Hey guys,
Lovely showcase of the new technology and the future of Unreal at Chicago Fest 2026!
I’ve been experimenting with ray-traced translucency in my project by enabling it in the Post Process Volume. The main issue is that it’s currently too expensive for real-time games. According to Stat GPU, it’s costing around 8ms.
I don’t really want to switch back to raster translucency because other translucent objects don’t appear correctly through my front-layer translucent materials, and mirrors/reflections also don’t show up properly. Ray-traced translucency solves these issues and displays everything correctly.
As far as I know, ray-traced translucency was introduced in UE 5.6 and hasn’t seen much development since then. I tried using the cvar r.RayTracing.Translucency.DownsampleFactor and setting it to 2. This reduces the cost from roughly 8ms to around 2ms, which is great, but the image becomes very noisy and shimmery. Since the cvar is marked as experimental, I assume the implementation is still a work in progress.
I’d like to know whether there is any active development planned for ray-traced translucency and whether we can expect the quality of Downsample Factor 2 to improve by UE 5.9 or future versions.
Thanks!
While I don’t know the active status of RT translucency development, I would just like to offer a small bit of insight as someone who’s also been following work like this for a while.
Real-time translucency is a REALLY hard problem in computer graphics, and while RT translucency is a great solution for a number of reasons, it’s also a staggeringly unperformant one. Lumen works by caching diffuse hits in the surface cache, and while there’s several promising attempts to cache specular hits via neural networks or other spacial partitions, none of them are in production as far as I know. As per Daniel Wright’s OG lumen presentation, consoles really have a 0.5 ray-per-pixel budget, and any kind of translucency would blow that budget apart. That’s not even considering the fact that translucency means constantly evoking anyhit instead of the faster inline tracing that makes Lumen possible on consoles, and needing to smartly compensate for lost energy because even four bounces of refraction in a glass of water may not be enough to resolve radiance.
The downsample factor makes a big difference because you’re more or less cutting your performance costs by 3/4 (hence 8ms to 2ms), but because of the wildly divergent paths that refraction ways take, upsampling and denoising is an even harder problem than with diffuse GI or reflections. What I’m saying is that I’m not sure this is a problem that Epic will actually invest significant resources in, because it’s such a hard problem and using it in games makes little sense. I think RT translucency largely exists for cinematics and offline work that wants faster rendering than the path-tracer, but higher quality than 60FPS lumen.
I love this stuff too and would love for them to make it work, but I’m not particularly optimistic that they’ll get to it anytime soon. I’d love to be proven wrong 
rt translucency is very expensive. watch a “how pathtracing works video” on youtube. think about the rays you have to shoot. if you hit a surface like that, computationally you have to generate a ray for the refraction and the reflection. that’s already 2 rays per pixel you have to shoot into a bvh structure and trace and solve the texture and lighting equation (the bsdf) at the next hit point. doing this in lower scale and temporally accumulated is currently the fastest solution. the amounts of ray bounces are gnarly.
that being said… in my test setup i shot some rays thru a normal mapped refractive glassball into a mirror and into a window to capture something outside the window. it’s pretty good, but in fullscreen it is taxing the rt cores alot. even more if your geometry is nanite and has alot of detail.
it works tho. with low geometry detail. on a 3060 with just 6 GB of vram. albeit framey. gotta be used well authored. hmm.
While I understand that ray-traced translucency is expensive, I would love to see Epic add another quality option between Downsample Factor 1 and 2.
Similar to how Lumen has r.Lumen.Reflections.DownsampleCheckerboard, it would be great if ray-traced translucency had something like r.RayTracedTranslucency.DownsampleCheckerboard 1.
However, Downsample Factor 2 looks extremely low resolution and noisy like a 360p image.
A middle-ground option could potentially reduce the cost to around 4 ms while providing noticeably better image quality, stability, and less noise than Downsample Factor 2. I know downsampling is more complex than simply lowering resolution, but having an intermediate checkerboard or hybrid mode could make ray-traced translucency much more practical for games.
Alternatively, improving the visual quality of Downsample Factor 2 would also be a welcome improvement. Some noise is expected, but if the result were closer to a 640p–720p quality level instead of looking extremely low resolution, it would be much more usable in real-time projects.
Hopefully Epic considers adding a feature like this in the future.