Hi
I understand the wait for GPU is the cpu waiting for the gpu to “Catch up”. The thing I don’t understand is that even with my gpu at 10ms (deep profiler) the CPU is consistently waiting a minimum of 10ms for the gpu.
I’ve tried disabling most objects in the scene with little luck.
Thanks for any thoughts.
One additional note.
Created a new blank scene with a single OVR controller.
My GPU time is still at 3ms in a blank scene 0_o and my OculusWaitForGPU has not gone down at all.
What is strange though is in a blank scene I’m getting a GPU spike once per second. The “total gpu ms” in the profiler lists 23ms, but adding up the ms of the itemized list at the bottom of the profiler everything still only adds up to 9ms.
I believe that Oculus Rift has a 90hz refresh rate, giving you around 11 ms to wait if nothing is going on.
Heya Jaimi,
I’m guessing I shouldn’t be throttled at below 60 fps by my CPU though? My GPU is well below 13 ms (except for the mentioned spikes) but my CPU is well above 13ms due to the “wait for gpu” calls. As a result the profiler has my app pegged well below 60 fps.
I’ll post some profiler pics tonight when I get to my pc.
With nothing going on, your wait call will need to pause to slow the framerate to coincide with the refresh rate. Various sources say the refresh is “90” or “75, 72, or 60”. So, I don’t think the wait call is the problem, even though it appears to be, it just means that nothing is happening, so your main thread has to wait so it doesn’t outrun the GPU. If you are seeing performance issues, I would think the problem would be somewhere else.
Heya,
here is where I’m at. my performance goes down dramatically with the headset turned on due to the “WaitforGPU” call. I’m sure there is something else going on, but I’m not rendering that much yet. Also its odd that deleting the objects that are rendering or adding 40+ high rez objects barely effects the FPS. Here is a shot of the profiler. I believe the “Spikes” in the GPU are caused by the editor as turning on profile editor seams to mirror the pattern. (also when hovering over a spike the itemized ms list still is around 11 ms.
I’m running a I5 GTX750
Any thoughts from the image?
One more pic. Switched quality settings to fastest. Profile lists my gpu at 10ms. (in the list the gpu is only 7ms). That said I’m still getting over 10ms WaitForOculusGPU calls. Could the editor running be causing this performance issue? I am running with game window set to maximize on play.
Still working on this. Here is a shot of my scene almost 100% gutted. The CPU is still getting hit by wait for GPU. In the profiler 8 of the CPUs total ms is wait for GPU though the GPU is well under 13. Any ideas what I should be looking for? It seams like I need to hit 9ms on both the cpu and gpu to get wait for GPU to go away… could there be some kind of loop where the cpu is missing a frame render and thus waiting for GPU which causes the GPU to wait for the CPU etc…
Its strange to me that I’m getting wait for GPU when the GPU performance is so much better then the CPUs.
Sorry If I’m hammering on this but I’m really confused. I’ve read up on what wait for GPU does and from my understanding I shouldn’t see it unless my GPU is the bottleneck.
Two additional shots:
Completely Blank scene still 8 MS wait for GPU
This is a development build. You can see my total ms is well below 13 but I’m getting 10 ms wait for GPU on the CPU.
