Seymour Cray's accomplishments
The Univac 1103 was one of his early accomplishments. And the CDC 6600 certainly was an impressive accomplishment for its day.
But the CDC 6600 was just a big computer - the same way that the NORC, the LARC, and the STRETCH before it were big computers. IBM, with the 360/91 and then the 360/195, significantly improved on Cray's design - so the Pentium and its descendants largely follow the lead of the 360/195. Some say that the 6600 was the first out-of-order machine, even if IBM, with the 91, later perfected that, but the STRETCH also was out-of-order to a limited extent.
The Cray I, however, was a unique design; its use of vector registers meant that it far outclassed previous attempts at vector-oriented supercomputers. The design was widely copied, for example for add-ons to the Univac 1110, the IBM 3090, and the VAX.
So I think that the Cray I stands as where he made the most original and greatest contribution to the science of computers.
As for the Roadrunner and similar machines - thanks to Moore's Law, eventually specialized hardware got beat by just throwing large numbers of commodity microprocessors at the problem. Although good interconnects between the many parallel processors are valuable, and require design and engineering expertise - Cray himself, just before he died, tried his hand at participating in this new era - the credit for this type of computer taking over lies in the gradual improvement of the capability of microprocessor chips, not primarily some ingenuity on the part of supercomputer designers, at IBM or anywhere else.