ИИ процессоры, серверы и суперкомпьютеры для бизнеса

2021. Amazon Unveils New Machine Learning Chip for AWS



Amazon unveiled the Trn1 chip, specifically designed for deep learning tasks. Observers say this product will directly compete with Nvidia chips. The manufacturer expects training machine learning models using Trn1 to be 40% cheaper than on competing platforms. Instances will offer network throughput of up to 800 Gbps and can be used to form clusters of tens of thousands of systems. Trn1 instances are available in preview. Despite developing its own chips, AWS continues to work closely with Intel, AMD, and Nvidia. Brown stated that his company is making every effort to maintain competition in the segment by offering consumers a choice of processor platforms.


2021. Google is launching a new generation of its AI chips



Google unveiled the next generation of its Tensor Processing Units (TPUs) AI chips. This is the fourth generation, and according to Google, it is twice as fast as the previous version. As Google CEO Sundar Pichai noted, these chips are then combined into blocks with 4096 v4 TPUs. A single unit delivers over one exaflop of computing power. Google uses these AI chips to power many of its own machine learning services, but it will also make this latest generation available to developers as part of its Google Cloud platform. While others, including Microsoft, have opted to use more flexible FPGAs for their machine learning services, Google has bet early on on these custom chips. They take a bit longer to develop and quickly become obsolete as technology changes, but they can deliver significantly better performance.