https://fxtwitter.com/_x1m4/status/1815747514844291122

💬 3 🔁 8 ❤️ 105 👁️ 3.9KRendering portals becomes straightforward if your engine is based on ray tracing! #gamedev #webgpu

xima (@_x1m4)

Rendering portals becomes straightforward if your engine is based on ray tracing! #gamedev #webgpu

AquaOP•7/23/24, 3:40 PM

this stuff annoys me to no end

AquaOP•7/23/24, 3:41 PM

ray tracing has nothing to do with this

AquaOP•7/23/24, 3:41 PM

this is just a render target and a second camera

AquaOP•7/23/24, 3:41 PM

infact you can literally tell because the portal does not "recurse"

Greem•7/23/24, 3:42 PM

i thought that was a minecraft shader

AquaOP•7/23/24, 3:42 PM

it does look it

Greem•7/23/24, 3:42 PM

hotbar and everything

AquaOP•7/23/24, 3:43 PM

very original indeed

AquaOP•7/23/24, 3:43 PM

normally i would not say anything about this being an obvious minecraft clone but don't brag about it being something it's not

Greem•7/23/24, 3:44 PM

agreed

Greem•7/23/24, 3:44 PM

Lol this reply https://fxtwitter.com/anarchyarcade/status/1815751879017796028

FxTwitter / FixupX👁️ 101

Anarchy Arcade (@anarchyarcade)

@_x1m4 Ironically, it's also straightforward if your on an ancient cell & portal engine like Sith engine :D

AquaOP•7/23/24, 3:45 PM

haha, yeah

AquaOP•7/23/24, 3:45 PM

portal did this infinitely better as well, you can create far better portals using the stencil buffer

Greem•7/23/24, 3:46 PM

No modern GPU necessary :weirdchamp:

AquaOP•7/23/24, 3:46 PM

you can easily "recurse" them, plus the resolution doesn't decrease when you get closer, and there's no additional memory needed for the potentially 4 or 5 render targets if you are using deferred rendering

GGreem No modern GPU necessary :weirdchamp:

AquaOP•7/23/24, 3:46 PM

yeah lol

Endlight Dev•7/23/24, 3:48 PM

hello again

back from school.. thinking about OpenGL and Silk.NET

yesterday i started working on adding textures to objects

AAqua ray tracing has nothing to do with this

TechPizza•7/23/24, 4:23 PM

wdym? when a ray hits a portal you can just move it over, that's why it's trivial

REM•7/23/24, 4:25 PM

yepp, and the reason it does not recurse is because there is a limit to how many times the ray can be: reflected, refracted and moved/teleported

AquaOP•7/23/24, 4:26 PM

to me, this looks like it is quite obviously just a render target

TechPizza•7/23/24, 4:28 PM

they could show it off better if they got more light through the portal

AquaOP•7/23/24, 4:30 PM

if they can prove it is raytracing then i would love to see that. although i don't really know what the point would be considering that you can do the same with a render target or stencil buffer

AquaOP•7/23/24, 4:31 PM

but i'd be happy to see me proven wrong

TechPizza•7/23/24, 4:32 PM

it would just be a lot more unnecessary work to add render targets when dealing with rays already

AquaOP•7/23/24, 4:32 PM

i suppose if the entire scene is raytraced then yeah

TechPizza•7/23/24, 4:32 PM

i guess a good showcase would be putting an object in the middle of the portal

AquaOP•7/23/24, 4:33 PM

yeah

TechPizza•7/23/24, 4:33 PM

you would not need to render objects twice like with render targets, it should "just work"

TechPizza•7/23/24, 4:33 PM

although not really because it would be in two locations at once...

TechPizza•7/23/24, 4:34 PM

raster enjoyer yells at rays

AquaOP•7/23/24, 4:34 PM

rofl

AAqua https://fxtwitter.com/_x1m4/status/1815747514844291122

queenvicky•7/23/24, 4:34 PM

I mean you could do this recursively with N render targets and then put them into each-other. Wouldn't that allow portal recursion (You set a recursion-max yes, but you also set a ray-bounce-max and there's no point in going beyond a certain depth)

AquaOP•7/23/24, 4:35 PM

you can, that would be potentially a lot of memory though

queenvicky•7/23/24, 4:36 PM

you really don't need a high resolution after the first idk. 3 cycles

AquaOP•7/23/24, 4:36 PM

the best way to render portals (in my own opinion) is absolutely the stencil way though. no additional render targets, just additional draw calls

queenvicky•7/23/24, 4:37 PM

I don't know if it's the best - but I like the stencil buffer so yes

AquaOP•7/23/24, 4:37 PM

portal uses it, imo that's about all the endorsement it needs lol

sowa705•7/23/24, 4:54 PM

https://ai.meta.com/blog/meta-llama-3-1/ who has 4x H100 to run a 405B model?

Meta AI

Introducing Llama 3.1: Our most capable models to date

Bringing open intelligence to all, our latest models expand context length, add support across eight languages, and include Meta Llama 3.1 405B— the first frontier-level open source AI model.

TechPizza•7/23/24, 5:21 PM

should get one of those GPUs with an nvme slot and use flash as memory :zigfast:

sowa705•7/23/24, 5:29 PM

I'm testing the 8B now and it's pretty awful, hard to say if its the model being bad or the software is broken currently