I have a similar experience to Dodfr, but instead it is about two WDBU6Y0040BBK-0A (WD Elements 4TB) that I had the opportunity to benchmark.
The disks inside are the exact same model (WD40NMZW-11GX6S1) and the exact same firmware too. The difference is on the outside - one is the 2016 model with white painted logo and the other is the 2017 model with engraved logo. I am not sure that they are exact 2016 and 2017 models, but I have seen that is what they are called.
I suspected that they could be SMR so I ran some tests on them - a test with fio in Linux that wrote 1,5 TB randomly in a 500 GB location (see How to determine whether hard drive uses SMR - Unix & Linux Stack Exchange), and then some benchmarking with HD Tune.
Both tests gave completely different test results.
For the 2016 model I got normal(?) write speeds of 70-110 MB/s all through the test. For the 2017 model I got normal write speeds until I hit 25%, then it dropped to 0-50 MB/s. Then I thought I understood the case - the 2016 model was PMR and the 2017 was SMR. But that was before I benchmarked in HD Tune.
The 2016 model has a very SMR like curve to me, even if the average and minimum speeds looks PMR at 80 to 105 MB/s.
But this is nothing compared to the 2017 model, which clearly is an SMR even in this case with a straight graph and speeds of 220 MB/s.
I have tested a 2 TB WD Elements and it has a normal graph that falls down the closer it gets to 100%.
I can conclude that anything above 2 TB is not PMR and that the 2017 model is SMR. It acts like an SMR disk - when the unknown cache is exhausted, then it gets very slow.
But I am not sure what the 2016 model is - it did not show any speed drops with fio but it still has an SMR graph.
