The SMs are counting the Long Scoreboard Stalls on each cycle where they are waiting for their data read to come back from memory. They don’t have direct visibility into what is taking so long. So as a user, you need to look at other metrics to try and infer what is causing them. For example, if you have a lot of long scoreboard stall, you can look at the memory chart and tables to see if a specific bus is saturated or if there a lot of cache misses. Those are the types of things that could cause the stalls. But there isn’t a one-to-one mapping that allows the hardware to know what was the cause of any specific stall.