Some Thoughts On PacBio's Computational Challenges

A few folks have commented on the Discord that the majority of the cost associated with the Revio is due to the compute requirements. This is plausible but suggests to me that compute cost has been deprioritized which respect to other factors. In particular squeezing a more accuracy out using DeepConsensus and other ML approaches.

In any case, I thought I’d try and throw some numbers around to get a rough sense of the computational challenges involved. This is complicated by the fact that raw PacBio traces have not been made available for some years now. A quick google search bring up traces from 2010 like this one: