Hi @csantill ,
I had a strange behavior of Intel MKL recently with R 4.0.3 on Ubuntu 18.04.
The results of a simple matrix multiplication of two matrices of size (1600,1600) * (1600,20000) was simply not correct.
I was lucky that I had a step to take the square root of the result and it gave nan because one of the entries was negative. Otherwise, I would never be able to find the error. I compared the output with MATLAB and Python base libraries and in R using the base library as well and I am confident that Intel MKL in this setting gives completely wrong results.
I just wanted to post this here for anyone as a warning.
Thanks!
Hi @csantill ,
I had a strange behavior of Intel MKL recently with R 4.0.3 on Ubuntu 18.04.
The results of a simple matrix multiplication of two matrices of size (1600,1600) * (1600,20000) was simply not correct.
I was lucky that I had a step to take the square root of the result and it gave
nanbecause one of the entries was negative. Otherwise, I would never be able to find the error. I compared the output with MATLAB and Python base libraries and in R using the base library as well and I am confident that Intel MKL in this setting gives completely wrong results.I just wanted to post this here for anyone as a warning.
Thanks!