A more correct headline would be: "An algorithm beats a poorly optimized GPU implementation on a narrow problem that uses an extremely sparse dataset using $8.5

A more correct headline would be: "An algorithm beats a poorly optimized GPU implementation on a narrow problem that uses an extremely sparse dataset using $8.5K worth of CPU". This is for recommenders only, and it does not translate to anything else. I don't know why everyone seems to misrepresent this as NVIDIA's undoing. Read the freaking paper, people.