Best use of NVIDIA Graphic Card for Open Roads Designer Connect

Happy to share Mark. The value in the forums is really dependent on us offering our experiences. I am regularly grateful for you contributions here as well. 

Steve

:-) :-)

Hi Cem,

Cem Donmez said:

I am wondering if this is not filling up the graphic cards GPU

I think all these settings have nothing to do with how specific application use GPU (because it's solely controlled by the application itself).

The tool primary task is to allow to define, in situation when 2 different GPU are available, what GPU will be used by particular application (e.g. MicroStation) in certain situation (running on battery...). It solves a question how an application, not knowing what GPU is better (or even not aware of there are more GPUs available), should choose GPU.

In the case of desktop, I do not see any reason to switch between Intel and Nvidia. On notebook, I use something like "use Intel, when on battery, otherwise Nvidia" for all applications.

Cem Donmez said:

And does Bentley have suggestions/best practices to have the best profit of Nvidia graphic cards ?

I am not aware of such recommendations, plus it has to be always clearly specified what application will be used.

Be aware ORD (and all Power platform based products) is solely DirectX application. Unfortunately Nvidia does not focus on DirectX too much and has long history of bugs, performance issues and even laying about support of specific DirectX features. On the other hand, for last few years Nvidia provided the best hardware in terms of  performance and computation (now it seems AMD offers performance-equal HW finally), which is in parallel with Nvidia focus to OpenGL applications, AI and similar tasks.

So when talking about ORD, check what DirectX related settings your driver offers. Of course ensure you use the latest driver and also you card has the latest firmware. MicroStation uses specific DirectX featuers to move computation (e.g. tessellation) to GPU, also I guess some printing related task are handled, when possible, by GPU.

Different situation can be when you will use LumenRT for visualization, because it's able to use Nvidia GPU (specific types) for some rendering processing. See Technology preview feature documentation of this enhancement.

Cem Donmez said:

Any additional configuration propositions to spped up the processing and visualitaion time will be wellcomed. 

As recommended already, to use fast SSD is important. The most of motherboards and notebooks today allow to use M.2 SSD (or adapter can be used), which is substantially faster than older SATA 3.

With regards,

  Jan

Hi Steven,

Steven Litzau said:

These likely aren't GPU heavy operations. 

you are right. GPU, through DirectX, is used to display content, plus several extra tasks, but to load and analyze data, CPU is used.

Steven Litzau said:

I think most of these settings are not likely to impact performance in ORD

Yes, ORD performance depends on other things like:

• Performance of PowerPlatform (engine) itself. It's still under development, but in recent versions (especially U13 and U14), MicroStation engine performance was increased a lot. But it's really complex issue, because in some releases, heavy internal changes were implemented (allowing to do some tasks e.g. 1000x faster), but it's not always immediately visible for a user. Plus, when there will be internal change with speed up one million, for a user it can be only 1% difference, because every operation consists from hundreds of even thousands steps.
• Performance of ORD code, which I think is today the biggest bottleneck and are that has to be optimized a lot. ORD uses data model based on EC data, implemented on top of DGN elements and geometry. It means to load ORD file requires to load DGN (which is typically extremely fast and simple), but also to analyze extra data stored in DGN file and create civil model object model in memory. Plus, to keep all templates (again, it about complex model built on the fly). My feeling (but I do not know ORD code) is that to implement (as fast as possible) features have been priority in the past, whereas to optimize it (which is boring, slow and expensive process, not clearly visible, because it's not "new feature") had not the same priority.

With regards,

  Jan