That's sort of the point though! We're actually selling these as the "board". So the right way to compare things is sort of DGX-1 style "deep learning rig" vers

That's sort of the point though! We're actually selling these as the "board". So the right way to compare things is sort of DGX-1 style "deep learning rig" versus a board of four TPU units (or several connected). The on-chip network is a big part of its overall efficiency. I don't recall what (if anything) we've said about how we link up the boards across racks, but the folks at Next Platform looked pretty carefully at the pictures: https://www.nextplatform.com/2017/05/22/hood-googles-tpu2-machine-learning-clusters/ https://www.nextplatform.com/2017/05/22/hood-googles-tpu2-ma...