This is one model we've explored. There are a few like this. What we're finding is that it comes down to distribution, really. Also, tuning GPUs to work 'conse

This is one model we've explored. There are a few like this. What we're finding is that it comes down to distribution, really. Also, tuning GPUs to work 'conservatively' is a non-trivial engineering challenge.