My first few months at AMD
This year, our team received AMD's Next 5% Award for our work on GEAK, or Generative Efficient AI-Centric Kernels , which became part of AMD's larger project on autonomously optimizing LLM workloads on Instinct GPUs, called Hyperloom . I joined AMD just 10 months ago, switching from a 15-year career in speech processing across both industry and academia. When I pause and reflect on this transition to the new field that I entered without even knowing what GPU kernels were or how they drive deep learning, a couple of observations stand out. 1. Choosing the right battle matters. For the longest time, especially in the industry, I was involved in building speech-recognition systems in different languages that went into their respective products in conversational AI. Some of those experiences felt rewarding as the solutions that we developed were adopted by customers, even when the customer base was small, and even though there was no formal recognition for our work. Recently, I was...