I was looking at this today & yesterday. Boding well, these results look right for GPUs because they're in parity w/ Duane Merril's GPU BFS work. Less clear is

I was looking at this today & yesterday. Boding well, these results look right for GPUs because they're in parity w/ Duane Merril's GPU BFS work. Less clear is whether MapGraph beats existing single node HPC systems. In particular, instead of comparing to GraphLab, I'd love to see comparisons to GraphChi or TurboGraph! The importance here is that huge graphs are now possible on single-node, and once these techniques get combined into distributed ones, most real-world graphs can be solved in real-time! (If you like building this kind of bleeding edge stuff in performance, ML, and/or infoviz, we're hiring at graphistry.com :)