Slate, a performance hog, according to "stat dumpframe"

Hello,

First I’d like to say that UE4 is awesome! (really, well, at least I like it a lot better than UDK which wasn’t to my taste) Won’t say much more about its awesomeness because you probably know and the community has likely already praised all the good things about it.

Anyway, there is a thing I’d like to know/shed light on. In my quest to find about why the performance in an nearly empty scene makes me cry every time I look at the frame render time, I stumbled across some interesting results when trying to profile the current frame, using the command “stat dumpframe” (and after failing to used GPU Perf Studio because it just freezes when capturing the frame).

It seems that Slate hits really hard on my GPU. According to the stats:

LogStats2: 74.00ms ( 66) - Thread_14c4_0 - RenderThread

74 ms for rendering, approximately 14 FPS. This is quite a big number considering that my GPU works incredibly well with other GPU intensive programs and runs latest games quite smoothly (well, no lower than 30 frames at least). All this was captured on the starter map (the one with chairs and materials and stuffs).

But when looking down, it becomes a lot more interesting
LogStats2: 11.02ms ( 1) - FDrawSceneCommand - FDrawSceneCommand - STATGROUP_RenderThreadCommands

Wait what? Real rendering taks only 11 ms? (I measured it many times to be sure, it varies between 11 and 20).
This is quite a good result (but believable) considering it’s a mobile GPU (but still, a powerful one!).

And then, it appeared, like a monster, who wants to eat all your frames:
LogStats2: 62.73ms ( 2) - SlateDrawWindowsCommand - SlateDrawWindowsCommand - STATGROUP_RenderThreadCommands

Ugh… 62 ms to render the UI? What the hell? Although I don’t really know how it works under the hood (and It might be linked to other rendering stuffs, so I won’t criticise that fast), it still seems like a lot to me.

System:
Intel Core i7-3635QM CPU @ 2.40GHz (8 CPUs).
8192MB RAM.
Windows 7 Pro.
AMD 8870M (is detected as an Intel 4000 because of the switchable graphics).

In the end, there is no real question, I’d just like to get some informations about this.

Kind regards,
Alexander

Thanks for posting this research, I look forward to the Epic Response!

Rama

use “Stat Slate” to see slate times as well as “Profile GPU” to see the GPU time. 62 milliseconds is pretty crazy.

I’d be interested to see the results of both of these stat dumps if you dont mind.
Currently with an Nvidia 670 I am getting 2 ms render thread time and 1ms gpu time to render the editor.

Forgive me, I’ve completely overlooked this issue since I have made a switch to another more powerful machine.
But, as far as I remember, the timings were on SlateUI even on profilegpu

Sorry but I can’t help more since I don’t have access to that computer.