[Docker-OSX]

I'm still working to get mine set up, but the vnc container does run and stays running unlike the standard one.

Here's what I did:

 

open the terminal on your unraid machine either through the web or ssh.
 

#entirely optional.. i just wanted an easy place to dump the dockerfile temporarily
mkdir /var/tmp/docker-osx
cd /var/tmp/docker-osx

wget https://raw.githubusercontent.com/sickcodes/Docker-OSX/master/vnc-version/Dockerfile

docker build -t docker-osx-vnc .
docker run --device /dev/kvm --device /dev/snd -p 8888:5999 -p 50922:10022 -d --privileged docker-osx-vnc:latest

 

that should get it to show up in your containers list in unraid. I didn't have to change any default settings or permissions. Just r

 

After it starts running, you will be shown a vnc password in the terminal. Copy than and then you can connect to it with VNC on port 8888. I just used RealVNC viewer to connect. Works a treat.

I'm looking at adding NoVNC support to mine later, but that will a problem for future darthcircuit to worry about.

 

 

[QEMU PCIe Root Port Patch]

Thank you very much for that. 

[QEMU PCIe Root Port Patch]

Fingers crossed

[QEMU PCIe Root Port Patch]

Is there any way of manually updating qemu to 4.1? I won't be using qcow2, especially compressed. I just pass through the whole controller for my windows 10 vm, and if my pcie lanes are only running at 1x speed, that will be a problem.

[QEMU PCIe Root Port Patch]

Interesting. Thanks for that. Would be nice to have the option anyway for those that aren’t using qcow2. I’m just passing through my nvme controller natively and a sata drive. 

[QEMU PCIe Root Port Patch]

Do we know why it was downgraded? I was hoping to have that fixed as well.

[Terrible Perfomance on threadripper]

On 10/23/2019 at 8:45 PM, chromedome said:

 

Don't worry use your desktop experience as your benchmark, if your mouse is smooth w/o any jitter while background apps are loading and your sound is fine - your system is responsive enough.   Play games for a few hours.  My Latency Mon numbers would say otherwise, but  I can play many games w/o any stutter, I haven't played one that did.  What I learned a few months late is that you don't need a lot of extra stuff with libvirtd for a smooth overall desktop experience.  

So start simple with your configuration.  

The more I played around with the CPU configuration, the more degraded my performance became.  I learned that by just installing win10 using a basic native qemu setup and using a simple XML and adding what I see from forums, but one line at a time - to troubleshoot. 

So here I am sharing mine.   I borrowed some CPU config lines from Symon's  XML.  Not all of them worked for me because as I'm using qemu 4.0.0
 

<snipped/>

 

 

 

Thanks for that

 

The last big issue i had that was noticeable was that my sata and usb controller were on a different numa node, so i'm ordered a combo sata/usb3 controller card. I'm thinking that will improve things a lot. I'm going to pass that through when it gets here, and then leave things as is until 6.8 comes out and i can see if there are any improvements from qemu 4.1...

 

I've been really pleased so far tho, and this is the most helpful community i've been a part of. Thanks everyone!

[Terrible Perfomance on threadripper]

Thanks everyone for your suggestions. I did try the msifix a while ago and that did make a difference too. I’m still getting latency quite a bit higher than I’d like (averages between 6-900), and between that and poor benchmark performance I didn’t do any further testing in most games. 

 

I bit the bullet and purchased unraid, hooked my rig back up to my tv and to my big surprise, every game I tried played almost flawlessly at 4k. I think I’ve been getting too hung up on the benchmarks. Even though ghost recon benchmarks horribly (between 4 and 30 FPS on 1080p), actual gameplay is fine with a pretty solid 45fps at 4k. Wolfenstein 2 played beautifully at 4k 60fps on ultra. 

 

Its not perfect solution  and I’m sure I’ll be tweaking it for a while, but I’m really happy right now. 

[Terrible Perfomance on threadripper]

Thanks for that suggestion :) I haven't had as much time as I'd like to be able to play with it this week.

 

I cleaned out the all my pcie slots with rubbing alcohol and cleaned the contacts on the card itself. I did find something interesting in that I had my gpu in an 8x slot. I've moved it to an x16 slot on the same node, and the command you had me run confirmed that.

 

 

I tried your xml change too, and that didn't seem to make any difference. My latency is about the same at around 1500. It seems curious to me that the latency peaks with the nvidia driver. I tried an older driver, but that didn't help...

 

In the bios, i forced the pcie slot to run at x16, but it didnt help.

 

Any other ideas?

 

I'm kinda thinking 6.8 will help since it's coming with qemu 4.1 @limetech - Sorry, i don't want to annoy you, but i don't know how far away we are from RC1 and my trial expires this weekend. I'm not ready to drop the $$ unless i can make my vm work right.

[Terrible Perfomance on threadripper]

Thanks for your replies! I've still got some wild latency, but your recommendations were a huge step in the right direction. I'm down to about 3500 on latency mon after 3 hours vs 15000 after 24 seconds. GRW doubled in FPS to 30 at 1080p, but realistically with this rig I was able to do 4k at 45fps on bare metal.

 

 

I unpinned 2 threads in my vm and assigned that to my emulatorpin and iothread

I passed through the controller for my NVME and installed the Samsung driver for that

I added the Epyc info to my <cpu> tag. 

I copied the <hyperv> section from the first reply

 

This is my pinned cpu's now:

 

My New XML

 

Current Numastat (no change)

 

 

anyone have any other ideas I could try to get my latency down?

 

I've checked my bios and that is up to date. Maybe I missed something there?

[Performance Improvements in VMs by adjusting CPU pinning and assignment]

21 hours ago, Nephilgrim said:

Im on mobile, but first thing I noticed is that you set the emulatorpin to a core/thread from node 0.

Try using 11cores on your vm from the node1 instead of 12 and change the emulatorpin to that free core/thread.

 

Probably that wil get rid of your latency issue since a core in node 0 is taking control of your vm in node1.

 

 

 

 

Thanks! I'll give that a try I deleted my post here the same time you hit submit so i could create my own thread. dunno if that was the right choice or not. didn't want to double post I'll let you know if that helped or not tho. Thanks!

 

EDIT: I've updated the post I made HERE

 

I think that helped a lot along with a couple other things, but am still getting some crazy latency. Thanks for the suggestion do you have any more ideas?

[Terrible Perfomance on threadripper]

Hi everyone

 

I'm pretty new to this still, but I'm kinda hitting a roadblock and could use some help. My brother @m0ngr31 is really selling me on Unraid, and I'm about to purchase a license if i can get my gaming VM working correctly.

 

I'm running a Threadripper 1920x, 12c/24t with 32gb of ram on an MSI x399 carbon pro MB. I'm passing through a gtx 1080ti to my windows 10 vm. I've enabled NUMA in my bios by setting memory interleaving to "Channel" as recommended by @SpaceInvaderOne - I've disabled CPU throttling and C-states in my bios. Made sure my SVM and IOMMU settings are enabled. I've changed a few other things over the course of the last month, but can't remember all what i've set in the bios. I've also got an nvme ssd and a standard sata ssd passed through directly to the vm.

 

I've been following various tips from the forum here.

 

Using LSTOPO, I've verified that my GPU and NVME are both running in NUMA node 1. As such, I've pinned, isolated and assigned every cpu for node 1. I've also followed @Nephilgrim post here, and edited the xml for numatune to pin my vm's ram allocation to node 1. I've verified this with numastat, and it is working.

 

 

The only things not in node 1 as far as i can tell are my sata controller and my ethernet controller. I've disabled my second drive that is running over sata, but it didn't improve my performance.

 

My pinned and isolated CPUs:

 

 

 

LSTOPO:

 

Full XML for my VM Here (pastebin)

(Sidenote, i decreased the vm's ram size before uploading this to see if it would help, but no dice. My xml reports 15gb of ram instead of the 16 i've been testing with)

Syslinux:

kernel /bzimage
append isolcpus=6-11,18-23 vfio-pci.ids=10de:1b06,1022:145c pcie_aspm=off kvm_amd npt=1 nested=1 avic=1 initrd=/bzroot

My vfio-pci.ids are my gpu and usb controllers that i'm passing through

 

I've also made a hook for qemu to change the FIFO cpu scheduling or something as per this guy's suggestion. The code for that hook is:

 

#!/bin/bash -x

if [[ $2 == "start"* ]]; then
  if pid=$(pidof qemu-system-x86_64); then
    chrt -f -p 1 $pid
    echo $(date) changing scheduling for pid $pid >> /tmp/libvirthook.log
  fi

fi

 

I've made sure to add the root port info for my pcie slots so they report full x16 pcie 3.0 speeds inside the vm.

 

latencymon and gaming deliver terrible results. Ghost Recon Wildlands nets me about 14fps, unigine gave me a score of about 5000.

 

I'm really at a loss. From what i read, making sure ram, cpu and gpu were all in the same node seemed to fix the problem for most people. I have done all those things, but my performance is just hot garbage. Can someone offer me some suggestions? I'd really appreciate it

 

Thanks in advance!

[Purchasing Unraid with bitcoin/eth etc...]

Ok thanks! I'll do a bank transfer then

[Purchasing Unraid with bitcoin/eth etc...]

Is it possible to purchase Unraid with cryptocurrency? I'm just about ready to pull the trigger, but I'd like to not have to do a bank transfer right now if I don't have to.

 

Thanks!