Rapids on ubuntu 20.04 with Geforce RTX 3070

tosupport.txt (1.1 KB)
What does it mean and how to cope with this:
Requirement not satisfied, you need 2.31, but Your installed version is: 2.31

This seems to be some generic, bogus message if conda fails to install any intermediate packages.
https://github.com/rapidsai/cuml/issues/3108

Thank you for your reply. I still have an issue.
I followed the enclosed link to generate extra information, but I got unexpected result here:
~$ ./print_env.sh
./print_env.sh: line 8: syntax error near unexpected token newline' ./print_env.sh: line 8:

I use conda to install tensorflow-gpu with rapids. Separately TF environment or Rapids environment on this machine are implemented with success (however TF is pip installed from PyPi wheels for 2.4.1 and 2.5.2, but I had to edit .condarc to put conda-forge to the end). If I try to top-up the cudann=8.1.* for the installed rapids environment - installation fails with the same message at the end, informing of 2.31, same if I try to install rapids into TF environment. I did not manage to get working tensorflow-gpu from conda-forge.
What should I look for and what info should I gather to proceed?
Running Ubuntu 20.04 on AMD Ryzen 5 3600 6-Core Processor with NVIDIA GeForce RTX 3070
Driver Version: 515.65.01

You should have followed the linked issue report
https://github.com/NVIDIA-Merlin/Transformers4Rec/issues/424#issuecomment-1147317944

Sorry, looks like you already did.
I don’t know of any further help, this is conda repo specific which has nothing to do with the nvidia driver.

Thank you very much for support. Following the first link You have shared in first response, I saw many people struggling with installation of rapids cuml. I have rsolved my issue by revising .condarc file and listing (defaults, rapidsai, nvidia, and last conda-forge). It was important to have conda itself from main channel, because else conda update conda would update conda from conda-forge and would not search for modules, whic is important for resolution. However, when installing cuml, it was important to list those channels and conda-forge for a speciffic libcumlprims multi-GPU library on x64 platforms even if most of us, students, are still on single GPU. When libcumlprims is generated ? · Issue #1518 · rapidsai/cuml · GitHub
If to follow explanation here
When libcumlprims is generated ? · Issue #1518 · rapidsai/cuml · GitHub
It appears that previousely cuml was built with single GPU flag. But documentation is not explaining that, unless you try compiling by yourself. Another worth comment was, "if you have conda from the main chanel, and install that libcumlprims on a single GPU, you might just skip that -c conda-forge and get rapids on single gpu x64 (at least Ubuntu 20.04) working.
Therefore thank you for support indeed.