Changelog

November 2020

• Updated performance tables to include numbers from 20.10-tf1-py3 NGC container

June 2020

• Updated performance tables to include A100 results

April 2020

• Improved Spark preprocessing scripts performance

March 2020

• Initial release

Known issues

• Limited tf.feature_column support
• Limited scaling for multi-GPU because of inefficient handling of embedding operations (multiple memory transfers between CPU and GPU), work in progress to cover all the operations on GPU.
• In this model the TF32 precision can in some cases be as fast as the FP16 precision on Ampere GPUs. This is because TF32 also uses Tensor Cores and doesn't need any additional logic such as maintaining FP32 master weights and casts. However, please note that W&D is, by modern recommender standards, a very small model. Larger models should still see significant benefits of using FP16 math.