Using Lavina on Google VM w/ NVIDIA Tesla T4

[mariocape]

Hello,

I am attempting to test Lavina on one of our VM configurations. The VM is running Windows Server 2019 and has a NVIDIA Tesla T4 GPU.

I saw the system requirements where it states driver version 451.48 is incompatible.

The problem is that NVIDIA only offers Tesla Drivers with the versions 451.48 and 451.82

Am I just out of luck to test this (for the time being)?

Or do I need to change the GPU? I have the option to use a K80, P100 or V100...

Thanks!

[Phillip Miller]

Our recent versions correct the incompatibility with 451.88
You should be fine by using our latest Beta version.
I will see if I can modify that previous post, as it's no longer relevant.

What VM software are you using BTW? (e.g., Grid, Citrix, VMware, etc.)

- Phil

[npg]

Hi

I assume this is the page you are referring to: https://docs.chaosgroup.com/display/...m+Requirements
It seems we forgot to update it. We released a hotfix - version 0.4.2 - which fixed the issue with the 450 series of drivers. So the driver should be ok. Just download the latest Lavina here https://download.chaosgroup.com/?platform=47&product=60

On the other hand, we haven't tested Windows Server or a Tesla GPU, nor running in a VM for that matter. Windows Server 2019 seems to match the minimum Windows version we require (build 17763).

I'm interested to hear if you managed to run Lavina successfully on this exotic setup

[mariocape]

Our recent versions correct the incompatibility with 451.88
You should be fine by using our latest Beta version.
I will see if I can modify that previous post, as it's no longer relevant.

What VM software are you using BTW? (e.g., Grid, Citrix, VMware, etc.)

- Phil

OK- I'll give that a try. We are not doing any real-time work, so we are using Microsoft Remote Desktop as our VM client to connect with.

[mariocape]

Hi

I assume this is the page you are referring to: https://docs.chaosgroup.com/display/...m+Requirements
It seems we forgot to update it. We released a hotfix - version 0.4.2 - which fixed the issue with the 450 series of drivers. So the driver should be ok. Just download the latest Lavina here https://download.chaosgroup.com/?platform=47&product=60

On the other hand, we haven't tested Windows Server or a Tesla GPU, nor running in a VM for that matter. Windows Server 2019 seems to match the minimum Windows version we require (build 17763).

I'm interested to hear if you managed to run Lavina successfully on this exotic setup

Thank you Nikola,

I download beta 4.3 from https://download.chaosgroup.com/?platform=47&product=60

Windows Server 2019, NVIDIA Tesla T4 GPU, Driver 451.48

On launch I am seeing this error message.

I'll try some other GPU hardware and report back...

[Phillip Miller]

Remote Desktop may not be giving access to the GPU. Only a few VM's do. That's why I mentioned VMWare, Citrix and Grid.

[Phillip Miller]

I just confirmed that RDP does not enable "GPU pass through" by default.
You may be able to enable it through adjusting the operating system on your cloud machine.
Here's a link for adjusting that setting on Windows Server that may provide some guidance: https://community.esri.com/thread/22...windows-10-rdp

- Phil

[npg]

You could also check the file %APPDATA%\Chaos Group\Project Lavina\debug_log.txt - the devices Lavina "sees" should be listed in the first few lines.

[dougWilkinson]

i've been testing cloud VMs with the T4 GPU. i havn't had any issues, though i haven't tested with Lavina specifically. I have used Teradici graphics agent and also the free version of Parsec and both seemed to work great. I agree with the above statements and would just add that making GPUs work right over RDP is a pain and when it does work, the responsiveness isn't great either.
Maybe this helps.

[mariocape]

Thank you for all the suggestions!

I tried enabling the Policy as described in the linked article - no luck. Same result when trying to launch Project Lavina - here the log:

Lavina v0.4.3 (build 746875, hash b76c70b2)
Build date: 2020.07.29
Window initialized in 1406.8 ms
NVIDIA driver version 8.17.14.5148 detected.
Driver check: OK
Windows version is 10.0.17763.
Device #1: Microsoft Basic Render Driver, 0 MB (DevId=0x8C, SubSysId=0x0)
No suitable GPU device found! A DXR-compatible device is required.
Windows version is 10.0.17763.
Windows version is 10.0.17763.

I am using Microsoft Remote Desktop on Windows 10 to connect to the VM.

I did try Teradici in the past, but honestly could not make out any performance improvement vs using the MS RDP client. I also have to say we are not doing real-time CGI work (yet)...

I am also a bit fuzzy on the virtual GPU. On this same VM configuration, Maya/3ds Max seems to work well, so does the GPU rendering using V-Ray - which was the main reason I expected Project Lavina to work...

FWIW I am getting the exact same error message switching the GPU on the VM from a T4 to a P100 or a V100. Windows 2016 does correctly show the GPU hardware in the Device Manager.

[npg]

It looks like DirectX only lists software rendering (pseudo-)devices on this machine. These could still work for rasterized viewports like in Max/Maya, although framerate should be lower, I suppose.

V-Ray GPU does not use DirectX, although it is quite strange that it can use a valid CUDA device, while there is no valid DirectX device, apparently.

Currently I don't have access to a similar VM to debug deeper.

[mariocape]

Understood - let me know if you are interested in debugging this further, and I can setup a VM for you to play with.

It will have the same hardware configuration...

[npg]

mariocape, we've just released a small update - 0.4.4. It includes a change to the device enumeration code - it will no longer skip a device if it is flagged as "software device". Previously those were not even listed in the debug log (with the assumption this was always a "microsoft basic render driver"). However in the log you shared earlier we can see device #0 was skipped (so it must have been software) and device #1 was not, even though it was the microsoft thing again. In any case if the device doesn't support DXR it will still not be used, but maybe we could see something about your device #0 in the log with this new build.

[mariocape]

This version is working! Thank you!!

[npg]

This is great to hear!

It would be helpful if you could copy the few lines in the debug log (Help->Debug Log) where devices are listed.

It looks something like this, for example:
Device #0: NVIDIA GeForce RTX 2080 Ti, 11048 MB (DevId=0x1E04, SubSysId=0x1E0410DE)
Device #1: Microsoft Basic Render Driver, 0 MB (DevId=0x8C, SubSysId=0x0)