LAS VEGAS — Machine learning is one of the fastest growing cloud services at the moment, and Amazon Web Services took several steps Wednesday to ensure it is keeping up with the competition.
The cloud market leader unveiled a new custom machine-learning chip called AWS Inferentia, designed specifically for inference tasks, in which machine-learning systems actually start to put patterns together after having been trained on a data set. And it plans to add GPU acceleration to its EC2 compute instances, the most basic unit of its cloud computing services.
In addition, Jassy said that AWS had developed new optimizations for the popular Tensorflow machine-learning framework that cut the training time in half compared to the Tensorflow systems developed by a company in “Mountain View,” a not-so-subtle allusion to Google. Google developed Tensorflow and later released it as an open-source project, and has made its machine-learning prowess a key selling point for its cloud services.
Specific details about the AWS Inferentia chip were not immediately available. The GPU acceleration capabilities will be known as Amazon Elastic Inference, and they will allow customers to pick the level of acceleration that makes the most sense for their applications and pay only for the performance they need, AWS said in a blog post.
Jassy also announced that AWS will launch a marketplace for machine-learning algorithms, something Google also did earlier this year and something that Seattle startup Algorithmia has been doing for a few years now. The AWS version will launch with 150 algorithms, Jassy said.