The Orbital Gateway - 08 - Platforms, Background, Optimisation

Update

Hey everyone. Quick update with a few things in my scene.

Main angle camera screenshot from Unreal 4. Progress from last post showcases lots of splashes of gold and yellow on the surfaces of the portal cylinder and its greebles. There is also some of that, together with sci-fi, square grid wall textures, in the background.

Visually, I spent my attention on mainly two things. First was working on the platforms that the humans sit on. I created a proper mesh to replace the previous placeholder. I also made a material with gradient I can tweak and put colors and noises on top of in order to get that force field effect. I experimented with a few colors for it but ultimately ended up with the one like in the original concept.

Close up screenshot of the floating platform. Its top and bottom are Zbrush sculpts showing metal with intricate carvings. In the middle of the platform stand 3 UE mannequins in an A pose. Around them is a strong, yellow emissive ring.

I then moved onto the background once again. I tried to improve the protrusion bits I had placed previously. I wanted the hard surface to look proper and as if part of the room and like weld metal, as opposed to the previous golden supports that looked just like clipped through.

Screenshot of the sides of the giant room, around the portal cylinder. The walls have square, greeble like flat texture with strong normals giving it dimension. Also occasional yellow lights scattered here and there. In the middle of the picture is a cylindrical opening and platform. That is where spaceships would float through, going towards the portal.

I also worked more on my trimsheet that I deploy in the background. I added a part to it that has panel-like details and greeble like features. You can see that above.

Optimization

And all of this takes us to today. Tonight I decided to try optimizing the scene a bit. I had spline meshes, non-static parts and materials in the scene that I wanted to improve tech wise. It would not be a successful video at the end, if I were to leave the performance as it were and recorded at 15 FPS. : )

Before. Constant at 75-80 ms:

SCENE RENDERING screen shown in Unreal 4 Editor. Lists and details RenderViewFamily, Draw calls, and other profiling stats. Most important one here, for this case study, is the stat third from bottom to top: Mesh draw calls. In this picture it reads 1,696.45.

After. Stable at around 18 ms.

Screen identical to the previous. Important stat is again: Mesh draw calls. In this “after” picture it reads 306.00.

With some more clean-up and with no stat scenerendering on, I’m sitting at a stable 10 ms. Quite better even than the previously mentioned 18 ms. This is not in standalone form so one could argue it might be even better there. I think for recording the video I’ll play a timeline through a cooked standalone project and maybe get a bit of a boost in ms from that too.

MS Comparison

All in all I think it will do for now. Here is one final comparison to an ms observation from a week ago to show you how slow it was, lol.

Two screenshots from the Unreal Editor. Left shows “stat unit” at 60.6 FPS, 164.97 ms (with hit mostly on GPU). Right image shows the same, but after fixes up. Reads 94.33 FPS, 10.60 ms.

Gif Progress

But away from this, and back to visuals. Here is a GIF:

Gif progress from the usual camera angle. This time it mostly showcases progress on the background and the walls. First they had placeholder, gray materials, but then were replaced for greeble texture details.

Hope you are doing well.

Thanks for reading,

Pete.

Go to next blog entry.

Commentary from 2023

I wanted to quickly mention how exactly I accomplished the big win in performance above. I think anyone with experience can derive and make an educated guess, just by looking at my stat scenerendering profiling pictures above. Yet I write and share these posts with a lot of students too and I wanted to explain it a bit better for them.

In the making of this project, I heavily relied on spline blueprints.

When there is a long platform, be it straight geometry or cylindrical motives, instead of placing by hand a repetitive, modular piece, I would feed that piece into a spline blueprint system of my own.

It made the art creation process much more quick and powerful.

There is one caveat though. When you work with spline blueprints like that you end up leaving all parts as “Movable” instead of “Static”.

To be able to edit a spline curve on the go too, the nodes derive tangent points.

All of those pieces stack up in draw calls. When you place one long platform that is a spline blueprint, instead of having say 1 to 5 draw calls you end up with 150-200, because of all of the spline components.

When you put an identical platform spline blueprint copy next to it, that gets its own 150-200 draw calls too. They don’t get bundle together with the first one.

I knew this from the very get go, when I started adding splines. I knew I will need to go back and clean up and “bake” my geometry into truly static geometry.

The “Before” and “After” screenshots above, you see me doing exactly that. I select my spline blueprints and I use the Unreal 4 Editor integrated“Merge Actors” function.

That turns my one, super long platform (for example) that has 150 draw calls into one piece of geometry, that is say mid-poly, but is 1 draw call.

I then reuse that super long platform, to the sides, by replacing the rest of the similar blueprints.

I make sure I don’t change the x y z scale of those copied platforms (otherwise the draw calls get separated again, even if not as bad as before).

Doing even more actions, similar to the above description, I optimized the draw calls in my scene from 1,696.45 to 306.00.

That on its own got me from 77 ms (12 fps) to a 10 ms (94 fps).

Unrelated to the Blueprints or Splines above, but if you are using the Unreal Geometry Brushes to block out spaces, bear in mind they have similar caveat of draw calls.

Each of those brushes, in its nature to be editable on the go, comes with dozens of draw calls instead of optimized just a few calls.

I often come across student’s block out projects that run at bare minimum FPS. They often ask and wonder why that happens and most often my advise is either to bake down all of their Geometry Brushes to truly static, non editable geometry, or to start blocking out using primitive Static Meshes.

Have fun in Unreal!

Pete.

Go to next blog entry.