Hi @rengolin @adam-smnk @rolfmorel @shahidact , should we rely on a100 alone or can we move to hopper partition available in pcl-tiergarten

The two a100 nodes should be fine.
We'll revise it in the future if we ever have problems with availability.

Not exactly sure why it fails.
Maybe sth clashes with the old cached LLVM build? You could try to add this change to the LLVM bump PR and see if it runs.

Also, now CUDA 13 being available, we should be able to use default system's gcc.
You could try to tweak setup_gpu_env.sh to stop sourcing older version.

Hmm, I see we also change gcc version in build_tpp.sh.
Then maybe try fresh LLVM build with a bump first before changing tooling versions.

Hmm, I see we also change gcc version in build_tpp.sh. Then maybe try fresh LLVM build with a bump first before changing tooling versions.

I think, this errors happening because of updates to pcl-tiergarten. This PR points to the old LLVM bump and older PRs before tiergarten upgradation passes. And, particularly this fails for pcl-spr* machines but passes for pcl-sprh*

Will investigate.

Merged the changes with PR: #1091. Hence, cancelling this.