Today, we are proud to announce a major upgrade to LightOn Cloud service, now available for the AI community worldwide. LightOn Cloud 2.0 features a substantially increased capacity with our latest-generation Aurora 1.5 OPUs, the latest version of LightOnML library and simpler pay-per-use payment, booking, and support processes. LightOn technology is now deployed in two data centers through partnerships with France’s largest cloud providers, OVHcloud and Scaleway.
LightOn Cloud 2.0 users will now have access to increased power and flexibility for the creation of their machine learning models, combining LightOn OPU with a high-end Intel CPU and a V100 NVIDIA GPU. In the past few months, LightOn Cloud users have built a portfolio of use caseswith the OPU technology, highlighting its advantage on several neural network architectures. Such examples include Natural Language Processing, Computer Vision, Computational Chemistry or Reinforcement Learning. Most of these examples are explained in blog posts on our Medium with code directly accessible through our LightOn AI Research public GitHub.
In such highly-demanding problems, typical training speedups are x8 to a whopping x40 compared to GPU only, at similar accuracies: results are obtained in minutes instead of hours, making the LightOn Aurora OPU the perfect technology for data or architecture exploration.
At LightOn, we support research through the LightOn Cloud for Research Program. Machine learning researchers working in academia or non-profit organizations can apply for free LightOn Cloud access. In particular, any use of LightOn’s technology on COVID19 related problems will be prioritized — The LightOn AI Research team has already demonstrated how the OPU can be used to detect conformational changes in coronavirus-related HPC simulations. You can find the article here.
LightOn Cloud 2.0 demonstrates that a hybrid approach, featuring beyond-pure-silicon technology, is already here to enhance large scale AI workloads while at the same time lowering data centers’ power consumption.
In order to provide the best quality of service for our users while allowing for the scalable growth of LightOn Cloud and attendant technology, applications to access the commercial service and the Research Program are required. Interested Data scientists and Machine Learning engineers can apply for LightOn Cloud here, and for LightOn Cloud for Research Program here.