IMO, this in this case, where if GPUPI is not allowed to evolve, eventually we end up going the way of an old ham radio club that did not evolve to digital. (no postcards please ). Make a shift date, freeze boints for versions before 3.3 and use 3.3 going forward.
It is inevitable that benchmarks will need to evolve with emerging architectures and platforms - old benches will not run new instruction sets and so on. But, there's two sides to this argument. IMO, keeping ancient "GPU" benchmarks like aquamark, MK03 etc, is off-kilter also - these simply are not GPU benchmarks anymore. Then, we have no AVX512 benchmark with boints either... etc, so plenty of opportunities to add benches that push modern hardware too. Seems a shame that more efficient coding (3.3 solves the same problem) is being "shat upon"
Lastly, try to keep it "us against the hardware" .