This is an interesting fact that perhaps some did not know:
Graphics cards (GPUs) that offer ultra-realistic graphics in numerous video games, they also offer great efficiency for the operations necessary to Artificial Intelligences (AIs) and high intensity predictive tasks.
In the world of high-performance computing, computing power is measured floating point operations per second (Floating Point Operations Per Second, FLOPS). The same metric that is used to determine the performance of graphics cards, which have improved greatly and rapidly in recent years thanks to the video game industry. Companies like Google and Facebook have managed to create artificial intelligences that seem out of science fiction thanks to these advances.
Beginning in 2007, huge advances were made in video card design, with the search for high-speed 3D rendering for games that require real-time rendering. This advance provided a great side effect, incredible speeds on machine learning tasks.
A couple of years ago, we looked at how artificial intelligence AlphaGO Google managed to beat the world champion of Go, a board game of Chinese origin that has the reputation of being quite complicated and with an immense amount of strategies and possible combinations (For further reference, I leave the following link). To carry out this feat it was necessary 1202 CPUs and 176 GPUs.
Therefore we see that it is every day that the ties between companies are strengthened as Google and Nvidia, to offer advances in the field of artificial intelligence. In an entrance of nvidia blog, the case where Google required about 2000 CPUs for its brain image recognition system is detailed, however managed to recreate the performance of 2000 CPUs with just 12 GPUs.
The current project DeepMind by Google, contains an infrastructure of about 176 GPUs and ensures that it provides the equivalent performance of 29333 CPUs. An extremely efficient figure.
What does this mean for you?
For those who do not serve as AI developers or high-performance computing experts, this means that every time they acquire a new video game console or buy a new video card, they support manufacturers so that they can continue to develop new and best video cards. Additionally, as we increasingly demand video games with higher graphic quality, we provide the necessary impulse for innovation in the field of GPUs.
For us lovers of free software, it means enormous progress. The vast majority of technologies supported by AIs are Open Source, Tensorflow of Google, Big Sur from Facebook and CNTK Microsoft, to name the greats. Additionally, all of these alternatives work on Linux, forcing video card manufacturers to offer Linux support. Filling with hope all those who hope to be able to enjoy video games natively in Linux (also remember Volcano).
So those of us who want to play with 4K resolution on giant screens are atsupporting progress and technological innovation!