Hallo Everyone, is there a method, that I can see the Roofline of my GPU for integer operations? I have only found in GPU SOL Section the Roofline for fp16, fp32 and fp64, which is not used in my programm. Or I can only learn from section Compute Workload Analysis and Memory Workload Analysis with Pipe Utilization Percentage and different memory throughputs?
And if my kernel function is only launched with small grid(for example 1grid with 1 block), is it better to refer to analyse section warp state satatistics? Or there is still better way. Many thanks!
Does anyone have idea or I shall post more details about my situation?
There is no roofline section for integer operations.
Which version of Nsight Compute are you using?
Starting with version 2023.3 the Summary page shows the most actionable optimization advices when a result row is selected. You can follow the rule links to see more context on the Details page.
In general you can look at the recommendations shown on the Details page. If a recommendation references another report section, it will appear as a link in the recommendation. Select the link to scroll to the respective section.
This topic was automatically closed after 10 days. New replies are no longer allowed.