#Written from scratch custom VRAD - XenorRAD:tm:
1 messages · Page 1 of 1 (latest)
XenorRAD on the left and VRAD on the right
XenorRAD took 17 seconds to compile, VRAD took 35 seconds
both -final
fixed lightmap filtering, now light scatteres more evenly around geometry edges
also added adaptable shadow sampling, meaning near geometry edges it uses more shadow samples to resolve multiple light hits and uniformly smoothes them
for example this now is set to 24 near edge and 16 for rest
if from scratch write it on CUDA then or something else
would better
my poor little amd gpu won’t like that
oh shit nice
bro that pfp 💀
https://docs.scale-lang.com/stable/ this does exist same with ZLUDA
but yeah its actually better if you do this stuff via the gpu ngl
its more fit for that kind of things
that actually a interessing project, will be open source ?
so it seems to currently be missing radiosity then
VRAD on the left XRAD on the right, improve the algo
no
since i learnt that people are ungrateful
thats one quick way to kill the project
i want to make my own map for my own server
well, have fun 
and VRAD is just too slow to work with and doesnt quite meet my expectations but worth to note, its already 30x faster than VRAD without SIMD improvements and i think ill also add OpenGL to accelerate some parts
omw to make my own vrad in luajit with gpu shit 🚎
as if vvis++ didint exist
yeah vrad is easy to parallelize, vvis depends on an algorithm that is almost entirely serial
the rope
does that work on gmod
because vrad++ doesnt i think
p sure it does
yay
yall are really supportive
because i never seen since someone to rewrite the VRAD
not rewrite but from scratch VRAD
templeos is opensource though
fact
even king terry the terrible understood the value of opensource projects
i want to first beta test it on larger maps and with experienced mappers so i wont open source some trash
and some cleanup in the codebase also will be fit
I wonder how possible it would be to create an end-to-end replacement for the map compilation pipeline where each individual step is replaceable
thats entirely fair 👍
yeah, the modularity of different .exes being used does make it easier
u just have to make engine compatible bsp
byte by byte identical literally
or else you gonna end up like me with 1 gb crash log folder until i got it just right
I experimented with VoxelGI instead of radiosity for gmod
but its hard to do so i scrapped it and continuing with just XenoRAD which is basically modern and efficient VRAD
How large are the final lighting textures?
basically i voxelized the whole map and calculated so by inject each voxel and doing light propagations
but its was so blocky that denoising and smoothing it would be inefficient and slower than regular radiosity
4900 produced against VRADS 8600 also the whole testing map is using power of 2 luxel scale
also im only shipping HDR only
I mean in terms of resolution and disk size
vrad does about 8mb whilst mine did 5mb
so it makes huge difference in larger maps
but its too early to compare it
since i also need to implement static props lighting
geometry lighting is now fully complete at this point
safe to say its better than VRAD could ever be without huge rewrite
it fully work on gmod, just remove your viscluster and go use it !
Make it open source.
Even if the codebase isn't in a great state.
Making it FOSS will be beneficial in the long run.


i forced vrad++ to work on gmod
workaround using sourcesdk2013mp
just mix and match your compilers til it’s all fine and dandy ❤️
how much faster is this gonna be than vvis++
how unoptimized and overly complex are your maps for gmod of all things that vvis is main point of slowdown/bottlenecking in your map compile
real
I never ran into vvis issues, but vrad was really slow in SDK
city map with open skybox
damn
i gotta make it GPU accelerated
cuz its only like approx. 10-15 seconds gains
also gonna add SIMD for vector math operations
GPU accelerated might be able to do it real fuckin’ quick
but its just direct lighting i need to also implement indirect
Added path tracing for indirect lighting calculations
This is 1024 rays per luxel
I now dont need to do Ambient occlusion pass since AO is results of path traced lighting similiar to blender cycles
This is 512 rays per luxel
14.9 million rays per second with total of 2,08 billion total rays casted
somewhat reasonable altough RTX cards do it like 100x faster
both with 16 progressive passes
holy shit
If it's not time consuming to do, it would be interesting to see a series of comparison screenshots taken from the same position/angle/fov with different ray counts
I'm curious about how quickly you get diminishing returns on adding additional rays
If the camera is aligned exactly the same between the images, it would be possible to compare two ray counts (e.g. 1024 vs 512) and subtract the pixels of the 512 image from the 1024 image to visualize exactly what the additional rays add to the final 1024 image
ill do it today or tomorrow
im guessing this is a private compiler
yes
oh i meant like open source, yeah im planning to release it, already got 2 beta testers
you planning to release it as a executeable?
yes
based
how fast in percentage wise is it compared to stock vrad, gmod vrad, vrad++?
map that im testing which is like 30 mb in vmf format compiles like in 49 minutes whilst mine without path tracing just direct radiosity did it in 6 seconds
and vrad++ is just vrad from source sdk 2013 improved
mine is literally written from scratch and i just use the same lightmap rgb encoding function since garrys mod is very picky to literal bytes
yeah
but it shouldnt be surprising since VRAD was written in like 2003 and was targeting Windows 95
mine is written using bleeding edge and the latest compiler tools since its written completely from scratch in Visual Studio 2026
but it wasnt like week project i was working on it for like 3-4 weeks dealing with constant crashes of gmod and bugs
and its still not complete for example i have to finish prop_static lighting and their shadow casting
idk really i just talk a lot in hl2rp hub and sometimes post executable for everyone to test
they usually give me feedback for example synapse t6 showed a larger gaps in my compiler which caused me to rollback to older version and rewrite the computer shader
once i will need testing with the new path tracing im gonna dm you with exe
there is a lot that might need implementation such as texture shadows, prop texture shadows etc
based 🙏
good work
@grand cloak keep us updated
threw away OpenGL and swapped to Vulkan API
also im gonna drop support for path tracing since its computionally very expensive and only people with RTX cards would have great compile times
okay
Partially implemented prop_static
prop_dynamics isnt handled (as far as i know) by the VRAD so these are untouched
the XenorRAD is now fully utilizing Vulkan Graphics API altough it needs a bit more care since Windows TDR doesnt like it (it falsely identifies it)

Added ETA approximation when the map compiliation will be finished in each batch dispatch
how does this compare to vrad++
vrad++ is just some fixes ontop of vrad isnt it
Its written from scratch using Vulkan Graphics API running on your GPU
VRAD on steroids
and using path tracing to calculate global illumination
based on that article
instead of 30 minutes of waiting to compile map its done under 30 seconds
i want to enter map contest later with my vrad
nice
I have decided to put this project to halt and instead directly improve and gpu accelerate vrad from source sdk and improve vbsp and vvis.
I will make another thread in waywo once I will have a ready to install version of vrad for public testing.
can i have mor rooms on darkrp_downtown_tits with this
actually not bad idea
i already have my own version of luajit where i introduced preprocessing statements and other modifications
meh dont edit the syntax pls, fuck that
but yeah adding lua into tools is POG
i added fn as alias for function keyword, so both function and fn are valid but im not gonna be changing syntax, only added // as valid comment, same as gmod
That's a great idea but rn ur vrad makes funky gi, you should make you own discord server for the project.
Idea is great but as any project it needs a lot of work done before release.(I started similar proj myself but it uses path tracing for baking, I won't finish it any time soon tho.)