You might have noticed that some work units complete in just a few seconds on GPU. It can't be that much faster than CPU, so I investigated it. There is no bug, it validates successfully against CPU version. The reason why it's so fast on GPU is because it has slightly improved filtering algorithm compared to CPU. Usually it doesn't give a big advantage, but sometimes the advantage is extreme: CPU does 100-200 times more work than GPU. I'll update CPU version today to fix this.