.3xlarge
| Instance type | vCPUs | Memory (GiB) | Arch | Accelerator | Count | Accel memory (GiB) | Network | On-demand $/hr | Cheapest spot $/hr | Discount | 24h avg spot $/hr | 24h avg score |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| dl1.24xlarge | 96 | 768 | x86_64 | Habana Gaudi HL-205 | 8 | 256 | 4x 100 Gigabit | $13.11 | $13.11 (us-west-2) | -0% | $13.11 | — |
| dl2q.24xlarge | 96 | 768 | x86_64 | Qualcomm Qualcomm AI100 | 8 | 125 | 100 Gigabit | — | $4.68 (us-west-2) | -48% | $4.79 | — |
| g4ad.xlarge | 4 | 16 | x86_64 | AMD Radeon Pro V520 | 1 | 8 | Up to 10 Gigabit | $0.379 | $0.077 (us-east-2) | -80% | $0.153 | — |
| g4dn.xlarge | 4 | 16 | x86_64 | NVIDIA T4 | 1 | 16 | Up to 25 Gigabit | $0.526 | $0.109 (sa-east-1) | -88% | $0.252 | — |
| g5.xlarge | 4 | 16 | x86_64 | NVIDIA A10G | 1 | 22 | Up to 10 Gigabit | $1.01 | $0.218 (ap-southeast-3) | -85% | $0.593 | 1.6 |
| g5g.xlarge | 4 | 8 | arm64 | NVIDIA T4g | 1 | 16 | Up to 10 Gigabit | $0.420 | $0.075 (ap-northeast-2) | -86% | $0.154 | — |
| g6.xlarge | 4 | 16 | x86_64 | NVIDIA L4 | 1 | 22 | Up to 10 Gigabit | $0.805 | $0.228 (sa-east-1) | -83% | $0.472 | 1.6 |
| g6e.xlarge | 4 | 32 | x86_64 | NVIDIA L40S | 1 | 45 | Up to 20 Gigabit | $1.86 | $0.597 (ap-south-1) | -73% | $1.32 | 1.7 |
| g6f.xlarge | 4 | 16 | x86_64 | NVIDIA L4 | 0 | 3 | Up to 10 Gigabit | $0.237 | $0.060 (eu-north-1) | -76% | $0.137 | — |
| g7.2xlarge | 8 | 32 | x86_64 | NVIDIA RTX PRO 4500 | 1 | 32 | 60 Gigabit | — | $0.413 (us-east-2) | -84% | $0.795 | 3.0 |
| g7e.2xlarge | 8 | 64 | x86_64 | NVIDIA RTX PRO Server 6000 | 1 | 96 | 50 Gigabit | $3.36 | $1.13 (eu-south-2) | -68% | $1.92 | 2.1 |
| gr6.4xlarge | 16 | 128 | x86_64 | NVIDIA L4 | 1 | 22 | Up to 25 Gigabit | $1.54 | $0.250 (eu-south-2) | -85% | $0.547 | — |
| gr6f.4xlarge | 16 | 128 | x86_64 | NVIDIA L4 | 0 | 11 | Up to 25 Gigabit | $1.07 | $0.150 (eu-north-1) | -87% | $0.597 | — |
| inf1.xlarge | 4 | 8 | x86_64 | AWS Inferentia | 1 | 8 | Up to 25 Gigabit | $0.228 | $0.024 (eu-north-1) | -90% | $0.056 | — |
| inf2.xlarge | 4 | 16 | x86_64 | AWS Inferentia2 | 1 | 32 | Up to 15 Gigabit | $0.758 | $0.076 (us-east-2) | -90% | $0.281 | — |
| p3dn.24xlarge | 96 | 768 | x86_64 | NVIDIA V100 | 8 | 256 | 100 Gigabit | $31.21 | $4.92 (us-east-1) | -84% | $6.80 | — |
| p4d.24xlarge | 96 | 1152 | x86_64 | NVIDIA A100 | 8 | 320 | 4x 100 Gigabit | $21.96 | $7.42 (us-east-2) | -66% | $19.13 | — |
| p4de.24xlarge | 96 | 1152 | x86_64 | NVIDIA A100 | 8 | 640 | 4x 100 Gigabit | $27.45 | $9.66 (ap-northeast-1) | -74% | $23.49 | — |
| p5.4xlarge | 16 | 256 | x86_64 | NVIDIA H100 | 1 | 80 | 100 Gigabit | $6.88 | $2.63 (us-east-1) | -62% | $5.34 | — |
| p5e.48xlarge | 192 | 2048 | x86_64 | NVIDIA H200 | 8 | 1128 | 3200 Gigabit | — | $10.63 (sa-east-1) | — | $26.61 | — |
| p5en.48xlarge | 192 | 2048 | x86_64 | NVIDIA H200 | 8 | 1128 | 3200 Gigabit | $63.30 | $10.98 (ap-northeast-2) | -87% | $31.34 | — |
| p6-b200.48xlarge | 192 | 2048 | x86_64 | NVIDIA B200 | 8 | 1432 | 3200 Gigabit | $113.93 | $40.89 (us-west-2) | -64% | $43.46 | — |
| p6-b300.48xlarge | 192 | 4096 | x86_64 | NVIDIA B300 | 8 | 2149 | 6400 Gigabit | $142.42 | $44.71 (us-east-1) | -69% | $49.51 | — |
| trn1.2xlarge | 8 | 32 | x86_64 | AWS Trainium | 1 | 32 | Up to 12.5 Gigabit | $1.34 | $0.134 (us-east-2) | -90% | $0.431 | — |
| trn1n.32xlarge | 128 | 512 | x86_64 | AWS Trainium | 16 | 512 | 16x 100 Gigabit | $24.78 | $2.48 (us-east-2) | -90% | $3.40 | — |
| trn2.48xlarge | 192 | 2048 | x86_64 | AWS Trainium2 | 16 | 8192 | 16x 200 Gigabit | — | $8.60 (us-east-2) | — | $8.60 | — |
| vt1.3xlarge | 12 | 24 | x86_64 | Xilinx U30 | 1 | 24 | 3.12 Gigabit | $0.650 | $0.171 (us-east-1) | -74% | $0.435 | — |