Have you checked other benchmark results?
LinX/y-cruncher seems to count on some cycle-based timers, which might cause wrong execution time and thus wrong GFlops.
You could try y-cruncher, which run-to-run delivers a more stable peformance than LinX.
(and count realworld time against the reported execution time)
It seems that LinX doesn't scale (well) with 6-cores.
So it might result in a very unstable GFlops because of resource contention, etc.
Bookmarks