GPU vs CPU time in SP/MP

Hi all,

I would like to have community opinion about a really unexpected drop in performance when I'm playing with my Oculus in Single Player or Multi Player mission.

I did a test and here's the result for the same scenario, using the Oculus Debug Tool.

With my current VR's settings here's the rendering time in SP:

GPU - 20/22ms

CPU - 10/12ms

For MP mission, in the same scenario (so, same airport, same hangar, etc. etc.):

GPU - 20/22ms

CPU - 30/32ms

As you can see, GPU performance is the same, it takes 20 to 22 ms to render a frame.

But, due to more than 2.5 time for CPU rendering (increasing from 10/12 to 30/32) my overall FPS drop, because it seems that CPU is the bottleneck.

I'm not a MP server, I'm connected just as client.

In the first case I can play steady with ASW at 45 fps. In second case I have to play less then 30 fps for MP.

I really do not understand why, in MP, CPU is involved in such a way to render the frame.

My CPU is a I7 overclocked @4200.

Thanks for help

Hi all,

I would like to have community opinion about a really unexpected drop in performance when I'm playing with my Oculus in Single Player or Multi Player mission.

 

I did a test and here's the result for the same scenario, using the Oculus Debug Tool.

With my current VR's settings here's the rendering time in SP:

GPU - 20/22ms

CPU - 10/12ms

 

For MP mission, in the same scenario (so, same airport, same hangar, etc. etc.):

GPU - 20/22ms

CPU - 30/32ms

 

As you can see, GPU performance is the same, it takes 20 to 22 ms to render a frame.

But, due to more than 2.5 time for CPU rendering (increasing from 10/12 to 30/32) my overall FPS drop, because it seems that CPU is the bottleneck.

 

I'm not a MP server, I'm connected just as client.

In the first case I can play steady with ASW at 45 fps. In second case I have to play less then 30 fps for MP.

 

I really do not understand why, in MP, CPU is involved in such a way to render the frame.

My CPU is a I7 overclocked @4200.

 

Thanks for help

My educated guess is that it is due to DCS being single threaded and having to handle the addition of the netcode stack in MP.

Thanks for your comment. In the meanwhile I did some other test with different MP servers and the result is heterogeneous: it seems that it depends from the kind of mission.

So, heavy missions impact more than I guess with client CPU performance

Hopefully the planned standalone server will include de-coupling of the "game" from the "server", thus at least allowing us to run the server AI loading on a different CPU core.