Amazon has unveiled its new custom chip called Trinium. This chip is supposed to enter EC2 instances by the beginning of the year and improve machine learning processes.
According to Tekna technology news service, Amazon’s annual conference was held with a group of company developers, during which Amazon’s new chip called AWS Trainium was unveiled.
Amazon’s new generation of custom chips is going to be used in the field of training models based on machine learning. According to Amazon, this chip can have more processing power in the cloud industry. Trinium chip is also able to support TensorFlow, PyTorch and MXNet.
The main purpose of releasing this chip is to use it in EC2 instances. Also, this chip will be placed in SageMaker, Amazon’s machine learning platform. The main advantage of this chip is its high processing speed and reasonable price.So that it can show 30% more throughput than standard GPU-equipped INSTs in AWS. Also, its cost will be 45% less per inference. New Amazon instances will be launched next year, and Amazon seems to be working closely with Intel on instances based on the Habana Gaudi platform.
