P2 instances contain high-performance NVIDIA K80 GPUs, each with 2,496 CUDA cores and 12 GB of GPU memory. There are three models of P2, as described in the following table:
Model |
GPUs |
vCPU |
Memory (GB) |
GPU Memory (GB) |
p2.xlarge |
1 |
4 |
61 |
12 |
p2.8xlarge |
8 |
32 |
488 |
96 |
p2.16xlarge |
16 |
64 |
732 |
192 |
These models with large GPU memory are best suited for training models. With more GPU memory, we can train the model with a larger batch size and a neural network with lots of parameters.