@blueberry can confirm that [org.bytedeco/mkl-platform-redist "2020.3-1.5.4"] contains mkl_rt.dll and example native code works with it on Windows 👍
So it is compatibility issue with newer versions of MKL
I'll create an issue for this
I'm just so happy this works now, I can finally start my own little hobby project 😂
And performance is indeed awesome, CPU BLAS is at Numpy level and it's super nice how easy it is to offload computations to GPU 👍