I hate to be the dismissive one, but I've seen a similar demo on a developer's laptop on CPU. On 32GB of RAM and 4/8 (core/thread) I have demos with billions of

I hate to be the dismissive one, but I've seen a similar demo on a developer's laptop on CPU. On 32GB of RAM and 4/8 (core/thread) I have demos with billions of rows of data that I can report on similarly. Here's a demo of Azure Analysis Services looking at similar Taxi data. Demo starts around 11:00 https://azure.microsoft.com/en-us/blog/christian-wade-shows-off-azure-analysis-services/ https://azure.microsoft.com/en-us/blog/christian-wade-shows-... Ninja edit: I missed the data load times. I have a current customer loading a (much larger than my laptop demo) 1.5B row dataset with large dimensionality into a single node SSAS instance. They can process the whole model (dump and reload data) in 1.5 hours. If I hear that 20 GPUs are necessary, I would expect multiple more orders of magnitude of data.