Jump to content

0.9.5 seems slower than 0.9.4


Go to solution Solved by Josh,

Recommended Posts

I hate to say it, but I'm seeing it a lot slower overall.  Actually, it seems 2.5x slower. :unsure:

These shots are in release.  In 0.9.5 I've had to remove the skydome due to some shader errors, and the terrain due to it not texturing.  The terrain was replaced with a single plane.  So 0.9.5 is doing less work but is a lot slower and the grass don't look as good.

Also in 0.9.5 I've seen the FPS fluctuate between 120 and 60 FPS like a seesaw even when the camera is still.

Scene in 0.9.4:

_094_scene.thumb.png.0400052d41cad410d79e2b655239191e.png

Same scene in 0.9.5:

_095_Scene.thumb.png.13e3d1af0390ec54b15a6d514b8627c3.png

Link to comment
Share on other sites

At the moment this isn't really actionable information. What is different about your demo and the performance benchmark apps? Are you constantly streaming geometry in? 

My job is to make tools you love, with the features you want, and performance you can't live without.

Link to comment
Share on other sites

Don't really see a problem with any of these stats.  Besides the FPS that is.  It gets down to 90 for a second or two then back up to about 200 for a bit.  I wish I could use FRAPS to verify the FPS but it doesn't seem to work with ultra.  It decided to work now.  FRAPS confirms the FPS is correct.

Performance_001.thumb.png.23f11f2a0cae028e8352f5cf7f558f12.png

Link to comment
Share on other sites

I can confirm that this is the case for me too (when I can even run the editor without continuous crashes to desktops) when it does run the editor is very much slower than it was before. The more assets that are added to the map/terrain then the more unstable the editor appears to become the more likely it is to crash the more likely it is that FPS count reduces etc. 

Link to comment
Share on other sites

I tested on my GEForce 1080, and these are the numbers I got:

0.9.4 / 0.9.5

Instances: 1206 / 2320
Lighting: 863 / 810
Animation: 1067 / 1032
Batches: 5199 / 3300

So on Nvidia it's faster in one test, slower in the other, and about the same in the other two.

When you are running at 1000+ frames per second, it's very easy for small things to have a disproportionate impact on the framerate. Adding something that takes 1/4 of a millisecond will drop 6000 FPS to 4000, so I don't think at that speed the results are very meaningful. Both the instances and batches tests could have been skewed by very small changes.

 

My job is to make tools you love, with the features you want, and performance you can't live without.

Link to comment
Share on other sites

 

If you look at GPU utilization on the 1080, the numbers are like this:

0.9.4 / 0.9.5

Instances: 96% / 85%
Lighting 96% / 96%
Animation 96% / 93%
Batches 76% / 67%

Again, they are for practical purposes identical in lighting and animation. The batches test had lower GPU usage, which may be caused by small changes in the visibility list processing step, since this involves 1000+ separate batches that have to be iterated through and processed in the rendering thread.

The GPU usage in the instances test is strange because 0.9.5 has lower GPU utilization but higher performance. This may be attributed to small changes in shader code, although I do not know what those changes would be. 0.9.5 does use one fewer vertex attribute, but I don't really see that doubling the framerate.

All in all, it looks like OpenGL 4.6 gives AMD a major performance boost and runs about the same as Vulkan on Nvidia cards in these tests.

 

My job is to make tools you love, with the features you want, and performance you can't live without.

Link to comment
Share on other sites

While putting together an example I have found something;

When the trees trunks are white, the FPS is stable.  When they are black, the FPS is a sine wave going from 90FPS to 280FPS and back and forth.

It's hit and miss which one I get, obviously there are missing shaders and I think something it's not being set correctly by default upon engine load.  It's interesting how the terrain is black also, I have not assigned any material to this terrain.

WhiteTrunks.thumb.png.05d82af71f255f578da06067ddd32deb.png

BlackTrees.thumb.png.b9892e5e59b569ada3c1ce60891e261b.png

Link to comment
Share on other sites

  • Solution

It actually makes sense why your colors were flipping back and forth.

The bindless texture override in the shader family was not being taken into account in the sorting, so it was just random. So sometimes your trees were being rendered with the more expensive terrain shader.

  • Thanks 1

My job is to make tools you love, with the features you want, and performance you can't live without.

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

 Share

×
×
  • Create New...