Amazon SageMaker provides a selection of instance types optimized to fit different machine learning (ML) use cases. Instance types comprise varying combinations of CPU, GPU, memory, and networking capacity and give you the flexibility to choose the appropriate mix of resources for building, training, and deploying your ML models. Each instance type includes one or more instance sizes, allowing you to scale your resources to the requirements of your target workload.
NOTE: Please refer to the actual pricing page for details on the support for each instance type in a particular AWS region and for a particular SageMaker function such as building, processing, training, and deployment of ML models. Not all instances are supported in all AWS regions and/or for all SageMaker functions.
Instance type |
vCPU | GPU | Mem (GiB) | GPU Mem (GiB) | Network Performance |
|---|---|---|---|---|---|
Standard – Current Generation |
|||||
ml.t2.medium |
2 | - | 4 | - | Low to Moderate |
| ml.t2.large | 2 | 8 | Low to Moderate | ||
| ml.t2.xlarge | 4 | 16 | Moderate | ||
| ml.t2.2xlarge | 8 | 32 | Moderate | ||
| ml.t3.medium | 2 | 4 | Low to Moderate |
||
| ml.t3.large | 2 | 8 |
Low to Moderate | ||
| ml.t3.xlarge | 4 | 16 | Low to Moderate | ||
| ml.t3.2xlarge | 8 | 32 | Low to Moderate |
||
| ml.m5.large | 2 | 8 | High | ||
| ml.m5.xlarge | 4 | 16 | High | ||
| ml.m5.2xlarge | 8 | 32 | High | ||
| ml.m5.4xlarge | 16 | 64 |
High | ||
| ml.m5.12xlarge | 48 | 192 | 10 Gigabit | ||
| ml.m5.24xlarge | 96 | 384 | 25 Gigabit |
||
ml.m4.xlarge |
4 | - | 16 | - | High |
| ml.m4.4xlarge | 16 | - | 64 | - |
High |
| ml.m4.10xlarge | 40 | - | 160 | - |
10 Gigabit |
| ml.m4.16xlarge | 64 | 256 | 25 Gigabit | ||
| ml.m5d.large | 2 | - | 8 | Up to 10 Gbps | |
| ml.m5d.xlarge | 4 | - | 16 | - | Up to 10 Gbps |
| ml.m5d.2xlarge | 8 | - | 32 | - | Up to 10 Gbps |
| ml.m5d.4xlarge | 16 | - | 64 | - | Up to 10 Gbps |
| ml.m5d.8xlarge |
32 | - |
128 | - | 10 Gbps |
| ml.m5d.12xlarge | 48 | - | 192 | - | 10 Gbps |
| ml.m5d.24xlarge | 96 | - | 384 | - | 25 Gbps |
| Memory Optimized - Current Generation |
|||||
| ml.r5.large | 2 | - | 16 | - | Up to 10 Gbps |
| ml.r5.xlarge | 4 | - | 32 | - | Up to 10 Gbps |
| ml.r5.2xlarge | 8 | - | 64 |
- | Up to 10 Gbps |
| ml.r5.4xlarge | 16 | - | 128 | - | Up to 10 Gbps |
| ml.r5.12xlarge | 48 | - | 384 | - | 10 Gbps |
| ml.r5.24xlarge | 96 | - | 768 | - | 25 Gbps |
| ml.r5d.large | 2 | - | 16 | Up to 10 Gbps | |
| ml.r5d.xlarge | 4 | - | 32 | - | Up to 10 Gbps |
| ml.r5d.2xlarge | 8 | - | 64 | - | Up to 10 Gbps |
| ml.r5d.4xlarge | 16 | - | 128 | - | Up to 10 Gbps |
| ml.r5d.8xlarge | 32 | - | 256 | - |
10 Gigabit |
| ml.r5d.12xlarge | 48 | - | 384 | - | 10 Gigabit |
| ml.r5d.16xlarge | 64 | - | 512 | - | 20 Gigabit |
| ml.r5d.24xlarge | 96 | - | 768 | - |
25 Gigabit |
| Compute Optimized – Current Generation | |||||
| ml.c5.large | 2 | 4 | Up to 10 Gbps | ||
| ml.c5.xlarge | 4 | - | 8 | - | Up to 10 Gbps |
| ml.c5.2xlarge | 8 | - | 16 | - | Up to 10 Gbps |
| ml.c5.4xlarge | 16 | - | 32 | - | Up to 10 Gbps |
| ml.c5.9xlarge | 36 | - | 72 | - | 10 Gigabit |
| ml.c5.18xlarge | 72 | - | 144 | - | 25 Gigabit |
| ml.c5d.xlarge | 4 | 8 | Up to 10 Gbps | ||
| ml.c5d.2xlarge | 8 | 16 | Up to 10 Gbps |
||
| ml.c5d.4xlarge | 16 | 32 | Up to 10 Gbps |
||
| ml.c5d.9xlarge | 36 | 72 | 10 Gbps |
||
| ml.c5d.18xlarge | 72 | 144 | 25 Gbps |
||
| ml.c4.large | 2 | 3.75 | Moderate | ||
| ml.c4.xlarge | 4 | - | 7.5 | - | High |
| ml.c4.2xlarge | 8 | - | 15 | - | High |
| ml.c4.4xlarge | 16 | 30 | High | ||
| ml.c4.8xlarge | 36 | - | 60 | - | 10 Gigabit |
| - | - | ||||
| Accelerated Computing – Current Generation | |||||
| ml.p3.2xlarge | 8 | 1xV100 | 61 | 16 | Up to 10 Gbps |
| ml.p3.8xlarge | 32 | 4xV100 | 244 | 64 | 10 Gigabit |
| ml.p3.16xlarge | 64 | 8xV100 | 488 | 128 | 25 Gigabit |
| ml.p3dn.24xlarge | 96 | 8xV100 | 768 | 256 | 100 Gigabit |
| ml.p2.xlarge | 4 | 1xK80 | 61 | 12 | High |
| ml.p2.8xlarge | 32 | 8xK80 | 488 |
96 | 10 Gigabit |
| ml.p2.16xlarge | 64 | 16xK80 | 732 | 192 | 25 Gigabit |
| ml.g4dn.xlarge | 4 | 1xT4 | 16 | 16 | Up to 25 Gbps |
| ml.g4dn.2xlarge | 8 | 1xT4 | 32 | 16 | Up to 25 Gbps |
| ml.g4dn.4xlarge | 16 | 1xT4 | 64 | 16 | Up to 25 Gbps |
| ml.g4dn.8xlarge | 32 | 1xT4 | 128 | 16 | 50 Gbps |
| ml.g4dn.12xlarge | 48 | 4xT4 | 192 | 64 | 50 Gbps |
| ml.g4dn.16xlarge | 64 | 1xT4 | 256 | 16 | 50 Gbps |
| ml.inf1.xlarge | 4 | 8 | Up to 25 Gbps | ||
| ml.inf1.2xlarge | 8 | 16 | Up to 25 Gbps | ||
| ml.inf1.6xlarge | 24 | 48 | 25 Gbps | ||
| ml.inf1.24xlarge | 96 | 192 | 100 Gbps |
| Accelerator | Throughput in FP-32 TFLOPS* |
Throughput in FP-16 TFLOPS** | Memory |
| eia2.medium | 1 | 8 | 2 GB |
| eia2.large | 2 | 16 | 4 GB |
| eia2.xlarge | 4 | 32 | 8 GB |
| eia1.medium | 1 | 8 | 1 GB |
| eia1.large | 2 | 16 | 2 GB |
| eia1.xlarge | 4 |
32 | 4 GB |
* FP-32 TFLOPS = trillion 32-bit floating operations per second
** FP-16 TFLOPS = trillion 16-bit floating operations per second
Learn more about Amazon SageMaker pricing