Nope, in my case disabling SpeedStep didn’t make any difference, exactly the same fps as before both in 0.4.2 and 0.4.3.
I really think it has to be something related to the way mirroring on and off is handled in 0.4.3… I find very strange that in 0.4.2 the difference between mirroring on and off is between 15 and 20 fps (more with mirroring off), but in 0.4.3 mirroring on and off gives exactly the same fps (with aero on, with aero off no mirroring gives 0 fps).
Something else that i think might be a factor is AMD hardware, the other person i’ve seen in the forums complaining about exactly the same problem is the user “knack” and i believe he’s got an R9 290 as well.
I attach some profiles i took at the exact same location (start of the level, without moving at all). 0.4.2 vs 0.4.3 and mirroring on vs mirroring off.
I’m also getting the same strange 37.5 fps cap in the sun temple demo with UE4.5.1 and OVR 0.4.3 + Oculus DK2. Disabling Vsync in the console and in my graphics card control panel help a tiny bit but I still can’t break into the 60+fps.
Ok i’ve taken a couple more profiles with SpeedStep disabled and i can see that (as already mentioned) it does help getting rid of the spikes in the InitViews when profiling (its effect is actually more visible in 0.4.2)…
Unfortunately, as i said in the previous post, this doesn’t seem to translate in any perceived improvement when actually playing the demo, still seems locked at 75 fps in 0.4.2 and 37.5 fps in 0.4.3.
I’ve also added another 2 profiles with vsync off, just to spice things up (sorry about the number of posts btw, limited by the maximum attachment size).
0.4.2 - DIRECT Mode - Mirror OFF - Aero OFF - SpeedStep OFF - VSync OFF
Game - AVG 11.264 / MAX 12.289
Render - AVG 11.249 / MAX 12.247
GameThread - CPU Stall - Wait For Event
AVG 10.049 / MAX 11.003 ms
RenderThread - Slate RT Present Time - RHIMETHOD_EndDrawingViewport_Flush
AVG 0.335 / MAX 0.684 ms
0.4.3 - DIRECT Mode - Mirror OFF - Aero ON - SpeedStep OFF - VSync OFF
Game - AVG 13.963 / MAX 21.752
Render - AVG 13.943 / MAX 15.697
GameThread - CPU Stall - Wait For Event
AVG 12.746 / MAX 13.581 ms
RenderThread - Slate RT Present Time - RHIMETHOD_EndDrawingViewport_Flush
Thanks a lot for all this info, a lot of food for thoughts. I am looking at all of those captures at the moment. Weirdly, but turning off SpeedStep does the trick for me… I use 780ti / i7-4770 / Win8.1 x64.
I’m not sure if this is anyway related or not, but I was monitoring my gpu (GTX 980) with evga precision X, and without the DK2 enabled it hits a steady 99% gpu usage and 98-99% power usage (getting around 200-210 fps in sun temple).
With the DK2 enabled though, it drops to around 60% at the highest with 75 fps and gets really unstable looking at the statue bouncing back and forth between 40 and 60% usages and 40-60 fps.
I would have thought that with the DK2, it would be using all the card’s resources but it doesn’t seem to be?
This is with UE4.5.1 and OVR 0.4.3, speed step disabled
This is a different issue, but also a known one and it should be addressed in 0.4.4 SDK. The issue is that SDK forces GPU to flush after each frame to achieve lowest latency. So, GPU is not 100% loaded. We are going to allow 1 frame of additional latency in order to improve GPU load.
I see. Too bad, just got used to this incredible low latency in 0.4.3. But If it’s the only way to get back some performance I guess you have no choice.
So, is there any way this could be solved by async. timewarp later in 0.5?
