Last night, I was testing an AMD FirePro S9300 x2 as a replacement for an nVIDIA Titan Z, when I received a PSOD. My hypervisor host is a ProLiant DL580 G7 running ESXi 6.5u3. I have 6 stable VMs that have been running for 15-20 hours per day, for at least the past year. The FirePro (similar to a Radeon Pro Duo) has a few important features:
- passive air cooler
- 2x PCIe 8-pin power cables required
- has 1x Mini-DisplayPort, disabled by default
The card is optimised for use in servers, and has dual-GPUs on it. In the past, I used the Titan Z to power 2 VMs simultaneously, and did so without issue. I've been doing this for space-saving reasons, in addition to other factors.
Last night, I had one GPU passed through to a Unix-based VM and the other to a BlissOS (Linux-based) VM. The Unix-based VM is one of the 6 stable VMs I mentioned before - no crashes or other major incidents. The BlissOS VM is a relatively new, and hadn't been powered on since passing a GPU through to it that day. The PSOD showed up when I tried powering on BlissOS:
Near the very top, it seems to mention a Windows 10 Enterprise VM. The only Windows 10 VM I have is another one of my 6 stable VMs. It wasn't given one of the FirePro's GPUs. In fact, it has its own dedicated GPU. I need assistance figuring out where I tripped up.
I did another test. This time, I powered off all other VMs, aside from Windows Server 2019. The Windows Server VM has no PCI passthrough devices. BlissOS was the only VM using the FirePro card (Unix-based VM was shutdown for this test). Still got a PSOD:
I'll have to compare the two PSOD's in the morning, for common factors...