A little hard to get too excited about something proprietary that I'll likely never see in the field. But it is cool. Voltron Data definitely seems to have a lock on where things are going, show up everywhere that's of any interest. Nimble/Alpha? Consulting there too. They're just everywhere, clearly highly respected, and even if their own stuff is proprietary they see that the ecosystem needs good interchanges & protocols for themselves to succeed, which is a not-bad trickledown from this amazing shop.
It's a bit drier & more abstract, but there was a talk on Velox Wave, which is Meta's attempt at doing parallelization & segment offload, strategies for targeting GPUs. Somewhere - I think maybe Q&A or close to the end - they admit the current implementation is built around Cuda, although most of the query planning & strategies could hypothetically be applied generally. https://www.youtube.com/watch?v=ACyhL9rdv-s#t=28m48s
They had a VelxoConf talk that definitely is impressive as heck in terms of numbers. https://www.youtube.com/watch?v=CXxDNWrdEyk
It's a bit drier & more abstract, but there was a talk on Velox Wave, which is Meta's attempt at doing parallelization & segment offload, strategies for targeting GPUs. Somewhere - I think maybe Q&A or close to the end - they admit the current implementation is built around Cuda, although most of the query planning & strategies could hypothetically be applied generally. https://www.youtube.com/watch?v=ACyhL9rdv-s#t=28m48s