2 ms·

I haven't personally used the Intel compilers in about a decade. I agree that they are very finicky. When I did regularly build code with them, my group would s

I haven't personally used the Intel compilers in about a decade. I agree that they are very finicky. When I did regularly build code with them, my group would stick with a particular point-release for months at a time. Even minor revisions could cause previously stable calculations to fail. But, like many others, we kept using Intel's troublesome optimizing compilers because execution time is very dear when you run compute-intensive code from several hours to a few days per run, sometimes on hundreds or thousands of CPUs.

gnufx 8y ago
Take it from current (HPC) experience/measurements that the big performance advantage of the Intel compilers is mostly mythical, but the unreliability of ifort isn't. (ifort is infinitely slower than GCC on most architectures, two of which are currently relevant to HPC.) "Just use GCC" solves quite a lot of problems, but people who don't understand the code and have measurements tend to insist ifort should be used even when the code won't build/run with it; really! It's quite likely that differences in compiled code quality are irrelevant anyhow, particularly considering communication with thousands of cores, and there are aspects like debugging, mentioned elsewhere.