Poor VM performance and reliability with GPUs

[turbooo]

I've been struggling with the reliability of my single Windows 10 VM and I'm at a loss for what to do. The GPUs are locking up the VM and the entire system, I think. The Windows 10 vm is a fresh install after the update. About half of the time Windows will go into a system recovery boot so I have to hard reset the server. Before updating to 6.4 I was using this vm as a standard desktop and mining when not in use. Now I can't mine for stretches longer than 3-4 hours before the system halts. 

 

I've done several Radeon drive uninstalls and reinstalls. When trying to play PUBG the game crashes either on boot or as soon as a game starts. Audio is another issue, it will cut out for 15 seconds and then continue. 

 

• i7-7700K OC 4.8GHz
• 32GB DDR4
• 5 x 3TB WD Reds
• 250GB Samsung EVO cache
• Two PowerColor rx580s 8GB in Crossfire (17.12.1 driver)
• 850W Gold power supply
• Unraid 6.4

 

Attached is the syslog from yesterday. 

syslog.txt

[SSD]

There are a lot of things that must be done right for a VM with passthrough to work properly. With the info provided it is very hard to know that you did all of those things. I am not even sure why you believe that the problem has anything to do with passthrough.

 

To confirm that, you could install something like NoMachine on the VM and another computer, and access it without passthrough. Then see if the VM locks up. If it doesn't, then you'd be right about passthrough being (at least part) of the problem. But if it locks up the same, your video passthrough is not the problem.

 

Not sure how you installed, but I would suggest following @gridrunner's excellent 2 part video series:

 

 

If you follow this, your audio should work much better and maybe whatever has gone wrong with your unstable VM will be corrected.

 

Unfortunately it is quite difficult to take vague symptoms, or even detailed symptoms, and trace that back to a root cause. It is easier just to do it right the first time.

 

If you follow that guide and still have problems - it could be something unique to your hardware. Or your overclock could be unstable. No way to know.

 

[turbooo]

I was able to find that I had other devices in my primary GPU's IOMMU group. Enabling PCIe ACS Override seems to have helped with the audio issues. PUBG is still force closing when I open it, but that might be an AMD driver and PUBG issue. 

[Maticks]

I would try not to enable ACS if you can, it can break other things. if it does then try moving the graphics card from Slot 1 to any other slot that usually fixes the IO MMU group issue.

You can usually do Slot 2 and Slot 3or4 depending on motherboard configuration in your manual you will get 2 8x slots and thats more than enough for those graphics cards.

 

Other suggestions for VM performance, always leave one core free for Unraid don't pass all cores to the VM.

If you are passing only 2 or 3 cores make sure you pass through the Hyperthread cores, in 6.4 these are now grouped together it will say something like 0/4 and you tick them in pairs.

 

As seen attached.

[Maticks]

Also crossfire i don't believe it officially supported that might be causing some of your issues.

[turbooo]

4 minutes ago, Maticks said:

Also crossfire i don't believe it officially supported that might be causing some of your issues.

 

Crossfire not supported with Unraid or PUBG?

 

For about a month I was smoothly running Forza 7 with crossfire enabled averaging over 100 fps. 

 

Quote

You can usually do Slot 2 and Slot 3or4 depending on motherboard configuration in your manual you will get 2 8x slots and thats more than enough for those graphics cards.

1

 

I only have two x8 slots and they're occupied by the rx580s. Should have just gotten a vega 56 last summer when I had the chance.

 

What problems can I run into with ACS enabled?