Amazon to launch new service AWS Trainium
The move lets Amazon enter the E-learning industry.
Amazon.com Inc. (Nasdaq: AMZN) has introduced AWS Trainium. This is a user-defined chip that is intended to enable cost-effective training of machine learning models in the cloud. This comes before the availability of the new Habana Gaudi-based Amazon Elastic Compute Cloud instances, designed for ML training and powered by the Gabana Gaudi processors from Intel Corp (NASDAQ: INTC). The company said that with PyTorch, TensorFlow, and MXNet support, AWS Trainium will perform better than the competition in the cloud. It will be available within Amazon SageMaker and EC2 instances. The company said instances based on the next-generation custom chips will hit the market in 2021. The advantage of custom chips is speed and cost. AWS says throughput will be 30% higher and cost per inference 45% lower than the current AWS GPU instances. Amazon claims that Trainium will deliver the most teraflops for ML instances in the cloud, with one teraflop equating to a chip that processes 1 trillion computations per second. Andy Jassy, the CEO of AWS, said the company wants to drive price performance in ML training and therefore needs to invest in its own chips. He said there is an unmatched array of instances in AWS that is accompanied by chip innovations. Andy spoke at this year's Amazon re: Invent developer conference. The new offerings will complement AWS Inferentia, which Amazon launched last year. Inferentia is an inferencing counterpart to machine learning pieces that also uses customer-specific chips. It is important to note that Trainium will use the same SDK as Inferentia. AWS noted that although Inferentia addresses inference costs, which represent approximately 90% of ML infrastructure costs, most development teams are still constrained by tight machine learning budgets. This limits the amount and frequency of training required to improve models and applications. Fortunately, AWS Trainium will address this challenge by delivering improved performance and lower cost of machine learning training in the cloud.