[Preview 2] Unreal Engine 5.9 Preview

I’ve signed up for the game jam, but I don’t currently have 200GB of disk space available to compile the engine for WebAssembly projects. Could you please upload a preliminary version that includes WebAssembly support, so that compiling the engine isn’t necessary? I’m really glad to see a modern version of Unreal with HTML support; I’ve consistently used HTML with Unreal for most of my projects on itch.io—specifically versions 4.23 and 4.27 (the SpeculativeCoder version).

An important aspect of the browser is its support for mobile touch; a well-optimized build for mobile should work very smoothly on any reasonably modern smartphone. I encountered quite a few issues when trying to emulate touch input in older versions of Unreal. Joystick support is also somewhat poor.
I was unable to get the API implementation—for example, for Facebook Game Room—to work.
I’m glad to see HTML making a comeback in Unreal!

Okay, now the question everyone wants answered about your new vibe-coded version of UE: does it start 4.6 seconds faster than Steam?

Let’s go sweeney, you can do it!

2 Likes

lmao this actually made me laugh out loud ty.

Soooo, to be clear, does the 5.9 preview release currently available on the launcher support shipping with WASM out of the box, or do you need to download the source version over at github to be able to use WebAssembly/HTML5 as a target?

I was about to say “don’t listen to the haters,” but… I can’t even find UE 5.9 for the Jam in the new vibecoded Launcher. So I guess the haters kinda have a point, and maybe you should listen to them.

Ouch, I don’t see that!

(And yeah, I have 5.8.4 already installed… haha)

Whops, and it’s showing it now, but in the normal category O.o

1 Like

why even let real ppl join the game jam, just have agents do it themselves at this point lmao. I’ve used Unreal since the first UDK version and its decline is very sad to see, please just keep the slop for 6 and make an actual LTS version for 5, that’s all we ask. 5.8 is great but i am not motivated to upgrade to 5.9 except for the wasm support

8 Likes

They need to press the button :joy:

But yes, please Epic, finally make a LTS version of Unreal. Thank you.

3 Likes

all I can say is, this is sad. UE used to be a powerful tool for motivated and talented devs, and now it’s being pitched as a middleware for slop. Not everything needs to be AI guys, I wish you could see that and save your engine from becoming completely irrelevant and hated by both devs and players.

5 Likes

The Model Context Protocol is missing in the Editor Preferences in 5.9 - How do we connect an LLM to this new version?

(post deleted by author)

Are you accepting Linux builds ?
It looks like you did not add Linux on the list:

This is the release I’ve been waiting for. I was in the middle of converting to Unity to use WebGL builds, but WebGPU is better and I don’t have to convert the project!

What about NvRTX? Is it packed in the 5.9 MegaJam too?

Hm? The Nvidia branch? @automata

Thanks for the preview and for being upfront that parts of the web renderer are prototypes. We have
been testing the WebGPU target on desktop Chrome and wanted to ask about one specific item.

In WebGPURHI/Private/WebGPUPrivate.h (5.9 preview), the threading comment says context commands are
currently queued as one closure per command, and that “P9 replaces them with recorded POD streams”.
Is that step planned for the full 5.9 release? If so, is there a rough timeline, or a branch we
could follow to track it?

Why we ask: on a large interior scene (mobile renderer, deferred tier) the frame is bound by the
browser main thread that executes the WebGPU calls, not by the GPU. The GPU takes roughly 6 ms of a
~40 ms frame. Main-thread cost fits roughly 15-18 ms fixed plus about 4 microseconds per draw (a
rough fit, r2 about 0.4, from Development builds with WebGPU stats enabled, so treat the absolute
numbers as approximate). At a heavy view that is around 4,300 draws and 12,600 WebGPU commands per
frame, and in a Chrome profile a large share of main-thread samples land in malloc/free. We suspect
the per-command closures contribute to that, but we have not proven it and are still profiling.

We can share a trace if useful. Any pointers on where this work is heading would be appreciated

[5.9 Preview][Wasm/WebGPU] Masked materials on VT-lightmapped meshes: black specks (deferred) / white fireflies and bloom discs (forward) along the cut-outs. Cause and a tested 2-line shader fix

Engine / platform: UE 5.9.0-60351404 (Launcher), Wasm target, WebGPU RHI, WGSL shader format, mobile renderer, both web tiers (deferred and forward). Chrome 157, Windows 11, NVIDIA RTX 4080 SUPER. A Win64 (D3D12) package of the same project is clean.

Symptom. Masked materials (opacity mask from a texture alpha, e.g. foliage-style cut-out decals) on static meshes with virtual-textured lightmaps show specks along the cut-out edges: black on the deferred tier; very bright on the forward tier, where bloom turns them into large white discs. Turning off lights, sky light, reflections or GI does not remove them; they remain with only emissive left.

Repro (from any template). r.VirtualTextures=1, r.VirtualTexturedLightmaps=1, static lighting built, a lightmapped mesh with a masked material whose mask is a texture alpha with a fine pattern. Package for Wasm and look at the mask edges in Chrome (forward tier shows it most).

Cause (engine source, 5.9.0)

  1. Engine/Shaders/Private/MobileBasePassPixelShader.usf:406-409: the masked clip (GetMaterialCoverageAndClipping) runs near the top of the shader for every compiler except GLES; :1318-1321 moves it to the end only for COMPILER_GLSL_ES3_1 (an Adreno workaround). COMPILER_WGSL (set at ShaderFormatWGSL.cpp:3777) is tested in neither block.
  2. The VT lightmap lookup comes after the clip and takes screen derivatives of the lightmap UV: VirtualTextureCommon.ush:169-180 (MipLevelAniso2D), 0.5f * log2(min(px, py)) / 0.5f * log2(max(px, py)).
  3. In WGSL the clip becomes discard; return;, and derivatives in a partly discarded 2x2 quad are undefined (the module also carries diagnostic(off, derivative_uniformity), so nothing warns). A zero or infinite derivative gives -inf - -inf (or inf - inf) = NaN in the mip fraction; the NaN blends the two lightmap mips and flows through the whole indirect term (LightmapCommon.ush:78-129). NaN * 0 stays NaN, which is why no lighting toggle removes it.
  4. Forward: SafeGetOutColor (MobileLightingCommon.ush:68-75) does min(OutColor, Max111110BitsFloat3 * 0.5f), which turns NaN into a huge finite value (firefly, then bloom). Deferred has no clamp and shows black.
    Desktop BasePassPixelShader.usf does the VT lightmap lookup (:947-953) before its clip (:1005-1012), which explains why Win64 is clean (from source; DXIL not inspected).

Cooked WGSL of an affected forward permutation (measured; condensed excerpt, nothing between the discard and the first dpdx derives):

if (((em.w - 0.3333f) < 0.0f)) { discard; return; }
...
let ev = dpdx(z);  let ce = (dpdy(z) * x);         // lightmap UV derivatives
let ew = (0.5f * log2(min(cf, cg)));  let ch = (0.5f * log2(max(cf, cg)));

Measured (interior scene, 14 fixed camera poses, Win64 package as reference)

  • Forward: largest near-white blob on frames with the cut-out decals up to 142,976 px (Win64: 2,064-5,792 px at the same poses). With bloom off and only emissive left, 15,870 px >= 100 in one decal crop.
  • Deferred: 385 exact-black (0,0,0) pixels in the same crop (Win64: 0).
  • Shader census of our cooked libraries: 266 pixel shaders in 34 shader maps (forward) and 60 in 30 (deferred) have a texture-alpha discard before a derivative, so other masked cut-outs (labels, logos) have the same pattern; only the decals were frame-checked.

Proposed fix (treat WGSL like GLES; tested by us):

-#if !EARLY_Z_PASS_ONLY_MATERIAL_MASKING && !COMPILER_GLSL_ES3_1
+#if !EARLY_Z_PASS_ONLY_MATERIAL_MASKING && !COMPILER_GLSL_ES3_1 && !COMPILER_WGSL
 	//Clip if the blend mode requires it.
...
-#if !EARLY_Z_PASS_ONLY_MATERIAL_MASKING && COMPILER_GLSL_ES3_1
+#if !EARLY_Z_PASS_ONLY_MATERIAL_MASKING && (COMPILER_GLSL_ES3_1 || COMPILER_WGSL)

Only MobileBasePassPixelShader.usf changes; we applied it to a copy of the 5.9.0 Launcher engine (shader text only, so no C++ rebuild) and recooked the mobile shaders. Result, same scene, one load per arm, unpatched build loaded in the same session as control:

  • Census: clip-before-derivative pixel shaders 266 → 0 (forward), 60 → 0 (deferred).
  • Forward emissive-only crop: 15,870 → 0 px. Largest forward blobs 143,120 / 54,144 / 59,456 → 48 / 3,152 / 4,432 px (Win64 2,064 / 3,488 / 5,792).
  • Deferred black specks: 135 → 0 px (one crop), 1,086 → 1 px (another pose).
  • No regression seen: mean CIEDE2000 vs Win64 7.25 → 6.67 (deferred), 9.45 → 6.35 (forward); no surface missing in a 14-pose eye check; cut-outs keep their shape. Per-pose differences were dominated by camera framing between loads.
    Optional hardening: log2(max(..., 1e-30)) in MipLevelAniso2D (catches a zero derivative, not an infinite one), and a NaN-safe SafeGetOutColor (min(max(OutColor, 0), ...)), which would only hide the defect.

Not a good workaround (measured): r.Mobile.EarlyZPass=2 removes the base-pass discard and the specks (15,889 → 0 px), but in our scene several large opaque surfaces disappeared with it (mean CIEDE2000 vs Win64 6.5 → 16.6). We did not investigate why.

Not measured / inferred: the actual derivative value Dawn returns in a partly discarded quad (we infer 0 or Inf; a finite wrong value would only give a wrong mip); Safari / iPadOS, Firefox and Android. Content-side alternatives (translucent cut-outs, or those meshes off VT lightmaps) were not tested.

Happy to share more data. Thank you!

1 Like

[5.9 Preview][Wasm/WebGPU] Virtual-textured lightmaps are sampled through an sRGB view: lightmapped scenes render ~1-1.5 EV dark with per-object tints

Engine / platform: UE 5.9.0-60351404 (Launcher), Wasm target, WebGPU RHI, mobile renderer, both web tiers (deferred and forward). Chrome 157, Windows 11, NVIDIA RTX 4080 SUPER. A Win64 (D3D12) package of the same project and cook settings is correct.

Symptom. Surfaces lit by baked lighting are much darker than on Win64, and surfaces that share one material come out with different tints (one object greenish, its neighbour bluish). Emissive surfaces are fine, so exposure and tone mapping are not the cause.

Repro (from any template). Enable r.VirtualTextures=1 and r.VirtualTexturedLightmaps=1, build static lighting in an interior scene, package for Wasm and Win64, and compare the same camera in Chrome and on Win64.

Cause (engine source, 5.9.0)

  1. Engine/Source/Runtime/RHI/Public/RHIGlobals.h:285: SupportsTextureViews = true (GRHISupportsTextureViews). The WebGPU RHI never sets it (no write under Engine/Platforms/Wasm; only OpenGLDevice.cpp:901 sets it false).
  2. Engine/Source/Runtime/Renderer/Private/VT/VirtualTextureProducer.cpp:96-97: with texture views “supported”, every VT physical layer gets an sRGB view, so VirtualTexturePhysicalSpace.cpp:153-157 creates every physical texture with TexCreate_SRGB, and the linear SRV is requested with SetDisableSRGB(true) (:177-183).
  3. Engine/Platforms/Wasm/Source/Runtime/WebGPURHI/Private/WebGPUResources.cpp: CreateTextureObject (:145-146) creates a TexCreate_SRGB texture directly in the sRGB format and adds viewFormats only for storage textures (:204-211). The SRV path (:1629-1633, GetView :706-724) always uses the texture’s own format, so DisableSRGB is ignored: the “linear” view is sRGB.
  4. Engine/Shaders/Private/LightmapCommon.ush:78-128: LQ lightmaps are log-encoded (L = exp2(16*Luminance(texel*Scale+Add) - 8)), so an sRGB-decoded texel loses stops, and with per-allocation, per-channel Scale/Add the error differs per object and per channel (the tints).

Any other linear VT layer in an sRGB-capable format (BC1/BC3/BC7/BGRA8 masks or normal maps in streaming VTs) should be affected the same way (inferred, not checked).

Measured (interior scene, 14 fixed camera poses, Win64 D3D12 package as reference)

  • Scene-linear median brightness, Wasm vs Win64: -1.23 EV mean (deferred), -0.98 EV (forward); 9 of 14 poses at -1.3 to -1.8 EV.
  • Lightmap light alone (GI toggled off and subtracted): 0.27-0.39 of the Win64 value.
  • Offline: decoding the cooked LQ lightmap texels through an sRGB curve reproduces -1.46 EV on the measured page; decoded linearly they match the HQ lightmap within -0.12 EV.
  • Format-independent: an uncompressed (BGRA8) lightmap build and a BC7 build look the same.
  • No dependence on local luminance (fit slope -0.14), so not local exposure.

Workaround we ship (project module, Wasm only; measured)
GRHISupportsTextureViews = false from PostConfigInit. The VT system then creates each physical space by the layer’s own sRGB flag, so linear layers get linear textures; no recook needed.

  • Caveat (measured): a plain global false breaks the mobile deferred tier. Its HZB compute pass (SceneTextureReductions.cpp:345-355) then binds a whole-texture SRV of MobileHZBFurthest while writing its mips; WebGPU rejects it (“includes writable usage and another usage in the same synchronization scope”) and every frame’s command buffer is invalid (black output). The forward tier is fine.
  • So we set the flag back to true for the duration of each scene render (an FSceneViewExtensionBase: true in PreRenderViewFamily_RenderThread, false in PostRenderViewFamily_RenderThread). It works but is a hack: a VT producer registered from another thread during a scene render would still get an sRGB pool (inferred; not seen).
  • Result: median -1.23 → +0.10 EV (deferred), -0.98 → +0.20 EV (forward); lightmap light 0.27-0.39 → 1.05-1.11 of Win64; mean CIEDE2000 vs Win64 21.0 → 6.6-8.3 (deferred, 2 loads; the rest is camera framing and mobile-vs-desktop renderer differences); the per-object tints are gone; 0 WebGPU errors in 6 loads. The same binary with the workaround switched off stays at -1.22 EV.

Suggested engine fix (not built or tested by us; the Launcher engine ships the RHI precompiled)
Make the WebGPU RHI honour DisableSRGB: for a sampled TexCreate_SRGB texture, create the texture in the linear format with the sRGB format in viewFormats (as WebGPUResources.cpp:204-211 already does for storage textures), and add the sRGB choice to the view key in GetView (:706-724). Texture views then stay “supported” and nothing else changes behaviour.
The one-line alternative (GRHISupportsTextureViews = false in WebGPUDevice.cpp) also needs the HZB and capture-mip fallbacks (SceneTextureReductions.cpp:283-355, ReflectionEnvironmentCapture.cpp:479-490, ReflectionEnvironmentDiffuseIrradiance.cpp:207) to stop sampling a whole-texture view of a texture they are writing.

Not checked: Safari / iPadOS, Firefox, Android; other VT content types (only lightmaps were measured).

Happy to provide more data or test a fix. Thank you!

The added Kimodo plugin does not work on my GPU RTX 5090 due to the pytorch installation not supporting sm_120.

I’ve attached logs which I ran python commands in to try and troubleshoot.

[2026.11.15-18.35.39:567][297]LogPython: import torch
print(torch.cuda.is_available())
if torch.cuda.is_available():
print(torch.cuda.get_device_name(0))
print(torch.version.cuda)
[2026.11.15-18.35.39:587][297]LogPython: True
[2026.11.15-18.35.39:589][297]LogPython: Warning: C:\Users\Connor\Documents\Unreal Projects\AIAnimation\Intermediate\PipInstall\Lib\site-packages\torch\cuda_init_.py:235: UserWarning:
NVIDIA GeForce RTX 5090 with CUDA capability sm_120 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_61 sm_70 sm_75 sm_80 sm_86 sm_90.
If you want to use the NVIDIA GeForce RTX 5090 GPU with PyTorch, please check the instructions at Get Started Locally - PyTorch
warnings.warn(
[2026.11.15-18.35.39:590][297]LogPython: NVIDIA GeForce RTX 5090
[2026.11.15-18.35.39:590][297]LogPython: 12.4
[2026.11.15-18.36.18:565][422]LogPython: import torch
print(torch.cuda.get_arch_list())
[2026.11.15-18.36.18:565][422]LogPython: [‘sm_50’, ‘sm_60’, ‘sm_61’, ‘sm_70’, ‘sm_75’, ‘sm_80’, ‘sm_86’, ‘sm_90’]

A workaround I managed to get working is disabling run pip install on startup located at Edit > Project Settings > Plugins > Python.

To prevent it overwriting me forcing it to install pytorch 12.8 from pytorch 12.4 which the editor forces.

And then I can run “C:\Unreal Projects\AIAnimation\Intermediate\PipInstall\Scripts\python.exe” -m pip install --upgrade torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128

And it works.

A proper fix would be shipping unreal 5.9 with pytorch 12.8.

1 Like

Hi, Stone Labs Studios here (Rate Pengi Laced Up by stonelabsstudios for 2026 Epic AI MegaJam - itch.io). After the deadline a player found a blocker: an invisible collision fence seals the entrance to battle 6, so the game can’t be finished. The fix is a one-file collision change, with no new content or features. Could you unlock uploads for our entry so we can replace the build, or tell us if this isn’t allowed? Happy to share the diff. Thank you!