Request for information from users with need for parallel GPU processing

As a part of the One-U Responsible AI Initiative GPU node purchase, we are considering adding a high-speed interconnect to purchased GPU nodes for faster inter-node communication. This interconnect is very important for good performance with communication-heavy algorithms, such as neural network training, for applications that use more than the eight GPUs that are present in a single node. If you have run—or are anticipating to run—in parallel on more than eight GPUs, please let us know at [email protected]. Your response is critical in informing and supporting the decision to pursue additional funding for the high-speed interconnect.