Abhinav Vishnu
About Abhinav Vishnu
I am a Fellow at AMD Artificial Intelligence Group (AIG). I work on a variety of strategic initiatives to optimize performance of Deep Learning (DL) workloads, including Training and Generative Inference, on AMD GPUs. I lead the performance modeling of DL workloads on AMD GPUs (such as MI100, MI250X, and upcoming MI300X) for AIG, where my team conducts modeling on SoC and large-scale system architecture, both AMD-specific and competitive architectures, to understand the impact Training and Generative Inference workloads. My specific interest and expertise are in large-scale systems beyond the scale of the largest Supercomputers today. The workloads include well published open-source models by hyper-scalers and partners such as Huggingface. My performance modeling work is used for cross-validation with silicon, and drive the software optimizations on customer critical workloads. The work is disseminated to customers, collaborators and by-products are usually in terms of patents and publications. A few examples are available at my home-page: https://abhinavvishnu.github.io.