GPU tests failing with OOM errors

[simone-silvestri]

since a week or so, the GPU tests start failing randomly with OOM errors. Examples of this failure:
https://github.com/NumericalEarth/NumericalEarth.jl/actions/runs/24768246993/job/72467804322
https://github.com/NumericalEarth/NumericalEarth.jl/actions/runs/24667361659/job/72128408807

Maybe we are at the limit of what the GPU worker can handle? Should we increase the cache size or clean the cache before starting a job?

cc @awiteck @giordano

[giordano]

Not sure how the cache would help. As I mentioned in #179 (comment), a possible strategy is to use fewer concurrent jobs, 7 instead of 8 should already help a lot, we had a similar situation in Breeze.jl

[simone-silvestri]

Ah stupid me I thought that comment was about reading the crash output 😅
I ll open a PR to reduce the number of jobs