June 2025 Progress Update

• Added user customizable settings for stars; star size and star density. (Settings are available in Atmospheric Settings > Atmosphere Settings)
• Added block light clipping to the campfire to improve visuals in challenging lighting conditions
• Fixed unmapped blocks (such as glow lichen) being treated as lightning bolts and glowing far brighter than intended
• Fixed various small bugs, and change some description text to better match Continuum Legacy’s latest versions

And a few screenshots of this release:

dotModded has been swamped with personal / health stuff as of late, so I chose not to bother him for a direct update this month, but we’ll still be hearing from Jake later on! Before that though, dotModded’s work on the 1.20.4 version of Focal finally reached our final testing stage, just not quite soon enough to make the cutoff for June’s updates. We’ve been testing and fixing small bugs with that internally for a bit now, and are fairly confident it will be ready to go for July’s update window (maybe a bit sooner  👀).

Dot’s focus is now on finishing the 1.21 Focal update, with some remaining UI and mapping work left to do there. If life lets up a bit, we should be able to get that out the door soon as well 🤞. Once that’s done, we can shift almost all of our Focal Engine development focus back onto fully decoupling ourselves from Optifine, which should further help simplify and speed up our update process (among other benefits)…more info on that soon™.

But enough from me, here’s Jakemichie:

Alright so, if you read last month’s progress update, you’d know that I was chasing down an issue with how POM interacts with forward projection in the Continuum RT rewrite, causing temporal gradients to erroneously fire and reject lighting within POM as you move the camera. To keep it short: my idea didn’t work. I suspect this is due to subtle differences in how pixels are reprojected during the forward projection pass, but I don’t know for certain. Either way, rather than wasting even more time trying to come up with a new fix that we’re not even sure will work, we’ve decided to just accept this issue for this release. To reduce the impact of it, we have split off light interacting with POM into its own setting that is disabled it by default. We’ll likely come back to this issue later, which I’ll touch on a bit below when I discuss the possible short term future of Continuum RT, but for the time being you will have to deal with it, assuming you enable the POM Lighting setting and load a texture pack with heightmaps of course.

Unfortunately I burned through most of the month trying to fix this one issue, eventually throwing in the towel only a few days before we usually get builds ready for release. Near the end of the month we uncovered several issues of varying severities with the shader, including a pretty severe issue that will require a Focal Engine update for the 1.20.1 and upcoming 1.20.4 builds to fix. We seem to have fixed it internally and we’re working through the issues with the shader, but we may take some more time to polish it as best we can with the time that we have. Hopefully we can get an updated build out soon, but I guess we’ll see.

Once the rewrite is released, most of my time will likely go back to Focal Engine. We’ll still release smaller RT updates, but for the time being we can’t really make substantial updates to the shader without further Focal Engine updates. We do have a bit of a short term roadmap for Continuum RT, but most of it requires us to at least move away from Optifine’s settings.

As a bit of a teaser of what we have planned, we want to:

• Overhaul the voxel data structure that we use. I’ve been looking at something along the lines of a 2- or 3-level brickmap/64-tree, which looks to strike a good balance between traversal speed, memory usage and editability. We’re still not 100% sure what our traditional shaders will look like further into the future, but it does look like we may adopt some form of voxel data structure in an eventual Continuum 2.2/3.0, so I’m also thinking forward a bit with the one that we will end up using in Continuum RT before we move to Vulkan and gain access to hardware RT.
• Overhaul the path tracer. I’m still not sure which direction we should go here, but the current one is just utterly abhorrent when it comes to performance. I’ve tried to squeeze as much performance out of it as I can, but its very design just doesn’t scale well on GPUs. The two directions that I’m thinking of is to either move to a global ray buffer (which may have some L2 contention issues with atomic accesses), or ditch the recursive design and move back to a pure stochastic path tracer (which may increase noise). The latter also opens up in-place ray sorting, which could be a pretty decent performance win.
• Replace A-SVGF with NVIDIA’s ReLAX denoiser. A-SVGF is a pretty elegant denoiser and works extremely well, when it works. Unfortunately there are situations where it doesn’t, as the forward projection is very temperamental and doesn’t efficiently scale with the number of independently reprojected signals (in the Continuum RT rewrite we currently have 3 signals, but forward projection only uses 1 as the basis for reprojection). Looking at the literature, it seems like A-SVGF is a bit of a dead end, so we’ll probably move off of it. ReLAX looks to be a solid alternative for us, as it’s based on SVGF (not A-SVGF) with a number of improvements, including one that’s sort of a sidegrade/minor downgrade to temporal gradients (temporal gradients respond to lighting condition changes within 1 frame, ReLAX responds within several frames).

That’s all for now though, see you all next month!

A few screenshots we’ve taken while testing RT Build 21 this past month, including our first reveal of Weather in Continuum RT, and a comparison of the default POM, as well as POM + the POM Lighting setting (and it’s issues), we mentioned above: