Intel Corp (INTC)

[chipguy]

No, you still have to use MPI, etc. to access the parallelism of KNL. You're still using standard(ish) tools/libraries to do it, but "old" code doesn't just convert automatically to massively parallel.

Many big customers are already running their MPI based (or whatever)
parallel codes on Xeon clusters.

Going to KNL means using a different version of the threading and
synchronization libraries and a bit of fine tuning. That is way less
effort and than rewriting to CUDA and verifying the code still works.