Performance Difference Between MKL Shared Objects

[StressedGradStudent]

Hi,

First time posting here, so hopefully I'm posting in the correct location. I am currently in the process of benchmarking two pieces of software against each other, one that I have implemented and another that is the current state-of-the-art. Both of use standard BLAS routines such as dgemm. The performance of both softwares is heavily reliant upon the BLAS installation that they use. I need to make sure that I am making as fair a comparison as possible by using the same BLAS installation for both. Both utilize CMake to collect and link the necessary packages. I have built both linked to MKL. However, (using ldd and grepping for mkl) I noticed that one has linked to libmkl_rt.so, whilst the other has linked to libmkl_intel_lp64.so, libmkl_intel_thread.so and libmkl_core.so.

I originally made several unsucessful attempts to build each with the same shared objects. I have spent a while going through the documentation and cannot find anything about the exact differences in performance of these shared objects. Are there any significant performance differences between them? Do I need to worry about making an unfair comparison between the two softwares in the event that I attempt to publish the results?

Thanks in advance!

[MRajesh_intel]

Hi,

Thanks for reaching out to us, we will get back to you soon.

Regards

Rajesh.

[Gennady_F_Intel]

Benjamin,

the performance could be the same on the same platform and the same inputs.

[Gennady_F_Intel]

The issue is closing and we will no longer respond to this thread. If you require additional assistance from Intel, please start a new thread. Any further interaction in this thread will be considered community only.