This session is part of a long research program aimed at scrutinizing how different components of server hardware and software stack affect deep learning model’s training time and cost. Even a very simple set up where one needs to choose a cloud VM with a single GPU secondary parameters (like RAM, CPU performance, disks, etc.) might have a significant impact on the cost and duration of the training and this impact will vary from one neural network architecture to another.
CEO and Founder, RocketCompute
Redwood City, California United States
Modern Enterprise IT
Performance Engineering and DevOps