I've been able to port NR to GPU's which with sufficiently powerful hardware can run simulations at about ~30fps with raytracing, to simulate binary black hole collisions. You need something around a top end consumer gpu at the moment. Phone hardware needs a while to catch up, there's an absolute minimum memory requirement of ~8gb vram, and you need a lot more bandwidth than they currently support

Awesome! Is it published anywhere? All the stuff I'm familiar with is aimed at old-school clusters.