Something I saw many times and which is confirmed by multiple benchmarks: Xeon cpu and more generally Intel cpu targeting the server market have slower per thread performance than a CoreX cpu.
Even a $117 22nm Core i3 Ivy Bridge cpu will typically run python workloads faster than a $2000 10nm Xeon Cannon Lake cpu. And it’s not even with Turbo Boost mode enabled!
Except in the case of python (where the language doesn’t have proper multithreading support) server workloads are more multithreaded and more multiprocess than the games and workloads run by an individual which explains why they favour sacrificing single thread performance in order to have more cores.
While it’s already know that Intel and other hardware maufacturers can no longer increase performance using single core designs, what (in details) does decreasing per thread peformance for the same microarchitecture brings? Why not continue to just add less but faster core per chip for the same price?
No comments:
Post a Comment