Platinum — Path Tracing in Metal

[winged veldt]

[quick zealot]

oh okay I'll try

is the scene somewhere before you eep?

[winged veldt]

#ray-tracing a while back

[quick zealot]

got it

Nothing at all 🙁

With sampling on emissives allowed

Same without MIS

[winged veldt]

ye, i clearly have some issue

i gotta fix the sampling on emissives as well

[bronze sedge]

I wish I understood the words used here

I will some day

[spice frost]

Woah these are the same sample counts?

[spice frost]

you don't happen to have gmon on?

I get just a few fireflies... I wonder if I also have a problem

[quick zealot]

No GMoN

[quick zealot]

how many SPPs?

[spice frost]

10k

but the fireflies increase over time

[quick zealot]

that's 10k

[quick zealot]

maybe some slight numerical instabilities somewhere

[spice frost]

dang

[quick zealot]

that pop from time to time

[spice frost]

also, do you use agx?

[quick zealot]

no I use some scuffed reinhart stuff

exposure + gamma

don't even remember what impl that is exactly

[spice frost]

ahh okay, they looked similar enough so the fact it's not identical got me worried haha

[copper halo]

https://tenor.com/bUd5v.gif

[quick zealot]

I'll get around GT7 when I decide it's worth the time : D

[copper halo]

now

now is the time

[quick zealot]

😳

[copper halo]

i need gt7 impled when i'm back next week

[quick zealot]

oh boi

and a bvh-built light tree with that I guess

[spice frost]

does gt7 actually refer to gran turismo 7?

[copper halo]

sure if you can do both in a week

yes

it's the tonemap they made specifically for gt7

[quick zealot]

ain't happening x)

[copper halo]

^

[quick zealot]

[spice frost]

color is such a rabbit hole

stay away from it

[quick zealot]

but

https://tenor.com/view/cat-death-penalty-gif-25754511

surely I can just poke my hand in the rabbit hole, grab GT7 and run away right?

[spice frost]

https://tenor.com/view/mistwol-cat-his-greed-will-consume-him-eating-silly-gif-14271929065765266047

[keen bear]

god that gif has horrible quality

https://tenor.com/view/cat-box-cute-death-penalty-gif-19440000

have an actually good one

[keen bear]

this is me

[quick zealot]

actual masterpiece that gif

[old hollow]

Hey mate @winged veldt I will have an m4 Mac mini tomorrow (however can’t unbox or use it until next week, long story) but it’ll get delivered to my auntie who lives near us along with a cheap wireless keyboard and mouse with no need for a usb dongle

So I will be testing your project with hw rt support finally

[old hollow]

Very excited to be on the m4 hype train hah

[winged veldt]

ooh very cool

m4 is neat yeah

m4 pro is like 30% berf uplift over my m3 pro lol (probably good bit of that is it being 8p/4e vs 5p/6e)

lmk if you find the firefly causing bugs

i haven't had time to look into it more so far

[old hollow]

Yeah, and well I’m still surprised the m4 isn’t faster than M1 Pro in my MacBook (except it has hw rt so your app will run better hehe) also it supports 64 bit atomic so I can finally run unreal engine 5 with nanite on mac :3

M4 pro however is

Do you like Mac OS? I love it tbh, it’s like the perfect middle between windows which is clunky and Linux which is somewhat obscure and a bit tricky to know where to find stuff

[winged veldt]

yeah macos is like

most of the good stuff from linux (you have a posix aka real os) but with a gui that works

windows is just ugh

every version gets worse and since 10 it's actively hostile to the user

my only real complaint with mac is developing for it can be kind of a pain at times

apple really wants you to use xcode

but xcode is ass

[old hollow]

Yeah Xcode sucks

But yeah like, windows 11 update bricked my ssd

[bronze sedge]

I'm sorry that happened, that's crazy

really?

like windows 10 to 11 bricked your SSD?

do you know why?

[timid frigate]

what is better for you on mac for dev?
I worked on mac years ago but now I use win 10 and dont want to use 11 or more. I use vscodium and console compile

[winged veldt]

i just use neovim and clang

metal compiler through cli for shaders

[timid frigate]

cli is good thing

I use visual studio c++ compiler with cli

[copper halo]

same

clangd is great

slow as shit to index though

[winged veldt]

no issues on platinum guess it's not that big of a codebase

ime clangd is way faster with neovim as a client than clion

for some reason with clion it would sometimes eat 30gb of ram and shit itself

never did that with nvim

[old hollow]

No a specific windows 11 update, Microsoft deny it but plenty of people have said about it so I guess that’s what happened to mine

Thankfully not my windows install ssd

[winged veldt]

thankfully? i'd rather bindows delete itself than my data

[old hollow]

Haha yeah I mean the data hasn’t been corrupted I don’t think cos after the drive was inaccessible it came back fully working showing in crystal disk as fine

[old hollow]

The Mac mini has landed at my aunties

Btw the “long story” short, we have rats in our home so we are living for a week in a hotel while they’re poisoned

Landlord paid for us to stay at the hotel nearby for a week heh

[copper halo]

yeah I'm talking about using it on UE

for my dingus shit I use crab

r-a is ass

[winged veldt]

nah it's good ime

[copper halo]

for meme projects sure

[winged veldt]

but it lacks the cargo check diagnostics

[copper halo]

anything serious it just shits itself

[winged veldt]

and cargo check is quite slow

[copper halo]

I've had the thing import something with a longer name than use the thing already in scope

[bronze sedge]

I don't have any clang issues

or clangd issues

[copper halo]

yeah it's fast for meme projects

[bronze sedge]

10k loc only though

whats it bottlenecked on?

[copper halo]

indexing ofc

[bronze sedge]

gopls just consumes GB of memory

[winged veldt]

go pls

[bronze sedge]

ok

[winged veldt]

no not like that

[winged veldt]

tbf there's a big gap between meme projects and ue

[copper halo]

well yeah true

[spice frost]

I've finally figured out where my fireflies came from... added a check to terminate paths that generate directions sampled from brdfs that were not in the same hemisphere as the geometric normal and they're gone

not sure what the best way to deal with that is, but I'll leave it like this for now since it did not visibly darken the image

[winged veldt]

hm how did you get those paths?

that feels like it shouldn't be happening but ima check later

[spice frost]

i think it's both the microfacet normal and the shading normal deviating from the geometric normal that can cause new rays to skip the surface and go inside/past the object

inside the brdf sample function i only check that wi and wo are in the same hemisphere

which doesn't do anything torwards that problem i think

[keen bear]

@winged veldt what's the status on your firefly problem

[winged veldt]

no updates, haven't had any time to work on it

wanted to see if i can try some stuff later tonight actually

i'll be posting any updates here for sure

[keen bear]

in the scenes that show up with firefiles, do you think you have any materials that have significant contributions from bsdfs involving microfacets

I noticed that pbrt does something in its microfacet distribution thing to avoid numerical problems when roughness is very anisotropic

https://github.com/mmp/pbrt-v4/blob/f140d7cba5dc7b941f9346d6b7d1476a05c28c37/src/pbrt/util/scattering.h#L117

[winged veldt]

hm, i haven't added aniso roughness to platinum yet actually

i do technically use anisotropic ggx but the x/y roughness is always the same

[keen bear]

i c

then nvm

[winged veldt]

it should be really shrimple to implement, but i wanted a cleaner way to do anisotropic rotation than what i used in yart cause that was stinky

but also i have at least some firefly issues outside the bsdf cause the first test i did was to make everything lambertian and some fireflies show up with mis still

here

so i'm gonna look into fixing that first then enable bits of the bsdf one by one and make sure they're firefly free

[old hollow]

Btw @winged veldt you tried Liquid Glass yet in your project

[winged veldt]

hm yes imgui liquid glass

[old hollow]

Ohhhh sorry I forget you’re using imgui not native uikit

Aw well you could try make Liquid Ass with imgui user callbacks ha

I wanna do callbacks to do 3D imgui

Btw I’ve decided to use pure metal api to make my v2 engine

If my OpenGL engine was a car it’d be a lada from 1976 Soviet Union

Metal engine will be a Bugatti

Gee I hope you do fix your fireflies

It’s an annoying issue

Similarly I have NANs from my ray-cone/sphere based volumetric raymarching but not on metal translation of gl

Only on windows heh

It also happens if you have a light exactly attached to the camera origin heh

Again, not on Mac OS

Guess metal has more protection from nasty black squares

[winged veldt]

yee i just have little to no time to work on it but i'm trying to make some progress even if super slow

[winged veldt]

what language will you be working with?

if you're doing c/cipipi i'd recomment avoiding metal_cpp and just writing the metal bits in objc, it's kinda clunky

metal being an objc api is kinda ass for interop tbh

[old hollow]

C#

Using Maui

Microsoft metal bindings lol lol

[winged veldt]

oh didn't know that was a thing, interesting

[keen bear]

@winged veldt I wonder have you considered doing some unification of things that use microfacet models (metals, the smooth part of the dielectrics, coat, etc)? I found the way cycles organizes its code is kinda interesting, it has a "microfacet bsdf" that's used for everything involving microfacet models, and it's just fresnel models that are different

like metals use something called f82 tint (if it's part of principled bsdf, standalone metallic bsdf also has an option to use "physical conductor" in which things are specified differently), dielectrics use something else, etc

so things end up imo neatly separated

[winged veldt]

you mean like, evaluating ggx once and reusing it a bunch of times? i haven't done that yet, it's an optimization

i do intend to at some point, though

[keen bear]

no

just moving code around

basically factoring out differences between conductors and dielectrics into types with a smol interface (fresnel models)

[winged veldt]

oh not sure what you mean then? i already have a ggx class for the microfacet distribution

[winged veldt]

ah right, i see

[keen bear]

and then plugging that into microfacet bsdf which glues that + microfact distribution

[winged veldt]

that could be interesting yea

i should look at how cycles does it

but also last time i went looking at cycles code it took me days to make any progress, it's not exactly easy to parse

for now i just want to fix the fireflies

[copper halo]

🔥 🪰

[spice frost]

fix the fireflies!!! look what you could be doing rn

[copper halo]

man so many cool things i could be girafficsing but no time

mfw i have work, work, school, school societies, applications for next year, my renderer, bevy, and basic necessities to survive on my list of things to do

and do none of them

[keen bear]

yeah. I got burned on material stuff so I'm just working, not even touching grass (excepting weekends)

well still progressing but very slowly

[winged veldt]

very cool

volumes is my next thing to implement so ye

[keen bear]

is that actually a volume

[winged veldt]

looks like it ye, shrimple beer's law with rough glass surface

yart does actually support that in a super hacky way

didn't add it to platinum cause i want to do better

[spice frost]

ye it's just beer's law

i'm still reading through the volume stuff in pbrt

[winged veldt]

some testing: removed direct light hit contributions entirely, the image is wrong (mis is still assuming they're being counted) but the fireflies are gone

@spice frost i'm also rejecting nee samples where the sampled direction is in the opposite hemisphere to the geometric normal

so basically firefly issues are somewhere in the direct hit part

[spice frost]

ah i was also doing that too... my problem was with the brdf samples

but you do not seem to have that problem

kinda weird now that i think about it

those planes shouldn't have interpolated normals that are different from geometric normals and the lights are a simple diffuse

maybe the microfacets can generate reflections that go outside of the hemisphere with high enough roughness but dunno

[winged veldt]

hrm, my problem now is with the brdf samples when the ray hits a light directly

[spice frost]

Could it be the brdf pdf giving you the wrong mis weight?

I'm going to bed now but I can look through your code tomorrow if you want an extra pair of eyes

[winged veldt]

sure, that always helps

nano's comments have been great

i think i'm onto something here, i clamped the cos / pdf term when calculating attenuation to 1 and the fireflies are gone (still some slightly brighter pixels but nowhere near as bad)

overall brightness now matches naive (right) perfectly as well, was a little too bright

so i think some of my pdfs were too close to zero and the attenuation factor was blowing up (and going > 1)

[winged veldt]

alright, clamp cos/pdf and also clamping rr termination probability to 99.9% seems to have done the trick

this one is way oversampled (10k) but it shows any bright pixels that are left do go away as the image converges, so they're officially not fireflies™

oh it also destroys indirect lighting

at least i now know it's small pdfs causing the fireflies

[winged veldt]

hm, rejecting bsdf samples at extreme grazing angles works well for the veach scene but does nothing for the dragons scene

[normal solar]

if you use surface form (judging by that emissive polygon in the cornell box) check denominator

[spice frost]

also remember having a two sided light being that close to the ceiling while still having a gap increases variance in my experience

[keen bear]

btw you need to apply coat to emission

which you don't rn

afaiu

[winged veldt]

oh? wdym

yea my emission lobe is kinda jank

i usually only ever have emission on diffuse materials with roughness 1 so i didn't bother yet

[keen bear]

I done cancer

struct composedBxDF {
    // TODO: would be nice if we could put it behind a pointer tbh
    Registers regs;

    uint8_t bxdfids[5]; // TODO: should be a global ptr
    uint8_t regoffs[5]; // same as above, probs interleave the two
    uint8_t n;
    // uint8_t contribregs;

    BSDFFlags Flags() {
        uint32_t result = 0;
        for (uint8_t i = 0; i < n; i++) {
            int regoff = regoffs[i];

            switch ((bxdfid)bxdfids[i]) {
            case bxdfid::diffuseBsdf:
                result |= (uint32_t)diffuseFromRegisters(regs, regoff).Flags();
                break;

            default:
                /* abort(); */
                break;
            }
        }
        return (BSDFFlags)result;
    }

    BSDFEvaluatedResult Eval(float3 wo, float3 wi) {
        BSDFEvaluatedResult result = {};
        for (uint8_t i = 0; i < n; i++) {
            int regoff = regoffs[i];
            float contrib = 1.0 / n;

            BSDFEvaluatedResult tmp = {};
            switch ((bxdfid)bxdfids[i]) {
            case bxdfid::diffuseBsdf:
                tmp = diffuseFromRegisters(regs, regoff).Eval(wo, wi);
                break;

            default:
                /* abort(); */
                break;
            }

            result = result.Add(tmp.Scale(contrib));
        }
        return result;
    }

    BSDFSample Sample(float3 wo, float uc, float2 u) {
        uint8_t i = min(uc * (float)n, n-1); // TODO: this is horrid, actually

        int regoff = regoffs[i];

        switch ((bxdfid)bxdfids[i]) {
        case bxdfid::diffuseBsdf:
            return diffuseFromRegisters(regs, regoff).Sample(wo, uc, u);

        default:
            /* abort(); */
            return {};
        }
    }
};

I can now almost have weird bxdf compositions fed by the interpreter.

you need this too

contrib should be read from a register and sampling should consider contribs

[winged veldt]

wot's this

[keen bear]

atm I implement programmable materials with interpreter only

Registers is interpreter state beside the instruction pointer

[old hollow]

hey bluescreen I'm ready to try it, what were the steps to get it to run again?

[keen bear]

so the program runs and when it termiantes, registers contain values that are inputs to weighted bxdfs

so like e.g.

this funny chungus would make it so the interpreter fills in the values that are inputs to the "shader" nodes here

and the contribution of each "shader" node

which is 0.5 for both in this case

and then the composedBxDF thing would implement bxdf interface in terms of the values in these registers

it's remotely comparable to how cycles' svm (its material interpreter) outputs a sequence of closures, except without unnecessary copies

and I guess also somewhat less elegant idk

tbc I'll also cook up a strategy for using api shaders for materials but interpreter is a much less complex way to approach it imo and also you don't get the issue of waiting for api shaders to compile

still not entirely sure where to fit edfs but I probs just need to thonk a lil

[winged veldt]

oh hey

so uhh i redid the build a bit cause i'm no longer using clion

so assuming you have the xcode cli tools for the metal compiler and stuff you should be able to just clone the repo and make

oh might have to make configure first

idk what i called the cmake config target

[winged veldt]

ah right

is that at runtime? or does it compile a specialized shader

[keen bear]

it's at runtime

[keen bear]

see

[winged veldt]

oh

how's berf

[keen bear]

bad

next question

I mean it's usable

on steam deck

but it's bad

interpreter takes up a significant chunk of time of an an rt dispatch

you can literally feel each interpreter instruction

but I mean

[winged veldt]

oh right you're doing realtime too

[keen bear]

blender's interpreter is even worse

so w/e

[winged veldt]

[spice frost]

[old hollow]

omg ios provisioning is a nightmare :c

rider keeps saying the provisioning profile isn't valid :C

[winged veldt]

shrimple, don't develop for ios

[old hollow]

:’(

You know if it wasn’t for the Mac keychain having a billion places to drop your cert files it wouldn’t be hard to do

[winged veldt]

oh ffs

macos 26 broke metal-cpp

[spice frost]

how badly?

[winged veldt]

i think it's a cmake issue actually

doesn't find system headers

[winged veldt]

got it building again, just a cmake cache issue + some include paths changed for mysterious reasons

macos tahoe has enbiggified my traffic lights, now they're no longer centered in the menu bar

[copper halo]

what's the goofy ahh weird outline there

[winged veldt]

in the corner? just my terminal window behind platinum

some apps still using the old style sadge

fixed 😌

now back to fixing ze fireflies

[copper halo]

glhf

[winged veldt]

@spice frost think you might be right, with the light further from the ceiling i'm not getting any fireflies

overall variance is higher tho

probably because of the additional indirect paths

with more samples it cleans up nicely and no fireflies

that's good, means it's a scene issue

honestly i think ima just make all lights one sided for the time being, i literally don't have a single scene where i want two sided lights

[quick zealot]

why fireflies on the veach scene though?

[spice frost]

try this one

[winged veldt]

they're gone

[spice frost]

but yeah those images of the dragon scene after your fix just look like regular variance and not fireflies

[winged veldt]

given the fix was rejecting extreme grazing samples, probably numerical instability with a very low pdf

or wi.z near zero

[quick zealot]

so rejecting when cos_theta of the sampled direction is near zero?

[winged veldt]

i also clamped rr termination probability to 99.9%

to avoid dividing throughput by a tiny number and blowing it up

[winged veldt]

yea

like 1e-4

[quick zealot]

but you didn't have the fireflies with a brute force path tracer no? (which also included the grazing angle samples)

[winged veldt]

no, i think it caused issues in the calculation of mis weights, so they blew up

[quick zealot]

hmmm

1.0e-4f I think should be really fine for numerical stability though no 🤔

[winged veldt]

yea i'm being conservative

makes zero difference in the results

since those samples have near zero contribution anyway

1024spp with one sided light, looking clean

[spice frost]

volumes are next?

[winged veldt]

nah this is diffuse only, now i have to fix the bsdf fireflies

then volumes

but hey, progress

[spice frost]

small steps

[winged veldt]

still having major issues with the full bsdf

[spice frost]

can you check in the veach scene what it looks like for varying roughness conductors? i think my fireflies came only from high roughness causing rays to skip the surfaces

[winged veldt]

still have some fireflies

none with diffuse, though

i'll have to test bsdf lobes one at a time, probably gonna start with conductors as that's the simplest one and will let me make sure the ggx stuff itself doesn't have any issues

[quick zealot]

I'm a bit suspicious of that tbh
Is it not all a MIS issue somehow?

There are still some mild fireflies even with the rejecting grazing samples
You'd have to reject even harder at like 1.0e-3 if you want them gone and this doesn't sound like numerical instabilities anymore imo

[winged veldt]

those aren't fireflies, they do go away with more samples

but yea there's still issues

don't think it's mis tho

i've checked that code about a billion times and just do not see what i could be doing different

it's for sure not the nee path cause they go away entirely if i disable light hit contribution

might be a mis issue there, but the weight calculation is correct and there's nothing that could cause it to blow up

[quick zealot]

so it's BSDF sampling or BSDF MIS then?

[winged veldt]

yea but honestly idk anymore

this is the light hit path:

if (sample.flags & bsdf::Sample_Emitted) {
  if (bounce == 0 || lastSample.flags & bsdf::Sample_Specular) {
    L += attenuation * sample.Le;
  } else {
    // Calculate light PDF, BSDF weight and do MIS.
    // Sampling pdf is 1 / area, light sample pdf is power / totalPower
    // Because power = Le * pi * area, the areas cancel each other out
    // and we can simplify
    float lightPdf = (dot(sample.Le, float3(0, 1, 0)) * M_PI_F /
                            args.constants.totalLightPower) *
                            length_squared(lastHit.pos - hit.pos) /
                            abs(dot(ray.direction, hit.geometricNormal));
    float bsdfWeight = lastSample.pdf / (lastSample.pdf + lightPdf);

    L += attenuation * bsdfWeight * sample.Le;
  }
}

the bsdf weight can't go too high unless both pdfs are near zero

and sample.Le is literally just material emission

[quick zealot]

what's this?

dot(sample.Le, float3(0, 1, 0)

[winged veldt]

luma from temu

that's not the issue

it matches what's used for calculating light power elsewhere

[quick zealot]

and the * M_PI_F?

[spice frost]

i also did end up looking at your mis code and didn't find anything that stood out

[winged veldt]

light power is le * pi * area

i honestly don't recall why the pi is there

iirc some change of variable but as far as the weight is concerned

it's a constant on both sides, i could remove it from the power calculation entirely and the result would be the same

only thing that could make the light pdf near zero is the distance being very close (distance squared term)

but that would have to combine with a near-zero bsdf pdf, otherwise you just get a bsdf weight close to 1

actually save for some weird numerical instability the bsdf weight literally cannot be >1

so yeah it's gotta be the attenuation

the weird thing is that bit is the exact same as the naive integrator which doesn't have firefly issues

or maybe it does, the veach scene is so noisy i can't really tell for sure

i might just leave like a 200k sample render and go make lunch

[quick zealot]

I'm kind of getting these guys too with MIS so maybe it's instabilities yeah
but I don't have any kind of clamping though

[winged veldt]

at 20k i think it's clean enough to say there aren't any fireflies

[quick zealot]

that small pink light is oof without NEE

looks like debugging-pink-NaNs everywhere

[winged veldt]

they are pretty obvious with mis+nee

though, they do seem to get converged away eventually...

interesting is that they're mosty green

[quick zealot]

how many bounces?

[winged veldt]

anyway that's pretty much all i have time for atm i should like do stuff

[winged veldt]

50 max

[quick zealot]

and with 0 bounce? which disables bsdf mis right

[winged veldt]

should be way more than the scene needs

[winged veldt]

yeah, but i know disabling that works

[copper halo]

what's a lunch

[timid frigate]

where is cornell_dragons.glb or cornell_dragons1.glb?

[winged veldt]

i've been trying out zed as an editor, it's very nice

might pry me away from neovim, even

(this was largely triggered by nvim inexplicably spiking my cpu to 100% when i scroll after an update, and me not having time to debug my editor)

[hazy pelican]

use a nvim distro ?

[winged veldt]

nih'd config using lazy.nvim is what i've been using

[winged veldt]

anyway zed vim bindings are top notch

and ironically being a gui editor it can actually make better use of the keyboard

[hazy pelican]

I'll have to try it out

[winged veldt]

looks like the scrolling issue isn't even neovim, just a massive macos 26 bug

zed is still a very cool editor tho, i like it so far

apparently there's a funi macos tahoe bug where some new feature causes scrolling performance to degrade over time and eat up all the cpu

the workaround is to restart the app

which is just great for my average uptime of just over two months

[winged veldt]

@dense jacinth you've been on the tahoe beta for a while right? did you notice any impact on battery life?

i'm pretty sure it's significantly worse than it was on sequoia

[dense jacinth]

they look the same on my Mac

or am I blind

we have the same macbook so that should work

[dense jacinth]

uh I had tahoe beta on since the start..... i don't think there was any noticeable degration

fuck seeing this I want to do graphics again

but now I need to do PCBs and build a fucking car 💀

[winged veldt]

left is more saturated for me, it's rather subtle tho

maybe your display is sest to srgb?

[winged veldt]

huh

i swear my battery is draining quicker since i updated

[dense jacinth]

doubt it

[winged veldt]

i need to do uni and random jobs

still trying to fix the damn fireflies

[dense jacinth]

this is uni but requires 100% of my time

[winged veldt]

uni taking up 100% of my time rn tbh

[dense jacinth]

I only did uni at the side so I barely wrote the exams I should've

[winged veldt]

but that's me being dumb and doing more than a normal semester's worth

[dense jacinth]

did 2 out of 5 I think

[winged veldt]

got a job in automotive?

i hear it's

[dense jacinth]

it's not even a fucking job

[winged veldt]

what was the horrible auto embedded thing everyone uses

intern?

hope they're paying you at least

[dense jacinth]

nope and nope

[winged veldt]

[dense jacinth]

it's just a club type thing

https://www.ecurie-aix.de

[winged veldt]

ah so it's self inflicted

[dense jacinth]

yes

[winged veldt]

rwth aachen, nice

that was on my shortlist for exchange studies

[dense jacinth]

oh nice

[winged veldt]

will be going to lund tho

[dense jacinth]

well it is a really good uni

[winged veldt]

that looks fun tho

[dense jacinth]

well yes but it's also an insane amount of work

basically doing electronics and software for them for that

and i have the slight pressure on me that the last car was the fastest car ever built in this category

[winged veldt]

ok i'm pretty sure macos tahoe is actually destroying battery life

like what the fuck is this

seriously considering going back to sequoia even if it involves wiping the thing

[quick zealot]

[dense jacinth]

can we please return to basic iOS 7 UI

and destroy every piece of React that exists

and only have native applications

our energy usage will thank us 🙏

[winged veldt]

no more liquid ass for me, i rolled back to sequoia

[keen bear]

eat the liquid ass.

[spice frost]

not me actually googling liquid ass

[bronze sedge]

How do you rollback a MacOS OS update?

[winged veldt]

wipe the mac

oh well, a fresh os install every year or two does help keep things clean

(if you have a time machine backup pre-update you can just use that and it's a one click rollback afaik, but i didn't have one)

[bronze sedge]

Do you still get security updates?

Idk about this rollback thing

I think this also means you aren’t getting driver updates? Like new Metal features?

[winged veldt]

yeah

[winged veldt]

apple supports two os versions behind latest

so sonoma and sequoia are still supported

support meaning you get security updates but also new safari, xcode tools etc, so yea i should have the new metal stuff as well for another two years

ie until macos 28 releases

i can only hope they'll have fixed the issues with tahoe by then

honestly battery issues aside (battery life seems back to normal now, drained ~30% in just under three hours with an external display connected) sequoia ui looks so much better, the redesign is not good

i think the whole liquid glass thing was clearly designed for ios first cause it actually looks and works great on my ipad

tbf the fancy glass effects are not the issue, it's the layout changes to macos that i don't like (mostly making random things bigger for no reason)

[bronze sedge]

I couldn't tell any difference

maybe I don't have the update yet

[winged veldt]

it doesn't autoupdate, never has for major os versions

you can check by going to system update in settings

or about this mac

if you haven't updated don't, tahoe needs some time in the oven

or, like, make a time machine backup if you do

[bronze sedge]

nah that's my work laptop, I'm not going to update and hose my ability to get work done, thanks for taking one for the team

I'm gonna wait a few months or something lol

[winged veldt]

tbf it's not broken, everything still works mostly the same

[bronze sedge]

I need docker to work

[winged veldt]

only real issue i had was battery life being significantly worse

[winged veldt]

ye, docker works fine

[bronze sedge]

I'm going to wait a bit and maybe a coworker or two update and then I find out if it's safe

let them be the first ones

I'm already in enough trouble

[winged veldt]

but there's also no reason to update because it doesn't even introduce any new features

[bronze sedge]

well, you get a windows vista looking UI

[winged veldt]

other than the liquid ass which atm is a downgrade imo

the icons are nice tho

[bronze sedge]

idk it all looks the same to me

[winged veldt]

back to normal, tahoe is truly ass

[copper halo]

liquid ass very optimized

[old hollow]

@winged veldt maaaate wtf is happening :C

my specular is static

the world pos is correct, the camera pos changes when I move the camera

but this happens

holy shit this is annoying

[winged veldt]

not sure what i'm looking at here?