I hope this is the right place to ask, and thank you in advance for any advice.
I am testing a workload where Process A sends large np.array objects to Process B through a multiprocessing queue (Queue1).
The main cost is that these large arrays are copied extensively in memory across processes.
To clarify: in my observation, this is not a memory bandwidth bottleneck.
Instead, the issue seems to be one-sided CPU pressure during the copy/serialization path, which slows down throughput.
What I observed:
On Ryzen 5 6226R, this workload is noticeably slower.
On Intel i5-13400F, the same workload performs much better.
I want to understand what to look for before buying CPUs for this kind of task.
Could you please help with:
Which CPU characteristics matter most for heavy cross-process large-object copy workloads?
(for example: single-core IPC/turbo behavior, cache hierarchy, memcpy performance, inter-process transfer overhead)
Which published specs are actually useful for predicting this scenario?
Which benchmarks should I trust for this pattern (large np.array transfer via multiprocessing queue)?
If needed, I can share more details, including array size, queue usage pattern, process count, OS, and test method.
Thanks again. Any guidance is greatly appreciated.
