Open haixpham opened 4 years ago
If any makes a fork of cntk build on the lateset CUDA and python version that would be great! (i have no skill to do it)
@haixpham I'm still actively using cntk for my current and new projects. I don't have the incentives to switch yet. Python 3.6 is fine for me.
But one question that is on the top of my mind is whether RTX 3000 series will work with cntk or not? Does anyone know anything?
@delzac RTX 2000 series did not support CUDA 9.x - worked with 10.0. So my guess RTX 3000 will require more recent CUDA version.
I still hold my (little) hope that someone from MSR will release the design document of CNTK - so that someone can take over the development. Pytorch is feature rich, but using it feels like using Java all over again. And deployment is a pain. Good for research though.
I think ampere needs to run on CUDA 11 based on the screenshot above? And i was hoping to upgrading my GPU :(
Don't buy new if not required! Find a couple of 2nd hand Tesla P100 or Quadro P6000 cards with 16GB RAM each.
Now I'm trying to build CNTK with CUDA 11.1 for 3000 series of GPU. cuDNN and some related libraries are updated. Although I do not have knowledge of Cuda programming, I just want to build first without an error message.
Related issue what I opened, here #3835
I'm not using CNTK much these days, beside old projects. I still like it nonetheless, for its elegant APIs. Wish to use it with newer CUDA version (for new GPU compatibility), and maybe Python beyond 3.6.
So is there any brave soul taking the burden of maintaining CNTK on his own? @delzac: How are you doing these days?