If it's anything like the 3060 hashrate limiter it shouldn't affect general CUDA or OpenCL workloads, just if it specifically detects it as Dagger/ethhash. It could be different this time, but I would think if they're going to throttle all CUDA/OpenCL they would have said that as part of the announcement.
I finally heard what pattern is blocked here: lots of accessing big arrays in main video memory at random locations. Which is hardly efficient even without LHR. No application would use that pattern unless they had no other choice.