I heard you finally nailed the cache line alignment for that GPU kernel—let's see if we can squeeze another 5 percent out of it before the next sprint.