Ah yes, surely those 6B operations per second are very usable by the CPU and don't take a massive performance hit by syncing. It's definitely not an article to say "i let a fragment shader do very simple math and it was very fast".