Your question is Transformer Memory vs Compute. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
Why are transformer models often memory-bound rather than purely compute-bound on modern GPUs?
Asked in the Onsite System Design / ML Fundamentals stage. Exploring GPU bottlenecks and transformer performance. Reported follow-ups: How do kernel fusion, kernel launch overhead, and attention matrix materialization affect memory bandwidth and GPU utilization?