Language AI company DeepL announced the deployment of an NVIDIA DGX SuperPOD with DGX GB200 systems, the first of its kind in Europe. This will enable DeepL to translate the entirety of the internet – which currently takes 194 days of nonstop processing – in just over 18 days. This deployment is now operational at DeepL’s partner EcoDataCenter in Sweden.
In the short term, users can expect immediate enhancements, including increased quality, speed and nuance in translations, along with greater interactivity and the introduction of more generative AI features. Looking to the future, multi-modal models will become the standard at DeepL. The long-term vision includes further exploration of generative capabilities and an increased focus on personalization options, ensuring that every user’s experience is tailored and unique.
Technical Advancements
This deployment will provide the additional computing power necessary to train new models and develop innovative features for DeepL’s Language AI platform. NVIDIA DGX SuperPOD with DGX GB200 systems, with its liquid-cooled, rack-scale design and scalability for tens of thousands of GPUs, will enable DeepL to run high-performance AI models essential for advanced generative applications. This marks DeepL’s third deployment of an NVIDIA DGX SuperPOD, now surpassing the capabilities of DeepL Mercury, its previous flagship supercomputer.
“At DeepL, we take pride in our unwavering commitment to research and development, which has consistently allowed us to deliver solutions that outshine our competitors. This latest deployment further cements our position as a leader in the Language AI space,” said Jarek Kutylowski, CEO and Founder of DeepL. “By equipping our research infrastructure with the latest technology, we not only enhance our existing offering but also explore exciting new products. The pace of innovation in AI is faster than ever, and integrating these advancements into our tech stack is essential for our continued growth.”
The deployment of these new racks will make DeepL faster and more accurate than ever before. Standout capabilities of the new clusters include:
– Translating the entire web into another language, which currently takes 194 days of non-stop processing, will now be achievable in just 18.5 days.
– The time required to translate the Oxford English Dictionary into another language will drop from 39 seconds to two seconds.
– Translating Marcel Proust’s In Search of Lost Time, the world’s longest book, will be reduced from 0.95 seconds to just 0.09 seconds.
– Overall, the new clusters will deliver 30 times the text output compared to previous capabilities.
“Europe needs robust AI deployments to maintain its competitive edge, drive innovation, and address complex challenges across industries” said Charlie Boyle, Vice President of DGX systems at NVIDIA. "By harnessing the performance and efficiency of our latest AI infrastructure, DeepL is poised to accelerate breakthroughs in language AI and deliver transformative new experiences for users across the continent and beyond."
