Data-parallel distributed training of very large models beyond GPU capacity
This addresses the memory bottleneck for researchers and practitioners training large deep learning models, though it is incremental as it builds on existing distributed training methods.
The paper tackles the problem of training very large models that exceed GPU memory capacity by introducing Large Model Support (LMS), a tool that uses high-bandwidth NVLink connections to swap tensors between CPU and GPU memory, enabling data-parallel distributed training across multiple GPUs.
GPUs have limited memory and it is difficult to train wide and/or deep models that cause the training process to go out of memory. It is shown in this paper how an open source tool called Large Model Support (LMS) can utilize a high bandwidth NVLink connection between CPUs and GPUs to accomplish training of deep convolutional networks. LMS performs tensor swapping between CPU memory and GPU memory such that only a minimal number of tensors required in a training step are kept in the GPU memory. It is also shown how LMS can be combined with an MPI based distributed deep learning module to train models in a data-parallel fashion across multiple GPUs, such that each GPU is utilizing the CPU memory for tensor swapping. The hardware architecture that enables the high bandwidth GPU link with the CPU is discussed as well as the associated set of software tools that are available as the PowerAI package.