The evolution of the cloud has made available to any company the Artificial Intelligence engines of technological giants like IBM, Google or Microsoft. We can easily integrate a virtual assistant into our website, because the data processing needed to make it work is in the cloud. Any solution that uses machine learning must manage two distinct phases, with very different processing needs.
The first phase of training that needs large infrastructures and calculation models that are effectively carried out by GPU-based systems. Once the system has been instructed, and the model is executed, we pass to the inferential AI phase in which the request is processed to obtain the most effective result.
Qualcomm Cloud AI 100: designed for inferential processing in the AI environment
In Qualcomm’s vision, inference processing in ambio AI needs different characteristics than those offered by systems based on a combination of CPU and GPU present in today’s data centers.
Leveraging on the experience accumulated on over 1 billion installations of Snapdragon platforms for smartphones, Qualcomm has created Cloud AI 100, a dedicated component to be integrated into data centers and designed specifically to optimize performance per watt in inferential processing solutions for the Artificial intelligence.
Qualcomm Cloud AI 100 is based on a specifically designed chip, with a 7nm production process that allows to obtain performance per watt 10 times higher than the solutions adopted at this time. Furthermore, the new platform natively supports the most used software stacks in the AI environment such as PyTorch, Glow, TensorFlow, Keras and ONYX.
In Qualcomm’s forecasts, sampling to Cloud AI customers will start in the second half of 2019.