Here is a whitepaper
It's not a whitepaper, it's a blog post.
Since it is such a large number GPU's become very slow at attempting it.
What about ASICs? This does not look like a problem that would be very hard to design ASICs for and just skip over GPUs entirely.