Trn1 instances are suitable for deep learning training, which require a lot of hardware GPU resources, and ensure high performance to run tasks efficiently. Compared to previous solutions, you can reach the same efficiency in machine learning by deploying a smaller number of EC2 instances. Trn1 instances use Trainium chips developed by Amazon. Up to 16 AWS Trainium accelerators with support for wide scaling options. The high EFA networking throughput is up to 800 Gbit/s. Features of this instance family allow you to resolve more difficult artificial intelligence (AI) training tasks compared to the previous families of typical GPUs.
Amazon wants to crush Intel Xeon and AMD Epyc with new Graviton2 chips
2ff7e9595c
Comments