Update 26.08.05

The front page of the main menu isn't doing it's job, it should be getting you into a game you're interested in as easy as possible. I've made a few changes this week, and we'll keep changing it until it's useful.

Friends List

devultj

The friends list got a couple of upgrades this week as it was looking dated. Party members are grouped in the list and you can invite users to your game / party straight from the row.

Streamer Mode

Carson

We've added a Streamer Mode setting to the platform. When enabled, everyone (yourself included) will be given a generic anonymous name along with a procedurally generated profile picture.

Please note that this won't work for all games out of the box, so developers can check Preferences.StreamerMode to ensure they will never network player strings/profile pictures when a client has Streamer Mode enabled.

Mount Maps Thumbnails

solwllms

You can play maps from supported mounted games, these now have thumbnails shipped with them to make selection much nicer. Enjoy 😎

If you're developing your own mount, SceneLoader now implements IThumbnailProvider loading thumbnails from /thumbs/{Host.Ident}/{RelativePath.WithExtension( ".png" )} You can mass generate them with the mount_generatethumbs concmd which takes a thumbnail screenshot from a gameobject tagged map_preview or falls back on a player spawnpoint.

Faster Package Downloads

Downloading games and assets was leaving a lot of bandwidth on the table, especially on faster connections.

We raised the parallel download cap from 16 to 64, and moved to HTTP/2 so those downloads share a few connections instead of opening 64 separate sockets. Small files now come down about twice as fast.

We also stream files straight to disk instead of buffering them in memory first. A 610MB package used to spike our heap to 739MB, now it sits at 3MB.

Input Glyphs & Improvements

devultj

We've added support for the recent Steam Controller, now we have a unique glyph library for both the Steam Deck & Controller. While in the area, I also added support for the Switch joycons too.

Also, the Steam Controller / other generic controllers were not reporting their name at all, now they do, should make it easier to tell your controllers apart.

Vulkan 1.3

We’re raising our minimum Vulkan version from 1.2 to 1.3. Although that sounds like a significant jump, we don’t expect it to affect the overwhelming majority of players or meaningfully change our hardware requirements.

In reality this matches Minecraft's new minimum specs which we feel is a very sane standard.

Moving to Vulkan 1.3 gives us a cleaner, more consistent foundation for rendering and lets us rely on functionality that is already standard across modern hardware. For almost everyone, nothing changes. There are no new visual settings to configure, no expected change in performance, and no need to upgrade your PC.

Vulkan Modernization

We've started modernizing how we use the Vulkan API.
Nothing you'll see directly, but it sheds a lot of legacy code and gives the driver better information to work with.

VK_KHR_synchronization2 is now in use across the board. Sync scopes are specified per barrier instead of one OR'd mask per call, so the driver sees the real dependencies and can overlap unrelated work instead of over-synchronizing.

VK_KHR_dynamic_rendering is now mandatory, which let us delete the legacy render pass and framebuffer paths entirely.

All of this has been well supported across the three major vendors for years, so it's mostly about clearing out old code paths and laying the groundwork for further upgrades.

Slang Shader Compiler

We've switched our shader compiler from DXC to Slang, this is the same compiler used in upstream Source 2.

There shouldn't be any breaking changes to your shaders, Slang is fully backwards compatible with HLSL but adds:

Reduced shader compile times by up to 25% with the exact same HLSL
- This can be further enhanced with modules, generics and interfaces.
Intellisense support via Visual Studio & VSCode extensions (which we were already using)
Generics & interfaces reducing shader combinations and scaling large code bases with more modular code.

As always we're going to aim to stay up to date with the latest technologies, be that rendering, .NET or more.

DLSS

I've added a new DLSS upscaler alongside FSR3. These are optional upscalers that makes use of temporal techniques, so it will exhibit minor ghosting and artifacts. There is no frame gen and no plans to add it.

Upscaler Fixes

GPUs implicitly select a smoother, smaller version of a texture when the neighbouring pixels of the textures are closer when rendering at lower resolutions, upscalers render the game at lower resolution, the combination of both was causing textures to look overly smooth when you were using DLSS or FSR.

By applying a bias on how these textures are selected we can have the game's textures have the same sharp look as native resolution.

Specular is now enabled by default in complex shaders

We've made specular enabled by default in all materials that use "complex" shader. If you ever authored materials with complex shaders, then you know that "specular" used to be an opt-in feature that you would have to manually turn on in every material that you create with complex.

Specular is what makes models shiny and fancy. Everything is shiny. It didn't make sense that such important part of shading was optional, and I know from my personal experience that this can be quite confusing for many beginner artists. Also, nearly all other shaders, including custom ones, have specular enabled by default.

Will this affect my assets? It's very likely that this will NOT affect your materials, and games will continue to look the same. However there may be an edge case where this *can* change how your materials look, but only if they meet specific criteria:

Material previously had "Specular" feature turned off
No roughness texture set, used default texture which equals 0.5 roughness

In this case models might gain a flat plastic shine. This can be changed by setting roughness value in material settings to 1.0, or adding a proper roughness texture, should be a pretty quick fix hopefully.

Add tint support to fur shader

Fur shader now supports tinting from model renderer component. This was previously not possible, but now that we finally have a human readable code in fur shader, we can start implementing new features to it. I imagine this will be handy for cosmetic creators.

Static Shadow Caching

Now that we can flag objects as static we can do some nice optimizations around it.

For shadow maps we only need to render static geometry once and then cache it, shadows will then only re-render dynamic objects in shadow maps and you don't need to do much configuration or baking.

Rasterizer behavior is exact the same, will still reject occluded depth pixels

Hair Transparency Fixes

We rely on alpha to coverage for order independent transparency, but that technique only works when MSAA is enabled. With MSAA off we were essentially falling back to screen door transparency, which gives ugly stippling patterns.

Now when MSAA is disabled we alpha clip instead. That means hair is no longer semi opaque, but it also gets rid of the stippling.

In the future we may look into other order independent transparency techniques that don't rely on MSAA, but until then we will warn in the settings if MSAA is turned off.

Bloom 3

Our second bloom implementation, the previous one suited better than what came with the engine and was also a good example of the our postprocess pipeline, but was hard to configure, after recent changes it also lost a lot of visual oomph.

Bloom 3 is looks even nicer and is easier to configure, this implementation is more like how modern games do it.

Small very bright details wouldn't bloom, they do nicely now:

This also proved to be about 25% faster in our benchmarks.

Shadows

A few updates back we added Screen-Space Shadows, we want to expose the concept of Shadow Masks more generally gradually, a way to defer part of the shadow calculations to compute shaders with a screen space texture, so they can decimated efficiently and composited correctly.

I've redone how we do shadow masks internally and this also fixed the 1 frame delay that caused stuff like Screen-Space Shadows in VR to be rendered in the reverse order, now it's all fine and the API should be pretty much almost good to be made public.

I've also fixed artifacts across quad divergences in shadows, we use a technique called Receiver Plane Depth Bias which uses the depth difference between neighbouring shadow maps texels in a quad to adjust the depth bias accordingly, the problem is that ordering of which shadow maps are in that quad can diverge in a quad giving garbage data, even when we try to mitigate it hardly.

I simplified it and just used a technique similar to what Unity and Godot do by offsetting the shadow position by the PCF size and surface normal, solves both shadow acne and the quad divergence shit

Hammer Aggregate Meshes

Hammer maps will now build aggregate scene objects from meshes and props, aggregates combine multiple meshes and props sharing a material into a single specialized aggregate sceneobject, so the runtime submits one batch instead of one per mesh.

This results in less draw calls and a faster path for drawing static geometry.

This is semi-backported from Source 2 upstream, but without meshlets or GPU culling for the first version.

Hammer Open World Vis

We’ve updated how visibility clusters are merged during map compilation, using proven settings from Deadlock.

Open maps could previously generate many small, neighbouring visibility clusters, even when those areas could already see one another. The new pre-merge passes combine suitable open-space and small regions before visibility sampling begins.

This should:

Reduce unnecessary visibility clusters
Improve visibility calculations in large, open areas
Produce cleaner visibility data during map compilation

Maps must be rebuilt to benefit from these changes. Because this affects visibility across every map, we’ll also be validating the results through map builds and visual sanity checks.

Why update Hammer?

Our scenes will soon have a compile step that makes use of aggregate meshes and props.

Backporting them from upstream Source 2 into Hammer gave us an opportunity to test them out in our worst performing cases of static geometry.

Bad faces

Hammer would highlight bad faces so you could sort them out, so now we do the same.

Mesh hide faces

As well as hiding gameobjects, you can now also hide mesh faces temporarily.

Primitive Tool UX

The primitive tool in our mapping tool needed to be sorted out so I collaborated with the community to come up with something nicer.

Add import scale with unit presets

A little user requested QoL feature, import scale is now included in the create model wizard.

ModelDoc: auto-fill material textures by name

I've updated how DefaultMaterialGroup node works and now it can automatically assign materials if names are matching. So if your model's material names match with materials in your project, they will be auto-filled.

Previously this auto-fill feature worked only if material slots in model had full content path in them, which is obviously not very convenient for artists. Now it also works with bare material names.

This helper will work as long as following conditions are met: 1) material slot hasn't been automatically filled by some other function already, like auto-fill by content path, 2) material name isn't ambiguous, meaning that there should be only 1 material with this name in your s&box project

I've added a new option to context menu that allows quickly creating new decal definitions from current selection of textures. Just select 1 or more images in asset browser, right click on them and click "Create Decal".

Engine will try to automatically assign all textures to appropriate decal slots from selection. The only requirement is that your decal textures must have appropriate suffixes appended at the end of their file name, such as: _color, _normal, _rma, _emissive, _height. For example, "my_custom_decal_color.png" will be assigned to a color slot.

You will be notified in a popup window if something went wrong during decal creation, this will usually happen if any of the decal textures have incorrect/missing suffixes. This tool should make decal creation a significantly less tedious process, turning it into essentially a fast two-clicks procedure.

Physics Event Callback Optimization