No hardware acceleration on remote desktop w/multiple cards

[ShawnCG]

Dear Arction Support,

One of our customers is using our application over remote desktop connection to a host machine that has multiple NVidia cards. Specifically, it has three cards, two of which are set to TCC mode and the third allowed for display use. They are reporting very slow rendering and poor responsiveness which we can see is due to a fallback to software rendering.

Here is the LightningChart rendering info that was pulled when running over remote desktop:
Active graphics card: Software rasterizer
Optimal pure device mode supported: False
Shader model 3 supported: True
Anti-aliasing supported: True
Fast-vertex format supported: False
Hardware vertex processing: False
Index buffer max size: 16777215
Index buffers (32-bit): True
Texture max height: 8192
Texture max width: 8192
Texture sizes not power of two allowed: False
GPU chip vendor ID: 0
Graphics render device created: True
Rendering Tier: 0

Compare this to the output on the same machine from dxdiag.exe over remote desktop:
---------------
Display Devices
---------------
Card name: RDPDD Chained DD
...
D3D9 Overlay: n/a
DXVA-HD: n/a
DDraw Status: Not Available
D3D Status: Enabled
AGP Status: Not Available

The hardware/driver configuration seems to have possibly fooled LightningCharts to thinking it needs to use software rendering. On other machines, we have found the LightningCharts to work very well over remote desktop using the hardware acceleration available through the RDPDD Chained DD protocol.

Do you think this is a situation that will rely on better remote desktop implementation from Microsoft or is there something else that LightningCharts can do to find and use the RDPDD Chained DD adapter?

Thanks,
Shawn

[ShawnCG]

Correction -- Of the three NVidia cards on this machine, only one (Quadro K2000) is set to TCC mode. The three cards are:

--System Video Controller Device Info (1/3)--
Video controller description: NVIDIA Quadro K2000
Video controller RAM: 2147483648
Video controller status: OK
Video adapter compatibility: NVIDIA
Video adapter DAC type: Integrated RAMDAC
Video processor: Quadro K2000
Video adapter driver version: 9.18.13.4062
Video adapter driver date: 20140731000000.000000-000

--System Video Controller Device Info (2/3)--
Video controller description: NVIDIA GeForce GTX 660 Ti
Video controller RAM: 2147483648
Video controller status: OK
Video adapter compatibility: NVIDIA
Video adapter DAC type: Integrated RAMDAC
Video mode description: 2560 x 1440 x 4294967296 colors
Video processor: GeForce GTX 660 Ti
Video adapter driver version: 9.18.13.4062
Video adapter driver date: 20140731000000.000000-000

--System Video Controller Device Info (3/3)--
Video controller description: NVIDIA GeForce GTX 660 Ti
Video controller RAM: 2147483648
Video controller status: OK
Video adapter compatibility: NVIDIA
Video adapter DAC type: Integrated RAMDAC
Video processor: GeForce GTX 660 Ti
Video adapter driver version: 9.18.13.4062
Video adapter driver date: 20140731000000.000000-000

[ArctionPasi]

Please try forcing the hardware acceleration.

set chart.ChartRenderOptions.DeviceType = Hardware.

Now you should see them in "full" performance (relatively speaking with RDT of course), or not appearing at all. How does it work?

[ShawnCG]

Thanks, Pasi. I'll see if we can test this on the customer's machine with the issue.

[ShawnCG]

We had the customer run a test application with the forced hardware rendering on and off and that flag had no effect. The rendering was slow for both settings.

We also had the customer take out two of the installed cards and leave only one GeForce card. In that configuration, the render times were very fast, implying that the hardware rendering works with only the one card, but with the addition of a card for CUDA processing in TCC mode, the LightningCharts resort to software rendering.

Our customer noted that the card in TCC mode is device 0 on the system.

Do these additional details give you any ideas where the problem could be?

Thanks,
Shawn