The objective of IMPACT (Illinois Microarchitecture Project using Algorithms and Compiler Technology) is to provide critical research, architecture innovation, and algorithm and compiler prototypes for heterogeneous parallel architectures. We achieve portable performance and energy efficiency for emerging real-world applications by developing novel hardware, compiler, and algorithmic solutions.


Recent & Highlighted Items

Our Work PyTorch-Direct Upstreamed to AWS Deep Graph Library (March 1, 2022)

The GPU-oriented data communication architecture proposed in our previous works, "Large Graph Convolutional Network Training with GPU-Oriented Data Communication Architecture" and "PyTorch-Direct: Enabling GPU Centric Data Access for Very Large Graph Neural Network Training with Irregular Accesses," have been officially accepted by the AWS Deep Graph Library (DGL) and merged.

Our work highlights the use of the zero-copy access capability of NVIDIA GPUs to improve the data access efficiency for large sparse datasets, which is the perfect fit for large-scale graph neural network training (GNN). Our work improved GNN training speed by about 1.5x - 4.2x in various models and Single/Multi-GPU training setups. There are active discussions about further expanding our work in multiple fields of GNN training, so please visit for more information!

(View Archive of Highlighted Items)