letik

I made and interesting discovery while experimenting with older hardware HP MicroServer Gen8 (2012) that supposedly cannot run modern NVIDIA data-center GPUs like the A2 or L4 (2023) because the platform only exposes a 24GB PCIe prefetchable window. On paper these GPUs require far more address space, so initialization fails. However, there is a way to perform BAR negotiation during initialization that can fit inside the Gen8 window, which makes it possible to run these GPUs inside a VM on the Microserver Gen8. Result: a decade-old microserver running modern GPU workloads inside a VM. I wrote up the full explanation and setup steps for anyone still running these little machines here: https://medium.com/@aletik/the-impossible-nvidia-gpus-ca48ac425efc

I made and interesting discovery while experimenting with older hardware HP MicroServer Gen8 (2012) that supposedly cannot run modern NVIDIA data-center GPUs like the A2 or L4 (2023) because the platform only exposes a 24GB PCIe prefetchable window. On paper these GPUs require far more address space, so initialization fails. However, there is a way to perform BAR negotiation during initialization that can fit inside the Gen8 window, which makes it possible to run these GPUs inside a VM on the Microserver Gen8. Result: a decade-old microserver running modern GPU workloads inside a VM. I wrote up the full explanation and setup steps for anyone still running these little machines here: https://medium.com/@aletik/the-impossible-nvidia-gpus-ca48ac425efc

I think that what needs to be done is Unraid really needs to look into implementing better upgrade assistant. So users can save off whatever versions they want and upgrade/downgrade to any saved or new version. Would even be nice to be able to upload an older version through the web UI and downgrade if needed.

Problem solved. Installed Unraid ver. 6.10.3 on exactly the same server (I have 2 exact same ones) Basic settings applied to syslinux (intel_iommu=relax_rmrr video=efifb:off) and NVIDIA moved under a separate group. Pass through worked out of the box. Freaking fu$%^& insane if you ask me. How can you break one of the main functionalities/ advantages of Unraid and not look into it considering many people are clueless...

Wonder if anyone have any ideas? Thanks

Hi Guys, First time posting but have been using Unraid for a few years. For the passed several days tried to pass through the GV-N105TOC-4GL Nvidia GTX 1050Ti GPU card. I think that I tried a ton of stuff reading all GPU threads here. Maybe it's even not how VM is configured type issue but some BIOS-related...not sure.. I also booted with 2019 J6 and 2014 BIOS. In Bios VT-d is enabled and also GPU is enabled as Secondary... When I started going through these issues and as a developer who wants to use GPU for compute I had somewhat structured approach to play with settings but now a bit lost. GPU is functional. I set it up for testing as Primary and got video from it via HDMI then set it back as secondary. Also for me in System devices I can only isolate GPU in a single group if I select Downstream. If I select Both then VIDEO and AUDIO component go into 2 separate groups. But it makes no difference in the VM. Ubuntu doesn't see any GPU with lshw -C display command. If I boot into Win11 with GPU, the VM although in started state is unpingable (without GPU the RDM works fine) I also tried manually changing XML VM file adding multifunction and trying to set VIDEO and AUDIO component to same bus but then libvirt complains about double use even both are in the same IOMMU group. I have this in my syslinux kernel /bzimage append intel_iommu=relax_rmrr video=efifb:off pcie_acs_override=downstream vfio_iommu_type1.allow_unsafe_interrupts=1 initrd=/bzroot Diagnostic file attached as well as some screens and config files. Need someones help. I'm a bit lost Would love to understand where the problem may be. Thank you so so much.. nas-diagnostics-GTX1050Ti.zip ubuntu_vm_config.xml