Offscreen rendering

[davide445]

What about the offscreen rendering plugin. 

[silent]

4 hours ago, davide445 said:

What about the offscreen rendering plugin. 

Is there any specific use-case for the offscreen rendering? What kind of task you are trying to achieve - could you please describe in more details?

We are currently gathering information on the offscreen usage scenarios from our customers that will help us to develop right solution for each category.

Thanks!

[davide445]

As base doing Tracker work based on a configuration file without the need of a open Editor so without any opened UI, as assembled project.

As advanced option doing the same but interactively trough API control for the Tracker various options.

Having the possibility to launch different instances with different setups on the same host. 

Meanageable as server side component, with deployment and monitoring trough standard scheduler, i. E. In a remote cloud. 

Edited November 18, 2019 by davide445

[silent]

Hi Davide,

Thanks for the info.

What kind of cloud and OS you are going to use? You can configure Linux to run X server on a virtual screen via VirtualGL. This solution with VirtualGL is already possible with a current engine builds, so you can deploy on a cloud (that already have virtual machine with VirtualGL being set up).

However, if your Linux setup is not configurable and you can't enable X server here, you will need to wait. The same with Windows Server that requires session 0 support (DX 11.1+ feature). These still needs some additional time for the development.

Thanks!

[davide445]

The cloud part is so far never tested nor prototyped since there are other priorities, but needed for future commercial usage.

Our requirements are more about the first two rows, so doing Tracker activities without supervision as first requirement and controllable trough API as second.

[silent]

Not sure that I understand you correctly about Tracker. It sounds like a tool that you need to develop in order to modify track files on fly? And I also can't see how it's connected to the offscreen rendering in any way.

Right now Tracker tool is allowing you to play tracks via UnigineScript and you can write a callback functions to control it from C++. So, if you will write code that will generate *.track files for you by some rule you can already run and test it.

Thanks!

[davide445]

Maybe I'm the one not aware about what the offscreen rendering plugin is supposed to to. There is any description about his planned features anywhere?

[silent]

Well, basically offscreen is just a render to a texture without creating any external windows. Basically that's it :)

[davide445]

Sorry again and what's your use case based on was in development? My knowledge about these offscreen buffers was for storing realtime calculations such as VFX or Global Illumination, but in the plan I read "cloud rendering, computer vision learning". 

[silent]

Hi Davide,

Moved offscreen rendering discussion to another thread to avoid off topic in News section.

We treat 'offscreen' as synonym to 'headless' and imply the ability to run engine without any GUI window to it. Right now it's kind of possible if you are using Linux and Virtual GL, but true headless mode is not available yet (on systems with Session 0 requirement [Windows] and without X window server [Linux]).

In that headless mode you can have an access to any rendering buffer as you normally can get with regular engine instance (the same GI buffer is also available). You also can generate data for neural networks  (so called segmentation) or stream some real-time video for already built AI (that for example is driving a car and it needs sensors input that will be emulated).

Thanks!

[davide445]

I was thinking too we was derailing the News section, so thanks for creating the new thread.

What you describe is exactly what we need, the point is if will be available for 2.10.

[silent]

Do you need session 0 support on Windows right now? Could you please give us more technical details?

Right now (and in 2.10) you can use approach with Linux and VirtualGL installed in configured.

[davide445]

We are developing for Windows right now and even if we are thinking about Linux still that's not the case, so we need to use this on Windows environment.

About Session 0 I'm evaluating the needs for next project step, so will come back when finished the internal discussion on this.

[davide445]

@silent

For a new project proposal we are planning a web interface leveraging the logic of a VR 3d interactive application, of course adapted to work on a server.

Was reading offscreen rendering as work in progress during 2.15 release, there is any ETA/plan for his availability post 2.16.

[silent]

Hi Davide,

Offscreen rendering is still planned for 2.16 (but I'm afraid only for Sim edition at this moment). But this estimation is not yet confirmed.

You can already run any UNIGINE version on linux headless servers, but you would need to install VirtualGL and configure XOrg server to run with virtual screens instead. This method allows you to run any OpenGL application in headless mode. I've already described how to do this in previous messages in this post.

For Windows it's a bit different. In Sim version we have already a custom app implementation that can create offscreen rendering context, so you can check this app and create your own based on it.

Thanks!

[davide445]

32 minutes ago, silent said:

You can already run any UNIGINE version on linux headless servers, but you would need to install VirtualGL and configure XOrg server to run with virtual screens instead.

Having a Linux server it's not a problem. Due time constraint (need to close the proposal in a couple of days) did't have the time making tests so wanted to ask how much overhead you think this setup will add to Unigine.

In our setup we will use something like 2xA16 GPU (each A16 containing 4xA2, every A2 about the performance of a 2070) to serve 24-32 concurrent users visualizing - with a few other features addes - this kind of anatomic model

https://autode.sk/3v4HTyY

will be possible also in this setup to have a separated session for every user so to allow them to have a completelly autonomous experience. I suppose replicating Unigine installation on separate VMs, but eventually doing multirendering on the same instance will be another option.

[silent]

Taking in account that OpenGL is already in some cases up to 20%-30% slower than DX11, running it in VirtualGL will not make a huge difference in performance compared to the regular one.

What can is a CPU / GPU resources sharing between different processess running in the virtualized environment. That's would be really hard to predict.

 

Quote

will be possible also in this setup to have a separated session for every user so to allow them to have a completelly autonomous experience. I suppose replicating Unigine installation on separate VMs, but eventually doing multirendering on the same instance will be another option.

 

Well, I guess that would depend on the performance numbers that you would have with an multiple engine instances running on the same server. 24-32 users would mean that you need to run 24-32 engine instances at a single server, that may be very hard to achieve :)