Amazon EC2 GPU instances for machine learning (2022)

Recommended GPU Instances
docs.aws.amazon.com/dlami/latest/devguide/gpu.html

1. Amazon EC2 P3

aws.amazon.com/ec2/instance-types/p3

Accelerate machine learning and high performance computing applications with powerful GPUs

Benefits:

  • Reduce machine learning training time from days to minutes
  • The industry's most cost-effective solution for ML training

On-Demand Price/hr: ≥ $3.06.

2. Amazon EC2 P4d

aws.amazon.com/ec2/instance-types/p4

Highest performance for ML training and HPC applications in the cloud

Benefits:

  • Reduce ML training time from days to minutes
  • Run the most complex multi-node ML training with high efficiency

On-Demand Price/hr: $32.77.

3. Amazon EC2 G3

aws.amazon.com/ec2/instance-types/g3

Accelerate your graphics-intensive workloads with powerful GPU instances

On-Demand Price/hr: ≥ $0.75.

4. Amazon EC2 G4

aws.amazon.com/ec2/instance-types/g4

The industry’s most cost-effective GPU instances for machine learning inference and graphics-intensive applications

4.1. Amazon EC2 G4dn

It uses NVIDIA T4 Tensor Core GPU.

The industry’s most cost-effective GPU instances for machine learning inference and graphics-intensive applications

Benefits:

  • Increase performance and reduce machine learning inference cost
  • Cost-effective small scale training

On-Demand Price/hr: ≥ $0.526.

4.2. Amazon EC2 G4ad

It uses AMD Radeon Pro V520 which is 2 times slower than NVIDIA T4 Tensor Core GPU in Amazon EC2 G4dn.

On-Demand Price/hr: ≥ $0.379.

5. Amazon EC2 G5

High performance GPU-based instances for graphics-intensive applications and machine learning inference

Benefits:

  • High performance and cost-efficiency for ML inference
  • Cost-efficient training for moderately complex ML models

It uses NVIDIA A10 Tensor Core GPU.
It 2.5 times faster than NVIDIA T4 Tensor Core GPU in Amazon EC2 G4dn.

On-Demand Price/hr: ≥ $1.006

6. Amazon EC2 G5g

aws.amazon.com/ec2/instance-types/g5g

Cost-effective machine learning inference

It uses AWS Graviton2 CPU and NVIDIA T4G Tensor Core GPU.

On-Demand Price/hr: ≥ $0.42.