When we talk about NVIDIA and its GPUs, the first thing that most people think of is the gaming experience they are capable of providing. And it is, of course, logical, because since its inception, more than two decades ago, graphics adapters of the brand have always been characterized by great performance, to which many other improvements have been added over the years, such as all grouped under the acronym DLSS, which initially focused only on image scaling based on artificial intelligence, but over time grouped other notable technologies.

However, as we have seen in recent years, NVIDIA GPU performance goes far beyond the gaming experience on PC to the extent that it has made the technology one of the main references in the world of artificial intelligence. Aware of this very relevant role, and to offer information about all the possibilities of AI, a few months ago they started a series of publications NVIDIA AI Decoded, in which every week they tell us about artificial intelligence applied in different fields, new uses, tools, etc. You can find at this link all our publications on said content.

So this week is dedicated to that RAPIDS cuDF, an open source library developed by NVIDIA that offers an API similar to Pandas, but redirects the workload to the GPU instead of the CPU. Pandas, if you’re not familiar with it, is a Python library widely used in data science and data analytics. Its popularity is due to the fact that it offers data structures such as DataFrames and Series, which allow to manipulate and analyze large volumes of data in a very efficient way. This has led to him becoming a reference in statistical analysis, data visualization and machine learning.

RAPIDS cuDF offers in principle the same as pandas, but with the fundamental difference that I indicated above, namely that the load is directed to the GPU, and we are talking about the type of operations in which it has already been fully demonstrated that their performance is several orders of magnitude more efficient thanks to the architecture of these integrated devices. So according to the documentation provided by NVIDIA The speed of performing certain tasks can be increased up to 100x.

We have to add to that Data science increasingly relies on artificial intelligencewhich, as I already mentioned, is another area of ​​specialization of NVIDIA GPUs, so their use translates into further improvements in performing tasks related to information obtained after processing data processed with RAPIDS cuDF.