Ole Schütt

@mkrack, it seems when ELPA is built by the toolchain then the `pdbg` regtests [pass](https://storage.googleapis.com/cp2k-ci/run-cp2k-pdbg-4986d1b8_report.txt) even with a threshold of `1E-14` (see #1501).

> What's about the warnings printed by ELPA itself? I'm not sure which warnings you mean? In `cp_fm_elpa.F` there is a second [eigenvalue check](https://github.com/cp2k/cp2k/blob/master/src/fm/cp_fm_elpa.F#L341), but that one passes the regtests...

> I remember that there have been warnings from the ELPA routines itself. Maybe, they were not important or unrelated. I ran `grep -ir elpa *` and `grep -ir warn...

Yes, it would be nice to get some up-to-date GPU numbers on the wiki page. We do keep taps on our GPU performance via some smaller benchmarks that run daily:...

Since there seems nothing left to discuss, I'm going to close this now. Maybe with the arrival of new machines some of the performance numbers also find their way into...

> Yes, that's what the issue was opened for I'm happy to keep this issue open if you deem it useful. Personally, I see this more as a continuous maintenance...

> I got H2O-64 test case to crash.... I can confirm this. On my dual GPU system cp2k crashes when run with **more than one MPI rank and more than...

I also think this would be a great feature. It could be implemented like `--host-routing`. The advantage of cookie based routing would be that all users get the same urls,...

No, it should be enough to call `dbcsr_acc_clear_errors()` only once when DBCSR gains control. At the same time you should also call `dbcsr_acc_set_active_device()`. Basically, don't assume that the GPU's state...

I really like your idea of relying on the C API instead of dealing with external `.mod` files.