@@ -112,7 +112,7 @@ <h4>GPU queues</h4>
112112workload. In addition you can see if your application is using barriers
113113efficiently, allowing the queues to run their workloads in parallel.</ p >
114114< h3 > Profiling GPU memory bandwidth</ h3 >
115- < p > GPUs are a data-plane processors, so memory access efficiency is an important
115+ < p > GPUs are data-plane processors, so memory access efficiency is an important
116116factor for overall performance.</ p >
117117< p > < img alt ="5th Generation memory system " src ="./images/bifrost-memory-system.svg " /> </ p >
118118< p > Memory system performance outside of the GPU cannot be directly observed via
@@ -184,10 +184,10 @@ <h3>Profiling workload</h3>
184184context of API constructs, such as vertices, triangles, and pixels making it
185185easier for developers to understand the feedback.</ p >
186186< p > < img alt ="5th Generation shader core " src ="./images/valhall-shader-core.svg " /> </ p >
187- < p > Supplementing the workload size counters, the Arm GPU also provide counters
188- that indicate areas where content is not following best practice guidelines.
187+ < p > Supplementing the workload size counters, Arm GPUs also provide counters that
188+ indicate areas where content is not following best practice guidelines.
189189Improving these best practice metrics will nearly always improve your
190- application's performance.</ p >
190+ application's performance or energy efficiency .</ p >
191191</ div >
192192< div class ="lgc-section ">
193193< h2 id ="s_gpufrontend "> GPU Front-end < span data-bs-toggle ="tooltip " data-bs-delay ="400 " data-bs-placement ="right " title ="Copy to clipboard "> < a class ="lgc-link-copy link-secondary "> < i class ="fa-solid fa-link "> </ i > </ a > </ span > </ h2 >
0 commit comments