Debian 'Bookworm' 5.18 and nvidia Xorg Tesla 510.72 ... weird behaviour

[Trollti]

Hi,

i upgraded a machine to the new kernel 5.18.

unfortunately, nvidia tesla 510.47 exited with error at upgrade.

nvidia released a new driver 510.72... (which, by the way, screwed up the menu bars from my gnome flashback :/ )

when I launch nanominer, (3.6.5 cuda) everything seems normal but the speed... the programs says LHR unlock, but it runs at 'full' lhr :'(

the gpu is a gigabyte 3070 ti OC 8G... I tried (un)secure boot, CSM and EFI, motherboard is a MSI B550-A PRO (latest bios, normally ^^)

If I boot on a machine still with kernel 5.17 and nvidia 510.47... it's A-OK

I'll will investigate how to roll back to the 510.47

Have a nice day,

[Trollti]

I used the 'floppy disk' style... cloned a ssd with debian 5.17 and Xorg Tesla 510.47 to the ones 'infected' by the update, disabled in 'Sofwares' the automatic update feature. I am sure there are other ways to do that ... but my linux semantic has it's limits :)

[Trollti]

etc-lhrUnlocked-363.log etc-lhrUnlocked-nano365-nvidia510-72.log

etc-lhrUnlocked-363 log is the one from the cloned system; the other one is the one with the issue (69Mh/s -> 37Mh/s) the backup system is under kernel 5.17 and the problematic one was under kernel 5.18

PS: with ETH it was the same behaviour; POWs are not concerned PPS: I have also a machine running with nano364 and another with nano365... it's the nvidia driver which is the pain in the ass!