Right, the innovation is in using storage-compute collocation to allow common data operations to be structured into the way the storage arrays are set up. The paper is really about doing sparse linear algebra in hardware not PDEs (but of course after discretization that is all PDEs are--linear algebra).