topic Just an update: I'm currently in Intel® oneAPI Math Kernel Library

SGEMV and SGEMM performance degradation on Xeon E5-2609

Harr__Jacob — Wed, 03 Jul 2019 16:03:56 GMT

Hi,

We have noticed a performance problem on a particular CPU (Xeon(R) CPU E5-2609 0 @ 2.40GHz) that we have narrowed down to a change in either SGEMV or SGEMM (or both) that occurred somewhere between MKL 11.2 Update 2 (Product build 20150120) and MKL 2018.0 Update 1 (Product build 20171007) on Windows. The slowness (relative to 11.2) still exists in MKL 2019.0 Update 4. We are using the 32-bit versions of the libraries.

Is there any known change or issue in MKL that could account for this behavior? If so, can anything be done about this? Any amount of help would be appreciated.

Thanks!

-Jake

Which exact CPU type is

Gennady_F_Intel — Thu, 04 Jul 2019 03:15:34 GMT

Which exact CPU type is talking about? is that one of this CPU is referred to this intel.ark.page?- https://ark.intel.com/content/www/us/en/ark/search.html?_charset_=UTF-8&q=+E5-2609

What is the typical problem sizes where do you see this regression?

Could you switch on MKL_VERBOSE mode and give us the output?

Just an update: I'm currently

Harr__Jacob — Mon, 08 Jul 2019 17:30:30 GMT

Just an update: I'm currently working on getting some of the info that you requested. I didn't want to leave you without a response in the meantime.

The CPU is the oldest on that page (without a "v" suffix).

Thanks for the help!

The problem sizes vary, but

Harr__Jacob — Tue, 09 Jul 2019 18:05:46 GMT

The problem sizes vary, but the most common operation is SGEMM with 1280x1280 and 1280x9 matrices.

Still have to wait on getting MKL_VERBOSE output on the troublesome CPU.