RTX 4000 Mosaic SLI performance and rendering issues

Graphics / Linux Linux
1

Performance is not good, vkcube content dissapears when moving across GPU cards in 2x3 configuration. Please advise

Nvidia driver 535.161.07
Xubuntu 22.04 kernel 6.5.0-26-generic #26~22.04.1-Ubuntu

/etc/default/grub
GRUB_CMDLINE_LINUX_DEFAULT=“nomodeset drm=0 iommu=off”

/etc/initramfs-tools/modules
nvidia
nvidia_modeset
nvidia_uvm
nvidia_drm

I had to put my configured xorg.conf contents into:
/usr/share/X11/xorg.conf.d/10-nvidia.conf
or SLI mosaic would fail

/etc/modprobe.d/nvidia-graphics-drivers-kms.conf
options nvidia-drm modeset=0

/lib/modprobe.d/nvidia-graphics-drivers.conf
blacklist nouveau
blacklist lbm-nouveau
alias nouveau off
alias lbm-nouveau off

sudo cat /sys/module/nvidia_drm/parameters/modeset
N

uploading nvidia-bug-report.log.gz gets stuck
“Processing” on this forum

2

Please try to unzip and upload the log.

3

Please see attached.
nvidia-bug-report.log.gz (2.5 MB)
thank you

4

Setup looks fine but there are a lot of errors (memory unmap errors, Xid 32, 56, segfault of nvidia-settings, vulkaninfo) that would rather point to a hw issue, either pcie slot or system memory. Tried reseating/changing slots? Memory set to stock clocks or overclocked/xmp active?

5

Thank you for the suggestion. Stock clocks and I will try re-seating components or plugging them in one by one. Once the machine is available to me again. Please leave this open until I can report back next week.

6

Just to update in my first report I showed iommu=off in my grub config, but I actually used amd_iommu=off (as the motherboard is AMD based) to get to the state I was in during my report. Since then I’ve not had much time to disassemble the PC but I did update the bios and try some other linux distros (manjaro, debian) with similar if not worse results. I also updated the bios as I had to reflash. Here is an updated bug report and I’m no longer getting the Xid errors but getting segfaults from vulkaninfo and libnvidia-glcore.so.535.171.04
nvidia-bug-report.log.gz (2.2 MB) Performance is still bad with tearing as well and vkcube content is cut-off between GPUs.