I have a PC with integrated intel graphics along with RX Vega M GH. I also use an eGPU via thunderbolt, RX Vega 64 LC. It would appear that when benchmarking graphics I'm at the mercy of fate when it comes to the selected GPU for the tests. I noticed under the Advanced menu and GPU Compute it is possible to select the GPU used, though strangely it doesn't seem to make a difference. Is there a way to actually specify the GPU to be used in a multi-adapter system when performing graphics beenchmarks?
Announcement
Collapse
No announcement yet.
Selecting a specific GPU to test in a multi-adapter system
Collapse
X
-
Prior to DirectX12 there was very little control over what video card was used, often the only way was if the window was on the output connect to the graphics card then it should be using that card for the rendering.
DX12 introduced the ability to select which card to use, so you should be able to select this card in the advanced window and have it run. Does it have a monitor attached to it?
PerformanceTest may not be selecting it if there are no attached output devices.
On most laptop systems where the was dual integrated/discrete video options there should be driver options to allow the choice of which card to run the program on however in this case I imagine the drivers haven't been changed to recognise external GPUs.
If you launch PerformanceTest in debug mode, run 3D tests and send us the log files generated we'll check how the card is appearing to PerformanceTest.
-
Logs attached and screen capture of the display adapters shown by Device Manager. It appears that some of the tests are unaware of the presence of the Intel Graphic while other tests actually ran on Intel Graphics. My hope is that, similar to storage testing, it would be possible to specify the graphics adapter in settings. I wouldn't want to limit this to 3d only, but include 2d as well.
In windows 10 I could attempt to tell windows to use a performance GPU or otherwise for specific applications in PT. Though I don't think I could specify which one (Vega M GH vs Vega 64).
Essentially, the thought is to test performance individually, and "teamed" if allowed. I have found that programs such as Adobe creative suite and OBS Studio seem to be able to Team feature sets of different adapters to achieve a desired result.
Thanks for your timeAttached Files
Comment
-
This is pretty tricky situation as it's not a "switchable" setup and really a case where there are two video cards enabled.
Older version of DirectX will generally use the video card attached to monitor they are launched on however it seems that they are using the external card, all seem to be reporting the RX Vega in the debug log though some of the results don't seem to match. Potentially it's a display issue that isn't helped by AMD naming all their cards the same and using the same internal device IDs.
Can you please send us a baseline from each set of tests (with the external GPU attached and with it absent).
Also could you try attaching a monitor only to the eGPU, this will likely force everything to use the eGPU, and seeing if the results seem correct.
Comment
Comment