You can spin up a Cortex cluster on a variety of AWS instance types. If you are unsure about which instance to pick, review these options as a starting point. This is not a comprehensive guide so please refer to the full documentation on AWS for more information.
T3 instances are useful for development clusters that primarily run model inferences with low compute and memory resource utilization.
- Example: iris classification with scikit-learn
- Compute: low
- Memory: low
- GPU: no
- AWS cost: starts at $0.0052 per hour*
C5 instances are useful for clusters that primarily run model inferences with medium compute and low memory resource utilization.
- Example: language identification with fastText
- Compute: medium
- Memory: low
- GPU: no
- AWS cost: starts at $0.085 per hour*
M5 instances are useful for clusters that primarily run model inferences with low compute and memory resource utilization.
- Example: MPG estimation with scikit-learn
- Compute: low
- Memory: medium
- GPU: no
- AWS cost: starts at $0.096 per hour*
G4 instances are useful for clusters that primarily run model inferences with high compute and low memory resource utilization that can run on GPUs.
- Example: image classification with Inception
- Compute: high
- Memory: medium
- GPU: yes
- AWS cost: starts at $0.526 per hour*
P2 instances are useful for clusters that primarily run model inferences with high compute and high memory resource utilization that can run on GPUs.
- Example: text generation with GPT-2
- Compute: high
- Memory: high
- GPU: yes
- AWS cost: starts at $0.900 per hour*
* On-demand pricing for the US West (Oregon) AWS region.