[Plugin] Intel-GVT-g


Recommended Posts

Discovered this plugin a few minutes ago, works like a charm! (Pentium G5400 / UHD 610)

 

Thank you so much!!!

 

Just out of curiosity, why does the Overview list more available modes than actually available?

 

Overview of available modes:
i915-GVTg_V5_1: VMEM: <512MB, 2048MB>, up to 1920x1200
i915-GVTg_V5_2: VMEM: <256MB, 1024MB>, up to 1920x1200
i915-GVTg_V5_4: VMEM: <128MB, 512MB>, up to 1920x1200
i915-GVTg_V5_8: VMEM: <64MB, 384MB>, up to 1024x768

 

Following modes are available:
i915-GVTg_V5_4
i915-GVTg_V5_8

 

(I guess, the UHD 610 only supports i915-GVTg_V5_4 and i915-GVTg_V5_8)

Link to comment
5 hours ago, lixe said:

(I guess, the UHD 610 only supports i915-GVTg_V5_4 and i915-GVTg_V5_8)

its about available mem for the vgpu(s), so basically with the desktop cpu's we cant get more out of it.

 

even roughly the *_4 and *_8 (4 or 8 vgpu's) prolly wont work ... i can use 2 /_4) or 3 (_8) max, then i run into memory page errors ...

  • Like 1
Link to comment
On 8/28/2021 at 6:26 AM, alturismo said:

its about available mem for the vgpu(s), so basically with the desktop cpu's we cant get more out of it.

 

even roughly the *_4 and *_8 (4 or 8 vgpu's) prolly wont work ... i can use 2 /_4) or 3 (_8) max, then i run into memory page errors ...

 

I am getting exactly the same as @STGMavrick with Blue Iris and GVT-g... 

 

Tried all what has been suggested before, yet I cannot get rid of the freezing.

 

@alturismo are you also having the same symptoms? 

If yes, what do you mean by "i can use 2 /_4) or 3 (_8) max, then i run into memory page errors ..." - Does this mean you have a solution to it?

 

Edited by olympia
Link to comment
5 hours ago, olympia said:

are you also having the same symptoms? 

nope, luckily my gvt-g VM is running just fine

 

about the types

image.thumb.png.5a5c3298cc358029b4f4ed07bf8e4199.png

 

the i915...._4 <- means in theory 4 simultan 4 gpu*s are possible

the i915...._8 <- means in theory 4 simultan 8 gpu*s are possible

 

but i cant use that many simultan, if i add more and more simultan i run into memory page errors and the VM's crashes

  • Thanks 1
Link to comment
4 hours ago, alturismo said:

but i cant use that many simultan, if i add more and more simultan i run into memory page errors and the VM's crashes

 

Ahh, ok, so you (only) have the memory page errors with a VM crash consequence if you run more than 2-3 VMs at the same time.

Is libvirt also crashing at you if this happens and only complete restart of unraid helps?

 

Could this be somehow load related? With Blue Iris there is a constant load of 20-25% on the igpu. I don't know if this is much, but constant. At normal operation I get to the point of memory page errors within 24 hours, but for example if parity check is running (because of an unclean restart following a prior crash) then the memory page errors comes very quick (few minutes, but may an hour).

 

How much is your load on the igpu?

Also, are you using RDP to connect remotely?

 

@STGMavrick would you mind sharing your current status with this?

Edited by olympia
Link to comment
4 hours ago, olympia said:

Also, are you using RDP to connect remotely?

 

you can force this pretty easy with RDP when you have a higher resolution then 1920 x 1080 ...

 

so yes, i would prefer RDP due performance, but depending on client the resolution switch can hurt ... so currently i either use VNC or parsec (fluid experience)

 

4 hours ago, olympia said:

How much is your load on the igpu?

my load is pretty low, only when there is really activity on the VM, but mostly 0 as its in idle ;)

its my "Homeoffice" work VM in my usecase.

Link to comment
1 hour ago, alturismo said:

you can force this pretty easy with RDP when you have a higher resolution then 1920 x 1080 ...

 

so yes, i would prefer RDP due performance, but depending on client the resolution switch can hurt ... so currently i either use VNC or parsec (fluid experience)

 

What do you mean by force this? You mean force and easily reproduce memory page errors followed by the crash of libvirt?

 

I am still in the early few hours of testing, but the VM is now running with Blue Iris for 5:30 hours now in parallel with parity check and the only thing what I did is that I did not connect to the server remotely at all since it booted up. The unraid was booting up, VM and parity check have started automatically and seems to run smoothly now. The VM never survived that much time before with Blue Iris and parity check in parallel.

 

So it pretty much looking like that somehow I am hit by this old issue what MS has fixed in Theory long ago with KB4522355 back in 2019:

"Addresses an issue with high CPU usage in Desktop Window Manager (dwm.exe) when you disconnect from a Remote Desktop Protocol (RDP) session."

 

There was a lot of reports re to this back in 2019, but not so much since then, so I am not sure this is the same issue.

Do you have any clue based on this? Presumable I should also try another remote desktop solution?

 

Edited by olympia
Link to comment
22 minutes ago, olympia said:

What do you mean by force this? You mean force and easily reproduce memory page errors followed by the crash of libvirt?

when you not connecting to the VM you wont have any load on the gpu, by forcing means, RDP Protocol by default is using the resolution from client side, so lets say you sit on a 4k monitor, now the session will open in 4k instead the max 1080p and you will force a mem page error fault ...

 

and i dont think this is related here

Link to comment
2 hours ago, alturismo said:

when you not connecting to the VM you wont have any load on the gpu, by forcing means, RDP Protocol by default is using the resolution from client side, so lets say you sit on a 4k monitor, now the session will open in 4k instead the max 1080p and you will force a mem page error fault ...

 

and i dont think this is related here

 

This is 100% RDP related for me. Blue Iris in a Win10 VM with constant 20% load was running nicely in parallel with the parity check for 8:30+ hours, then following I logged in and disconnected (not logged out) using RDP I've got the memory page errors in 5 minutes and libvirt has got frozen. 

 

I also noted the following:

 - I have 4 cores allocated to the Win10 VM

 - the average CPU load on all 4 cores cores were around 15% when it was still running

 - If I logged in with RDP and just disconnected from RDP session (leaving my user logged in), then the CPU loads were more than double on all core

 - If I logged (signed) out my user in Windows instead of just disconnecting from the RDP session, then all 4 cores have got back to the normal 15% and not having the double load issue

 

Now the question is that why this Windows 10 build 1903 bug from 2019 is hitting me in 2021? :(

Le me see if the workaround from 2019 disabling "Use WDDM graphics display driver for Remote Desktop Connections" helps me now...

Link to comment
52 minutes ago, olympia said:

the average CPU load on all 4 cores cores were around 15% when it was still running

my gvt-g VM is in idle just core 1 which needs a little more as expected and i never resolved this yet

 

52 minutes ago, olympia said:

Now the question is that why this Windows 10 build 1903 bug from 2019 is hitting me in 2021?

you know there was a major change in july 2020 when i remember about RDP and GPU hardware rendering, just as note.

 

53 minutes ago, olympia said:

using RDP I've got the memory page errors in 5 minutes and libvirt has got frozen. 

i cant confirm this when i use RDP with a low resolution (max 1920 * 1080), i can confirm this when i use a higher resolution like 2560 x 1080 and up ... but this is also a (sadly) expected behaviour here as we are over the limits ... and my board cant assign more vram to the igpu here, at least i know in older days i could assign manually up to 1 GB RAM for the igpu, but my current mainboard is either auto or max 64 mb ...

 

so yes, RDP also causing issues here but ONLY when im over the gvt-g vGPU limits.

Link to comment
1 hour ago, alturismo said:

so yes, RDP also causing issues here but ONLY when im over the gvt-g vGPU limits.

 

...but contrary to you, the issue at me is not when I have the rdp session open (what I also had at 1080p, so not beyond the limit), but when I disconnect RDP without logging out from Windows. If I am logging out, then it seems like there is no issue.

Link to comment
2 hours ago, olympia said:

the issue at me is not when I have the rdp session open

 

this is while a rdp session is open here now while playing back a 1080p video in browser, the cores are actually double proven, 2 VM's using them ... one with a dedicated gpu /gt1030) as desktop PC annd the gvt--g  VM (workspace)

image.png.b439315a7a41775c50aad589ac892d18.png

 

image.png.b28e1cb987f0df246aee392db2c70449.png

 

image.thumb.png.2c1436ff92bdf9db3bd6ce8d741a7f2a.png

 

now, just closing the session without logoff, gpu is immediately offloaded, thats when i understand you correctly where your issues starting

 

image.png.15bdcb245818dfc89207fe7f3a26634f.png

 

and CPU is back to normal idle state here (2 win VM's in idle, core 1 (CPU5) is a little higher, thats the only recognition here), nothing more ... all good.

 

image.png.f46ff85db8fb3dd1ae9a478d4638abfa.png

 

here you see the current running VM's (AlsPC_Work is the gvt-g one)

image.thumb.png.3598ff7c2edbde3e140ffb5f94771fb7.png

here the CPU pinning

image.png.f028dd662cdd32df947048e0b6d970d6.png

 

as you see, AlsPC and ApsPC_Work using the same cores so thats all normal here, also after a RDP session with a close (no logoff).

when i logoff its even more in idle state, i guess the same to you as you say no issue when u logoff.

 

so, i tested now 20 Minutes RDP session from my laptop (also a 1080p device) so i dont run into trouble ... all fine here, so in the end i cant reproduce the behaviour you describe when only closing the rdp session without logoff, sorry but i cant help here.

  • Like 1
Link to comment

Wow! you made a really deep testing there! Thank you very much for your efforts! I really do appreciate this!

 

Getting late here, I will do the same with same screenshots tomorrow just for the sake of the records.

 

Probably also Blue Iris is behaving badly in this setup. It is running as a windows service and constantly using igpu, so it is not like the video in a web browser what should stop when you disconnect. 

 

...and your question: open means when I have an active RDP window open - no problem with this at all. 

When I just close the RDP window (disconnect from the session, but leave the user logged in) - that's when the problem starts.

If I connect back to the left sesion - no issue again until the window is open.

If I leave the session with logging out, then it looks like the issue is not popping up - still need some testing on this to confirm 100%

 

And also for the records:

There is no issue with RDP without igpu passthrough

 

Let me ask a few quick questions:

 - is your VM configured as Q35-5.1 with OVMF BIOS?

 - did you do any tweaks to the video drivers in win10 (e.g. removing QXL video device manually from the XML suggested somewhere on the forums) other than what is in the second post? 

 - what unRAID version are you running? I don't recognize the chart for the CPU load. Is this in 6.10rc1 or that is provided by some plugin?

 - do you have GuC/ HuC loading disabled?

Link to comment
7 hours ago, olympia said:

 - is your VM configured as Q35-5.1 with OVMF BIOS?

 

yes, but i tested all through and in the end, all worked, at the beginnings i only had success with seabios.

 

7 hours ago, olympia said:

 - did you do any tweaks to the video drivers in win10 (e.g. removing QXL video device manually from the XML suggested somewhere on the forums) other than what is in the second post? 

 

yes, i always remove the video and graphics part from the xml so there is no more 2nd display adapter, the recommensation is more or less from me ;) 

 

7 hours ago, olympia said:

 - what unRAID version are you running? I don't recognize the chart for the CPU load. Is this in 6.10rc1 or that is provided by some plugin?

 

im with this plugin from the 1st startup, so i tested this on 6.92 and now im on 6.10rc, all fine on both unraid versions (here atleast)

 

7 hours ago, olympia said:

 - do you have GuC/ HuC loading disabled?

nope

 

and also i have this running on 2nd mashine i serve, also all good there on different hardware with 6.92, but i can only speak about 9th or 10th gen cpu's.

 

may what you also could test, there is a "rdp hack" out there (rdp wrapper) so you can run multiply sessions simultan on win 10, then you could may follow the tasks what happens when a session is closed while you still on the 2nd session active ...

 

and btw, once im in the gpu page error loops here (like using too much ressouces on the VM due high res) then i cant get out of it ... in the end system will crash due memory page errors.

Link to comment
  • 2 weeks later...

I'll be frank, I don't know much about this at all but I am also getting the freezing when trying to run Blue Iris in a Windows 10 VM. Runs fine for a time, and then libvert crashes and I have to restart unRAID to get it back. This in turn causes an unclean shutdown as I assume libvert wont exit in time before it forces the shutdown. I have tried both i915-GVTg_V5_4 and i915-GVTg_V5_8 which give the same result.

 

This is the only VM I use so wouldn't have thought it was overloaded but Blue Iris is set up to use QuickSync so maybe that's contributing. 

 

Running an Intel Corporation Xeon E3-1200 v5/E3-1500 v5/6th Gen Core Processor

 

I assume there isn't much I can do apart from upgrade the CPU etc?

Link to comment
10 hours ago, ich777 said:

Can you try to install a VNC server inside the VM and connect to the VM through this VNC server and not over RDP and check if it also hapoens that the server hangs?

Tight VNC seems fine so far. I've gone in and out of the VM a few times etc to test and no errors in the syslog yet. I'll keep using it for now I and see how it goes.

 

EDIT: LOL, literally as I wrote the above, libvert has locked up and I wasn't even connected to the VNC. So same issues with VNC as RDP.

Edited by mgladwin
Link to comment
4 hours ago, mgladwin said:

EDIT: LOL, literally as I wrote the above, libvert has locked up and I wasn't even connected to the VNC. So same issues with VNC as RDP.

Did you log off or did you just close the connection?

Maybe it has something to do how you quit the session, that's just a really wild guess but maybe this has something to do with it.

 

May I also ask if you habe disabled sleep from the monitor and also sleep from the VM too?

Link to comment

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Restore formatting

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.