- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I have been using ifort version 19.1.3.304 and just updated to OneAPI ifort version 2021.3.0
and ifx version 2021.3.0 Beta. Doing some speed test of my production code that uses openmp I find that ifort classic is running much faster than the ifx.
ifort 19.1.3.304 and ifort 2021.3.0 run at exactly the same speed:
real 153m58.297s
user 3150m35.557s
sys 15m18.230s
However, ifx runs at:
real 237m0.759s
user 4655m44.994s
sys 20m52.255s
The compiler flags are -O3 -xHost -qopenmp for both runs. This is on dual cpu dell precision 7920 with Intel(R) Xeon(R) Gold 6256 CPU @ 3.60GHz. The library paths for both compilations point to the same place (is this correct?) in my makefiles. Am I doing something wrong or does the beta version have some debug features etc that is slowing the runs down. Thanks!
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
For this kind of comparison VTune profiler should have been used. The timing results as posted by you are too abstract to reveal the exact reason of apparent much worse performance of the code compiled by the 'ifx'.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Yes, naturally, but I found the following statement from Intel:
"ifort will remain Intel’s recommended production compiler until ifx has performance and features superior to ifort. In the future, when ifx functionality and performance equivalence or better against ifort is achieved, Intel may suggest using ifx, instead of, ifort for performance critical applications."
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
So the Intel answer posses some degree of clarification. More in-depth "digging" (as I stated) would require a profiler and combination of sampling and counting mode.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
The critical words in the text you quoted are "until" and "In the future". At present, ifx is an early beta compiler whose performance is definitely NOT superior to ifort. Feel free to try ifx, but ifort remains the production compiler of choice for now.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
The two Intel® oneAPI Toolkits, the Base & HPC Kits, deliver what developers need to build, analyze, optimize, and scale HPC applications.
Join Intel engineers Ron Green and Mike Lee for a tour of these development tools, including:
- How it offers expanded support for CPUs, GPUs, and FPGAs (this will be rolled into an overview of the oneAPI initiative)
- How existing Intel Parallel Studio users can easily transition to the new suite, including a step-by-step walkthrough of the process
- A detailed overview of Intel’s latest compiler technology for C/C++ and Fortran
- A showcase of the major new XPU development capabilities available through Intel oneAPI Toolkits
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page