Do they really?
What is better: 5% higher IPC or being able to clock 10% higher, within the same power budget and die area?
In case of BD, if the single threaded workload is mainly integer, it will be able to benefit from an aggressive turbo (possibly doing speculative execution with the help of the second core of the module too), if the workload is FP (and not AVX), it will be able to utilize both halfs of the FlexGPU, running two instructions in parallel, if possible. I am therefore not particualry worried about single thread performance. Yet.![]()
Bookmarks