For those who missed the joke, EPIC was Intel/HP's name for the Itanium ("Itanic") style of instruction set that attempts to moves most of the work for parallel instruction scheduling out of silicon and into the compiler. (And, HP later asserted that EPIC refers specifically to the Itanium instruction set, not just Itanium-like instruction sets.)

It wasn't a terrible idea at a high level, but the implementation was terribly complex and power-hungry, needing huge caches to compensate for its low instruction density. They also bet heavily on compiler advances that never materialized or materialized later than expected.

I can imagine an alternative history where Intel took the EPIC idea, but went more conservatively and focused on minimizing complexity and the total number of transistors, designing in such a way that allowed for complex power-hungry HPC optimizations later on, but didn't depend on them for the initial roll-out. This would have resulted in a lot of the potential of the idea for HPC being left on the table, but may have allowed them to have more initial success in the server market and would potentially allowed them to scale down to cell phones more easily than trying to scale Atom down to cell phones.