TECH

DeepL announces deployment of new NVIDIA supercomputing infrastructure

Translation technology company DeepL has announced the deployment of a new NVIDIA supercomputing infrastructure. This is the first commercial installation of a DGX SuperPOD system with DGX GB200 in Europe, located at the EcoDataCenter in Sweden.

With this new processing capacity, the German company claims that the time required to translate a volume of text equivalent to the entire internet has been reduced from 194 to 18.5 days. The investment aims to accelerate the development of more complex artificial intelligence models and expand the product offering.

The new infrastructure provides a significant increase in DeepL's processing speed. Translation of reference works, such as the Oxford English Dictionary, which previously took 39 seconds, can now be completed in 2 seconds. The current system succeeds the company's previous supercomputer, a DGX SuperPOD with H100 systems.

According to DeepL, the main objective of this investment is to provide its researchers with the computing capacity necessary to train more sophisticated language models. This infrastructure will support existing products, such as the contextual translation tool “Clarify”, and will enable exploration of new areas.

The company plans to use the new system to develop generative AI capabilities and personalization options for its users. The implementation of multimodal models, capable of processing different types of data beyond text, is one of the company’s strategic goals.

This implementation comes at a time of growth for DeepL, which in May 2024 raised $300 million in investment, reaching a valuation of $2 billion. The company, which competes with services such as Google Translate, provides technology to more than 100,000 enterprise and government customers.

The adoption of the new NVIDIA platform represents a strategic step for DeepL to increase its research and development capabilities. The investment is not only aimed at optimizing its current translation services, but also positioning the company to compete in the next phase of artificial intelligence, focused on generative and multimodal models.

mundophone