Lower performance in V4 compared to V3

[mtarabkhah]

Hi,
I am studying quantum simulation as part of my PhD. Recently, I compared the performance of the newest version of the QuEST with its previous version, and the V3 seems to be faster (at least on the multi-threaded CPU execution). Is this normal, or am I doing something wrong?
I have prepared a benchmark for this comparison using a QFT circuit:
https://github.com/mtarabkhah/QuEST_V4_vs_V3/tree/main

Here are the system details for the machine on which I ran the simulations:

• Hardware:
  CPU: AMD Ryzen 9 5950X (16 cores / 32 threads)
  Architecture: x86_64
  Threads per core: 2
  Total logical CPUs: 32
  RAM: 62 GB

• Operating System:
  Ubuntu 24.04.1 LTS

• Compilers:
  gcc 13.3.0 (Ubuntu 13.3.0-6ubuntu2~24.04)
  g++ 13.3.0 (Ubuntu 13.3.0-6ubuntu2~24.04)
  clang 18.1.3 (Ubuntu build)

As you can find in the run.sh, I am running both versions using the provided cmake, with GPU turned off, multithreading turned on, and the number of CPUs set to 8.

Please let me know if you need further information.

It would be great if you could check this and let me know your opinion.
Thanks a lot in advance, and looking forward to hearing from you.
Best,
Meisam

[otbrown]

Hi @mtarabkhah,

Not wholly unexpected -- see some of the relevant history in #631 and #638. Actually I'm pleased to see that your plots start to converge at 20 qubits!

I see you're also at UoE -- I'm in Bayes, and happy to chat in person if it's easier.

Cheers,
Oliver