Yongmei Zhang

I've spent much of my life chasing nanoseconds. Now I'm chasing tokens, too.

I'm a GPU/CPU architect and engineering leader with nearly two decades of experience in computer architecture, hardware-software co-design, and system performance optimization.

My career has taken me from ASIC verification and performance modeling to Intel Xeon architecture, NVIDIA GPUs, and AI inference optimization. Along the way, I've worked on everything from low-level hardware behavior to system-level performance, often crossing the boundary between hardware and software. I've learned that understanding a system from multiple angles is incredibly useful, especially when the performance numbers don't quite make sense.

One thing hasn't changed throughout my career: the bottleneck is rarely where you expect it to be. As the saying goes, the devil is in the details. Sometimes a small change in architecture or design can have a significant impact on system-level performance. And sometimes the best optimization starts with simply understanding what the system is actually doing.

That's what I enjoy most about engineering. I like digging into the details, questioning assumptions, building models, and connecting ideas across different layers of the hardware and software stacks. Whether it's memory latency, GPU scheduling, coherent interconnects, or LLM inference, I'm interested in understanding not just what performs better, but why.

These days, optimizing frontier AI model performance across heterogeneous hardware has given me plenty of new questions to explore. Models are evolving rapidly, workloads are becoming more complex, and the relationship between algorithms, software, and hardware keeps changing. Techniques that worked well for traditional computing don't always translate directly to AI workloads. It's an exciting time to be working at the intersection of AI and computer architecture, particularly because there are still so many interesting problems to solve.

I started this blog as a place to share what I've learned, explore technical ideas, and occasionally challenge assumptions I once thought were settled. Some posts will dig into architecture and performance tradeoffs. Others will connect lessons from traditional CPU and GPU design to the challenges of modern AI computing. My goal is to make complex technical topics easier to understand without losing the details that make them interesting.

After all these years, I've come to appreciate that good engineering is as much about asking the right questions as finding the right answers. And just when you think you've finally understood a system, a new workload comes along to prove otherwise.

Nothing makes me happier than a good speedup. Except figuring out why it happened.

Outside of work, I'm an avid hiker. The Bay Area trails are where I do my best thinking and connect with like-minded friends.