Starting with Volta, Nvidia GPUs have forward progress guarantee, preventing lockups when there’s thread divergence.

That doesn’t improve the performance of a well behaved and well written compute shader. But avoiding hard hangs IMO deserves the label “improved thread divergence.”