Saturn Cloud enables easy and powerful Nvidia GPU computing for everyone
Joshua Patterson of NVIDIA
Saturn Cloud, the GPU data science company, announces the launch of Saturn Cloud Hosted, an open platform for multi-node multi-GPU computing in the cloud. The platform is available to anyone and everyone: corporations, startups, academics, students, and hobbyists.
This new platform enables data scientists to spin up an Nvidia GPU instance and run their Python code from JupyterLab, and even launch clusters of GPUs with Dask, a Python-native parallel computing framework. Saturn Cloud Hosted comes pre-loaded with images for popular deep learning frameworks like TensorFlow and PyTorch, as well as Nvidia RAPIDS, an open GPU data science framework that allows users to accelerate array, dataframe, graph, and stream processing.
“We have seen scientific progress limited by insufficient computing power and downsampled data. We believe that democratising the high-speed computing capabilities of cloud-hosted GPUs will enable researchers to work with the fullest volume of scientific data, at the fastest runtime speeds possible, delivering more experiments and results for scientific exploration. The future will be less waiting, and more data,” says Sebastian Metti, one of the founders of Saturn Cloud.
Historically, GPU computing required the purchase of hardware or complex cloud solutions. With the new Saturn platform, anybody can scale their models with endless GPUs in the cloud at the click of a button.
“Speed is essential to enabling data scientists to analyse the vast and growing amounts of information created every day,” says Joshua Patterson, senior director of Rapid Engineering at Nvidia. “With RAPIDS and Nvidia accelerated computing on Saturn Cloud Hosted, data scientists can access the tools and performance they need to do their best work.”
Recent benchmark publications with Nvidia GPU-accelerated data science have stoked much excitement in the community, having shown up to 2,000x faster model performance over leading tools such as vanilla Python and Spark. This remarkable speedup paves the way for the next generation of fast computing, largely owing to the GPU acceleration enabled by Nvidia and the RAPIDS project.
An early notable study, which sparked further research, was published by Mike McCarty, director at Capital One. The benchmark study demonstrated a 100x faster XGBoost model, enabling more experiments and vastly superior model accuracy. “With these improvements, we have seen roughly 100x improvement model training times and costs have gone down 98%,” writes McCarty in his analysis.
In addition to Saturn Cloud Hosted, the company offers Saturn Cloud Enterprise, which includes all capabilities in Saturn Cloud Hosted plus virtual private cloud hosting, state-of-the-art security, and additional enterprise functionality. The two products are priced by usage, as low as $0.01 (€0.0085) per GB/RAM an hour on GPUs.
“The breakthroughs provided by Saturn Cloud only come around once every few years,” says Kris Skrinak, global machine learning segment technical lead at Amazon Web Services. “They provide an essential machine learning platform to accelerate workflows for data scientists, practitioners, and anyone who understands the value of data.” Users of Saturn Cloud include Mount Sinai Health System, Trimark, numerous AI/ML startups, hedge funds, and many others.
To try out Saturn Cloud Hosted, click here.
Comment on this article below or via Twitter @IoTGN