DeepL announced on the 12th that it will fully commercialize the 'DGX SuperPOD' system based on NVIDIA's 'DGX GB200'.
Through this implementation, DeepL has improved its performance to handle the entire web translation task, which previously took 194 days, in just 18.5 days. This system marks the first introduction in Europe and will be installed and operated at an eco-data center located in Sweden.
This infrastructure expansion serves as a foundation for enhancing DeepL's generative artificial intelligence (AI) capabilities and providing a customized user experience. It aims not only to improve text-based translation but also to strengthen interactivity and broaden the application of multimodal models. DeepL users are expected to experience changes such as improved translation quality and nuance detection, faster processing speeds, and expanded generative capabilities.
Through the DGX SuperPOD, DeepL is implementing a new model training method and has expanded its overall text processing capacity to 30 times that of the transfer. For example, the translation of the entire Oxford English Dictionary has been reduced from 39 seconds to 2 seconds, while the translation of 'In Search of Lost Time' has been shortened from 0.95 seconds to 0.09 seconds.
Jarek Kutylowski, CEO of DeepL, noted, "DeepL's ability to showcase a solution with higher performance than competitors in the market is due to our continuous research and development efforts," adding, "The combination of the research infrastructure established through this SuperPOD and the latest technology will not only enhance DeepL's existing products but also enable the realization of new and groundbreaking products."