https://hgpu.org/?p=5632
Brief announcement: better speedups for parallel max-flow