GPU Zen 3: Virtual Shadow Maps

[mystic lark]

Any updates btw? I'm half expecting them to say "nevermind actually, we don't like the article and scratch it lmao

[sweet nimbus]

None yet

[sweet nimbus]

how is it possible to refer to RAGE having virtual texturing when I cannot even find a first-party reference to it

[vestal snow]

true I remember there used to be a full presentation that was easy to find

idk where it went though, been googling

found this though: https://www.youtube.com/watch?v=M6y4glzIwP8

[sweet nimbus]

Ah yes, from John Rendering Pipeline himself

[vestal snow]

oh you know what there might be a quick YT presentation from actual john rendering pipeline himself

not ideal but maybe the original stuff got deleted for some reason 😦

[sweet nimbus]

Maybe you can find a link to the og stuff and use wayback machine

[vestal snow]

https://web.archive.org/web/20091007031619/http://s09.idav.ucdavis.edu/talks/05-JP_id_Tech_5_Challenges.pdf

you were right it worked

[sweet nimbus]

DX9 == nirvana for conventional hardware graphics

[hazy steppe]

shame this didnt lead to true opengl multithreading capabilities

[sweet nimbus]

we need to workshop this

[sweet nimbus]

I added the mention of the virtual texturing stuff

but we still need to reference that 2015 gpu-driven presentation that has 2 slides of VSM

[sweet nimbus]

ok I added the 2015 stuff too (line 98 if you want to look)

that should wrap up all the loose ends

[hazy steppe]

are you passively waiting for wolfgang to take a look or are you going to boop him when you are done?

[sweet nimbus]

well, it doesn't look like the other articles have received editing by the book people

idk

[hazy steppe]

: (

[sweet nimbus]

one of the articles is incredibly long- like 70 pages by itself

[hazy steppe]

oi

[sweet nimbus]

there is a lot of work for these people

our article is a meager 20 pages : D

[hazy steppe]

wolfgang (perhaps his or some of his book people too) is on the forge discord too, just in case

[sweet nimbus]

oh right, @acoustic bobcat @prime ice do you want to be acknowledged for your help in reviewing and editing?

[sweet nimbus]

I'll just email our contact

[hazy steppe]

i captured a glimpse into the future

vsm article release party https://www.youtube.com/watch?v=KBCvDJcrLgU

demongod @ 0:08, patrick @ 0:18, froyok @ 0:30, hatrickk @ 0:32 , devsh @ 0:34, saky @ 0:51, deccer @ 0:57, DR @ 1:02, jb @ 1:11, JStephano @ 1:30 and jaker @ 1:40, someonesomewhere @ 2:04 sean @ 2:06 😄

[sweet nimbus]

hmm potrick has less hair and bigger muscles

[humble shadow]

Can confirm this is accurate, those are my dance moves

[hazy steppe]

potrick @ 1:32 then

shit i forgor lvstri

@ 1:38 🙂

[sweet nimbus]

the furtive lvstri, so easily forgotten

[wooden jolt]

my name will go down in history when I ship nanite

(expected release date: 31/12/9999)

[sweet nimbus]

hehe I do wonder what some of y'all look like IRL

[hazy steppe]

ah good, i thought 32/13/9999

[prime ice]

really good music

[sweet nimbus]

it's kinda funny that we have this medium of communication where we can become friends with people without even knowing what they look loike

[mystic lark]

but it's also written by like 10 people

[humble shadow]

The inner beauty is what is charming all of us together

[sweet nimbus]

that and the shared neuron probably

[wooden jolt]

was about to say

we share one neuron

we are one

[hazy steppe]

Doofenschmirtzes neuron apparently 😄

[humble shadow]

Is this one split across of us, or one that pass by one by one ?

[sweet nimbus]

it's guarded by a mutex that we have to acquire before use

[hazy steppe]

like an umbelical cord

[hazy steppe]

std::conditional_variable too?

[sweet nimbus]

pavlov's classic: std::conditioned_variable

[hazy steppe]

xD

[acoustic bobcat]

Sure I suppose

[acoustic bobcat]

Hmm

Don't put my handle in though, I can punch my name in when you need I guess

[sweet nimbus]

I thought I sent the email to our editor but it turns out I just left the draft sitting there

now it's sent

[sweet nimbus]

that was quick. he's asking for the PDF to review as he doesn't have access to overleaf atm

[sweet nimbus]

(I sent it btw)

[mystic lark]

Very sus communication

[tepid geyser]

How does VSM fare for memory consumption compared to other techniques

[sweet nimbus]

Best

[wooden jolt]

fixed cost, completely configurable

[sweet nimbus]

And basically no wasted shadow texels

[mystic lark]

Yeah a single 8k texture can easily fit 16 4k cascades, probably could also fit 16 8k cascades

[sweet nimbus]

It's about the pages that are resident really. You could have a thousand 16k² cascades

[wooden jolt]

mmm

pixel sized pages

[hazy steppe]

whats the status?

[sweet nimbus]

stasis

[sweet nimbus]

The editor got back with comments

I'll forward the email to the homies

[mystic lark]

Oh wow I half wasn't expecting a response haha

[sweet nimbus]

Email sent

[mystic lark]

Hmmm interesting, the later comments I don't really understand (in the results/appendix section)

[sweet nimbus]

I didn't look yet because I need to sleep/I'm on mobile

[mystic lark]

Not many comments

Like 15-20 at most

[vestal snow]

Maybe we just need to recap details? Not sure

Like for the comment about “seeing black pages” we could just mention that if they were in view, that’s a page fault, so data would have been generated

[mystic lark]

Same with the sun, we did say that the pages get invalidated on sun movement no?

Maybe we should explain better

[acoustic bobcat]

Is this somewhere I can see it or is it all in the master document

I might be able to take another editing pass over it this week

[mystic lark]

Jaker sent an email

[sweet nimbus]

I forwarded it to you too @acoustic bobcat

[sweet nimbus]

Should we put our affiliation as graphics-programming.org since Peter asked

[mystic lark]

I can ask my uni

And have that

[sweet nimbus]

I don't think we need to have an affiliation though

Since we're really not

Peter probably just thought we have one as we're a group of four

Little does he know the power of VSMs in making friends

[hazy steppe]

"deccer made us do it"

[wooden jolt]

the real VSM was the friends we made along the way

[hazy steppe]

shame i was not able to contribute anything

[sweet nimbus]

you contributed encouragement

[dark crater]

yall mfs are making me add vsm on my block game

is there a draft somewhere? or whats the most recent article

[sweet nimbus]

I can add you to the overleaf if you want

just gib your email

[dark crater]

neat

im not thinking of doing the whole thing (ill be re-rendering every frame), but i think just the clipmap version or similar would work well for what im doing

[sweet nimbus]

uncached increases render times significantly in our benchmarks

but it's still within a couple ms for 16 4k shadow maps, so it could be viable

[dark crater]

im mostly interesting on the culling part, i wanted to see how you doing it

so essentially doing cascaded shadows, but building a structure to cull it on tiles accurately

[sweet nimbus]

the culling is basically adapting hzb to work with a page table instead of a depth buffer

[dark crater]

yeh just saw, so kinda what i was thinking. Gonna try to implement that part

[sweet nimbus]

it's pretty easy, and is definitely the biggest optimization

if you have fine-grained culling then it should help a lot

[dark crater]

need to benchmark how does the atomic-mins work to build that depth buffer for shadow-space hbz

[sweet nimbus]

that's the most expensive part and is the reason why you need good culling to prevent excess drawing

[dark crater]

im doing voxel rendering and each chunk is pretty smol + cheap to draw, and the cull is indirect so all good

[wooden jolt]

depth HZB is trickier btw

but it makes VSM basically as fast as CSM to render

[dark crater]

i just want a better-culled csm

by building that tile structure i get something to accurately cull shadowcasters, or thats my thinking at least

[wooden jolt]

ye the HPB is the base culling that will help you a lot already

if you want more 🅱️erf you try doing HZB

[dark crater]

how do you cull your shadowcasters?

do you also do frustrum/etc?

[wooden jolt]

yes

[dark crater]

or just project box into shadow space and check bounds

how do you do frustrum on the shadowcasters?

are you going from the camera frustrum with some tricks, or just frustrum of the light

[wooden jolt]

the same way for the main view

just check it against the box plane's

[dark crater]

ah but thats of little use 🤔 you are drawing tons of shadows behind the camera that dont intersect and arent visible

atm i was using some of the camera frustrum planes to also cull shadowcasetrs

[sweet nimbus]

objects that shadow areas behind the camera are culled by hpb

[dark crater]

so stuff behind camera doesnt show

ah, true

[wooden jolt]

it's a side effect of the depth analysis

you only render what the player sees

[sweet nimbus]

you can't really screw with the light matrices too much since they need to be stable for caching

but since you don't care about caching then maybe you can do other memes

[dark crater]

i dont care about caching because ill have a day-night cycle and light is going to be moving

and cant cache stuff easily if you can destroy blocks/etc

i think trying to bother much with the caching in my case would just make everything much harder

[sweet nimbus]

caching is on a per-page basis

if you break a block then you'll affect a few pages at most

and day-night cycle can be coped by quantizing the sun movement

[dark crater]

yes but now i need to keep track of all that, plus the dynamic enemies and so on

[sweet nimbus]

indeed

[dark crater]

and im most likely going to have animated shadows on trees

[wooden jolt]

it do be a pain but caching helps a lot

[dark crater]

which means its likely everything is going to be invalidated anyway

thing is that my thing renders uberfast due to the whole "just cubes" part

the full 2k shadowmap rendering everything on a scale of 1 block = 6 pixels or so takes 0.5 ms to render (before even doing camera frustrum cull)

so im expecting that i can have 3 2k cascades below the 1 ms target once i have decent culling for them

[sweet nimbus]

why use vsm if you only have 3 2k cascades

[dark crater]

i mainly only want the page table part for accurate culling of shadows

[sweet nimbus]

aight

hmm you won't even need indirection actually

[dark crater]

no

[sweet nimbus]

you just need a way to mark active areas (what would be pages) and then make a hierarchical structure from that and cull against it

[wooden jolt]

true

there's no V in this VSM

[sweet nimbus]

it's just SM with better culling

the V allows you to control residency and have an assload of cascades with minimal memory/rendering cost

[dark crater]

which i might still do later

depends on brainworm

[sweet nimbus]

the V is by far the most complicated part

[sweet nimbus]

I actually wonder if games have implemented a culling system for regular SM like this before. it seems like an obvious improvement for most titles

I think it's worth noting in the article that the hpb can be used this way

[dark crater]

it requires indirect funtimes

tho you could also use it to prime the depth buffer for shadow

and afaik yes its a common thing, it seems that AC unity did it

are they doing that with the atomicmin thing?

[sweet nimbus]

reprojection 💀

I don't see how this is describing the page-based culling

this is regular hzb

man reprojection is kinda an overloaded term

I wish I had the speaker notes or the whole video for this

[mystic lark]

I was wondering btw, what if I bias the light matrices center forward in the viewing direction

So they no longer are centered on the main camera

And I get more resolution

[prime ice]

you need hiz or caching

[dark crater]

thats like.. normal?

[prime ice]

even in sponza we get ass perf otherwise

[mystic lark]

Not for VSMs

[sweet nimbus]

as long as you snap it to the grid it should be ok

[dark crater]

@mystic lark ah true

[mystic lark]

I think it should better utilize the resolution

Gib better shadows

[sweet nimbus]

perhap

[wooden jolt]

better than RT

[mystic lark]

Naaah you are right, probably smarter to rt

Than more VSM bikeshed

[wooden jolt]

no

VSM bikeshed

RT is cringe

it's equivalent to admitting defeat

[mystic lark]

Ye I meeean

[prime ice]

vblanco is in for a ride

next few months gone

[mystic lark]

Caching is the most difficult part, without it it shouldn't be such a hurdle

Just more granular caching

Building the actual hpb is also simple

[sweet nimbus]

I think the V is the hardest part

Caching is easy if you don't care about invalidation

[mystic lark]

Ig it all seems simple now that I know how to do it

[hazy steppe]

now i know what the v in vblanco stands for

[buoyant lily]

Rasterizing instead of raytracing, with voxels

[dark crater]

@buoyant lily i explicitly want to be doing rasterization

tho i might add raytracing logic for far views

thing is that i wont only have voxels

i will have a lot more complex geo like half sized blocks, stair blocks, ramps, and stuff like grass

and i think the architecture im going with is more scalable once i start adding more types of stuff

it currentl goes so fast its basically l2 bound

so having it raytracing would change very little

and i want to try supporting scaling down to switch level, which i guess current arch can cope with (at lower draw distances) but raytracing woudl be tricky on there

[dark crater]

0.02 ms (basically min size) for the atomic reproject for shadow culling (1080p buffer, 4070)

the barrier cost is the expensive one in fact

[prime ice]

barriers really get expensive with 40

all the compute passes go lightspeed

[dark crater]

also after more benching the atomic thingy can reach 0.3 ms

when tere is super high contention, like when you directly look at the ground

so i think ill make it just with the tile asignment as IDs and nothing more as you dont need atomics there

[prime ice]

dhow code

[dark crater]

nothing really strange.
for every pixel of the gbuffer depth:
find world position of pixel
reproject to shadow-space
atomic-min its depth in shadow-space

but thats 2 million atomic-mins that vary on a per-pixel basis, so its kinda mediocre

if im far from the ground so that those atomic-mins cover a bigger % of the shadow-space, they get spread out and its ok, but if you look at the ground at ground level thats 2 million atomic-mins on the same texel

[prime ice]

ok

[acoustic bobcat]

I ended up having to work all last weekend and didn't have time to edit the article any, does the overleaf still have the most recent version or were Peter's suggestions addressed in the master copy

[sweet nimbus]

Nothing has changed yet

I'm in cali atm but I might find time to edit the article on my dad's Chromebook

[acoustic bobcat]

Alright

I was just going to give it another read through and make comments since I haven't looked it over in a long time

[sweet nimbus]

Same

The book is supposed to be released now I'm pretty sure, at least according to the original timeline

But clearly that must've been a mere suggestion

[hazy steppe]

dont forget to put sunscreen on my orange 🙂

[wooden jolt]

time is but a relativistic illusion

[sweet nimbus]

Does anyone understand the meaning of the third comment on the first page? It says something about "US-based"? I'm quite conchfused

[acoustic bobcat]

It says something-bound

US-bound or vis-bound or something

I was confused too

[sweet nimbus]

Maybe it's alluding to the benchmarks later on

Whatever I'll just rewrite it

[hazy steppe]

could always send the frog who wrote the comment a message to clarify

[sweet nimbus]

true, I just don't like sending an email for something so small

[acoustic bobcat]

It's peter the scawy editor

So worth trying to decipher before bugging him about it

[hazy steppe]

: )

[sweet nimbus]

Do we really need a brief explanation of VSMs after its mention in UE5 if we explain it in detail just two paragraphs later

[acoustic bobcat]

It reads a little jarring to just dive into the technical details without knowing what VSM is

Read it from the standpoint of someone who has never heard of VSM and has no idea what it entails or what its purpose is

[sweet nimbus]

Maybe I could move the talk about virtual textures to that spot

Then introduce VSMs as virtual textures for shadow maps

[acoustic bobcat]

Yeah perhaps

Or just virtual memory/anything as a concept

[sweet nimbus]

Yeah

[acoustic bobcat]

It can be pretty brief it just needs to give people an idea of what it even is they're about to read about

[sweet nimbus]

@acoustic bobcat FYI there are a couple sentences about VTs at the beginning of the Implementation section

I think I'll move those, but I'm just letting you know for now

I need to go afk for a bit

[acoustic bobcat]

Ah yeah

No worries you don't have to keep any of what I'm writing verbatim I'm just spitballing

[sweet nimbus]

Btw I hastily added a tiny section to acknowledge your help at the end

I usually only have a few minutes to add stuff though because I keep getting interrupted lol

[acoustic bobcat]

Lol np

[acoustic bobcat]

Are others going to be acknowledged, if I'm the only one we can just drop it lol

[sweet nimbus]

No one else really directly contributed lel

[acoustic bobcat]

I feel strange having a whole attribution section to myself lol but I will think about it

When are you submitting the draft back to the editor

[sweet nimbus]

Idk I'll just upload it to the main Overleaf article sometime

[vestal snow]

For the comment about one of the diagrams

do we just need to increase the resolution or should we redo it

[mystic lark]

I have a question, I was speaking with some people at my uni and they said that if I write down uni affiliation and like a sentence in the acknowledgments section they could give me money for it (as a part of a scholarship something something), would y'all be okay with that?

[sweet nimbus]

Lol sure

You should get the details clarified though

[sweet nimbus]

I will respond to this later when I'm at a computer

[mystic lark]

Yeah they'll send me the details in an email, so I'll inform you when I have it

[sweet nimbus]

I just got this email

Hi Jake and Virtual Shadow Maps authors,

We need you and your fellow authors to please enter your bios into AboutAuthors.tex asap please.

@mystic lark are you up? I'm not sure what info is needed and I don't have access to a PC to do it atm

@wooden jolt @vestal snow too just in case

Ok I found a PC that I have access to for a minute

Basically they just need a paragraph about yourself

"X works as Y at Z corporation and enjoys doing W" sort of deal

[wooden jolt]

will do

[mystic lark]

Uhh lucky, I was almost asleep lol

Booting my pc

[wooden jolt]

man this makes me realize that my graphics credentials are null and void

[mystic lark]

yeah like what do I put in lmao

[sweet nimbus]

I also think maybe we should change the title of the article to just "Virtual Shadow Maps", but I'm open to ideas

[mystic lark]

did we put the title in or was it put in for us?

[sweet nimbus]

If it makes you feel any better, my employment status changed about a week ago so idk either

[sweet nimbus]

I think we can choose

[wooden jolt]

should I say "undergraduate" or does that assume that I've already graduated (I didn't yet, coming soon in december )

[sweet nimbus]

Just put undergrad I guess

Or "student at"

Maybe we can just say we work at the tree of life

[mystic lark]

I just put some garbo 2 sentences in

you can't just swing this on me at 11pm when I'm ready to eep

Reading what others put in I think it should hopefuly be fine

[wooden jolt]

ye I just did mostly the same

[mystic lark]

I approve yours LVSTRI

[wooden jolt]

thank you, I approve yours too

[mystic lark]

eyy okay, off to bed with me hehe

[wooden jolt]

gnite mf (my frog)

[mystic lark]

btw I hope there is still time before the final release? This makes me a bit sus, why do they need it so suddenly?

[hazy steppe]

will no longer be zero my frog

[sweet nimbus]

I feel the same way, but I get basically 0 info

[sweet nimbus]

You just need to increase the resolution of the mentioned image

we also need a regular CSM impl for the benchmark

[vestal snow]

Oh crap I think saky has that one actually

[vestal snow]

Wait do we need to ship it or can we just use one we have sitting around

[sweet nimbus]

We just need something for the benchmark I guess

Honestly though idk if it's that critical to even do. It was just one comment

[mystic lark]

Ye I have cascades shadow maps impl with sample distribution

But I'll be able to add it next week the soonest

[sweet nimbus]

Pogg

That will be perfect

I'm editing the article on my dad's Chromebook atm

Addressing the notes and doing some general cleanup

[sweet nimbus]

Most of them should be addressed now

I'll make a quick to-do list for myself/anyone reading this:

• don't need to fix code formatting since it's fine in the main doc (maybe shorter variable names would make it easier to read though)
• need to increase figure 4's resolution
• need to fill empty space on page 14 after Sampling (probably using less [H], etc. will do the trick)
• need to address dynamic day and night perf again
• need to address the second comment in the appendix. Idk how yet

Also the whole thing with doing a benchmark for CSM/SDSM to compare against relatively state of the art shadow stuff

[mystic lark]

I don't have any fancy culling for the csm or anything though

I can hookup frustum culling we have but that's about it

But ig we don't have any occlusion culling for VSM also so it's fair comparison

[sweet nimbus]

SDSM could even be ignored as it probably performs almost as well as plain CSM

It just has the reduction pass

[mystic lark]

Yeh true, it's just a quality improvement kind of thing

[mystic lark]

okay so I got information about the grant - all I have to do is put this sentence into acknowledgments:
Matej Sakmary was supported by the Grant Agency of the Czech Technical University in Prague, project no. SGS22/173/OHK3/3T/13.
and mark myself as being affiliated with Czech Technical University in Prague - usually either underneath the name of the author or into the footnote on the first page.

but I don't really see anyone writing their affiliation in their articles

[sweet nimbus]

Hmm

[sweet nimbus]

You can email Peter and Wolfgang if you want to ask how they want affiliations to be handled. I can send you their email (cc me if you do)

[sweet nimbus]

"wraparound addressing" sounds weird. Maybe "sliding window" would be a better alternative?

I made sure the subsection headers are consistently lowercase, but I'm not sure what the proper convention is here

Looking at the previous GPU Zen book, it looks like sections and subsections are uppercase, while subsubsections (I think) are lowercase

Nevermind, all headers are uppercase. The lowercase ones are the bold "mini section" starting sentences (idk the proper term lol)

[mystic lark]

That'd be great yeah

[sweet nimbus]

I removed the appendix section (but kept the figures) and the [H] placement option. Now there aren't any pages with massive empty space

[sweet nimbus]

This is kinda bugging me. @mystic lark do you have any thoughts on renaming it?

[mystic lark]

Oh sorry I missed the message

Hmm well in all the clipmap papers they call it toroidal addressing, so perhaps we could change it to that?

[sweet nimbus]

I've seen "sliding window" used in other papers

[mystic lark]

I avoided that term as it's kinda specific and I felt like wraparound is more descriptive

Sliding window is also fine

[hazy steppe]

"Persistent homology of toroidal sliding window embeddings"

[sweet nimbus]

I mainly dislike the current one we have because it's not a standard term

"Toroidal" describes it well, but it's not as intuitive as wraparound or sliding window

[mystic lark]

Yeah true, in hindsight wraparound is a weird decision for a name (although I still feel like it's pretty descriptive)

[mystic lark]

Yeah that was my main fear, that's it's another special term

And that's why ultimately I decided against it

[hazy steppe]

wraparound is much more intuitive, the other 2 are purly mathematical

[mystic lark]

Yeah, but I also understand Jakers concerns, sliding might be a good compromise - especially if Jaker already saw it in some papers/materials

[hazy steppe]

yeah

[sweet nimbus]

"Sliding window" has an intuitive interpretation

[mystic lark]

I agree with changing it to sliding window

As long as the mapping is sufficiently explained (which I believe it is) I think it will be fine

[sweet nimbus]

Ye I'll change it when I'm back. Or someone else can

[mystic lark]

PC is off for today, but can do it in the morning tomorrow

Sorry I've not been so active btw, exam period and all...

Next week it should improve hopefully

[sweet nimbus]

Well the original schedule on the website apparently didn't mean jack, so I can't blame you

[sweet nimbus]

I changed the wraparound thingy but kept the original mention in parentheses since it's actually called that in the geometry clipmap article that you referenced

[sweet nimbus]

I did a cleanup pass and copied everything except demongod's attribution (bc no real name yet) to the main article

I added a few paragraphs of content

[sweet nimbus]

I'm done for today

[acoustic bobcat]

bingus

[wooden jolt]

what kinda frog is this

[remote quest]

it's a ca||f||r||og||

[hazy steppe]

i read "i added a few propaganda to the content" 😄

[sweet nimbus]

I just noticed that there was an edit to our article on june 2 (five days ago) 👀

[mystic lark]

Big edit?

[sweet nimbus]

one character edit

[mystic lark]

Oh lmao

[hazy steppe]

so

whats next

when can i buy the book

[sweet nimbus]

https://tenor.com/view/sleeping-sleep-dog-dawg-eeper-gif-5083970977419902566

@mystic lark when will you review my changes

I can offer you my promise to start working on that mieplot article again

[mystic lark]

Okay I'll do it today

ngl I forgor about the article

[sweet nimbus]

frfr

[mystic lark]

I should also add the damn cascaded sms to Tido

but ahhh I'm so lazy hehe

[sweet nimbus]

I think it'd be cool but isn't critical

[mystic lark]

It shouldnt be a lot of work as I already have it implemented in other projects and porting code between apps using daxa is pretty easy

[sweet nimbus]

so what you're saying is that you have no excuse jk

[mystic lark]

yeah no I really don't

[sweet nimbus]

I currently have no excuse for not finishing my article for you

[mystic lark]

Today I'll also finish the last higher prio interview assignment so yeah, really no excuses left

[mystic lark]

Okay I reviewed changes and as per usual added my own

[sweet nimbus]

I forgor to look. Will try to remember to check this when I wake up

[soft island]

I take it this schedule is old?

can't find the book online

[sweet nimbus]

Yeah the schedule isn't remotely accurate

Idk what the current one is

[sweet nimbus]

I also forgor to do this again 💀

[vestal snow]

[sweet nimbus]

is it bad to begin a sentence with an inline citation

I'm doing it, can't make this sentence not suck without doing that

[mystic lark]

I always try for the sentence to make sense even without the citation

But again, stylistic choice more than anything

[sweet nimbus]

where would I put the citation then 🤔

[mystic lark]

What's the sentence?

[sweet nimbus]

it's the one that you had a note on, saying you didn't understand what it was supposed to say

I reworded it

~\cite{haar2015gpu} is the only resource we could find that mentioned VSMs, and it only does so in passing.

that's the extent of VSM literature we could find

[mystic lark]

Ah I don't think we need to cite there tbh

[sweet nimbus]

I kinda like having the citation though 🥺

[mystic lark]

This is fine then

I don't think it matters that much

[sweet nimbus]

I also reverted one of your commaisms to make the sentence clearer

[mystic lark]

Yeah no worries

I suck at comas

[sweet nimbus]

I try to keep a mental note of sentence complexity

[mystic lark]

My autocorrect makes them for me mostly 🤠

[sweet nimbus]

having too many commas in a sentence can make it hard to read

[mystic lark]

Good point, you are right

[mystic lark]

You can even say something like "To our knowledge Haar et al. were the only ones to mention VSM, ...., cite at the end here"

[sweet nimbus]

ah yeah I forgot that you could use names

[hazy steppe]

dont overthink it too much : (

the people reading this stuff wont even notice and if they do they should come here and show off their engines

[ruby trellis]

when is your guys article releasing roughly?

[sweet nimbus]

no clue

[ruby trellis]

i remember something saying may or so

[sweet nimbus]

yeah their website

[ruby trellis]

is it viewable already? or is it private

[sweet nimbus]

it's private

the book that is

I could send you a pdf if you want to read our article

[ruby trellis]

i meant the article

I was thinking about reading it next week when I have some free time

since I should have everything set up until then

[ruby trellis]

would be nice yeah

[sweet nimbus]

ugh it's slightly too big for discord dms

[ruby trellis]

done

thank

[sweet nimbus]

alright cool

btw the code snippets are messed up in the pdf I sent

they are fine in the actual book though

[ruby trellis]

eh its not that bad

[ruby trellis]

when sampling and figuring out which entry in the VPT I want, can I imagine that as being axis aligned to the xz-plane?

as shown in the big visualized image in the diagram in the 3rd section

or is this explained further down? ive only read til there

[sweet nimbus]

are you talking about the grid in figure 2

[ruby trellis]

the image where the grid is overlayed with the render

so essentially the cascade hierarchy thing

[sweet nimbus]

the shadow map is broken up into 128x128-texel pages, and those are what you see in that pic

[ruby trellis]

yes i get that

[sweet nimbus]

which coordinate space are you referring to here

[ruby trellis]

oh wait

so this is from the suns perspective

nvm

stupid question

[sweet nimbus]

it's all good

I see why you were confused now

[ruby trellis]

thr image made it look like it wad axis aligned

but you can see its angled on the wall in the back

[ruby trellis]

The first heuristic prioritizes achieve pixel-perfect shadows—a one-to- one mapping of screen pixels to shadow-map texels.
shouldn't this read achieving?

[mystic lark]

Yes

But iirc this sentence should be removed

[tame summit]

You said there's an article to read somewhere here, LVSTRI? That explains VSM more

[acoustic bobcat]

They're writing it

It will be published at some point

Seems like it's been delayed or something I haven't heard any news about the publisher

Assuming that is still happening?

[sweet nimbus]

Nothing has been cancelled

[tame summit]

Ah alright well I'll be here when it's available 🙂

[humble shadow]

You have to join the cult if you want to share the knowledge, it's the rule

[sweet nimbus]

useful reading for writing papers (criver just posted it in #mathematics)
https://cs.dartmouth.edu/~wjarosz/writing.md.html
https://people.csail.mit.edu/fredo/PUBLI/writing.pdf

[sweet nimbus]

@wooden jolt @mystic lark @vestal snow
I just got an email saying that we should address feedback on the article and do final touches

[vestal snow]

awesome

new feedback just dropped in the doc?

[wooden jolt]

pog

[sweet nimbus]

Yes I'll finally review saky's changes

[mystic lark]

https://tenor.com/view/its-been84years-its-been-a-long-time-rose-dewitt-bukater-titanic-gloria-stuart-gif-16850506

I should fix the code in Tido I guess

[prime ice]

killlee

[sweet nimbus]

!remindme 9 hours review gpu zen article NOW

[lost jay]

New Reminder | ID:73070177

Alright gpgpu, I'll remind you in 9 hours about:

review gpu zen article NOW

[wooden jolt]

ah yes

review the article NOW (in 9 hours)

[mystic lark]

I'll try to fix the code over the weekend, I've been incapacitated this whole week due to sickness ☹️

[humble shadow]

Looks like several folks in GP have been sick the past few days (me included).
Is this what it feels to be part of a "Superspreader Event" ?

[ruby trellis]

it’s called autumn

[mystic lark]

Yeah I think I did catch it from a colleague at work

[prime ice]

tons of people sick

[hazy steppe]

sickkos

[humble shadow]

sick's town

[sweet nimbus]

I will diff the actual article and the staging article then try to address the comments while inserting our latest revisions

[hazy steppe]

were there lots of comments from wolfangs's side?

[sweet nimbus]

no, each section has different reviewers

[hazy steppe]

i mean from people who are not frogs from here

[sweet nimbus]

ah

all but one of the comments were about the figures actually

7 comments

[hazy steppe]

i see

[mystic lark]

Hmm so no major text changes?

[sweet nimbus]

yeah, just some figures need to be mentioned in the main text

@vestal snow there are a few minor changes (e.g. font size increase, capitalization) that need to be made to some of the figures

man the diff is showing way more changes than I thought there'd be

@acoustic bobcat do you want to be attributed in the acknowledgements?

[acoustic bobcat]

Nah it's fine

[vestal snow]

Alright I can reupload the ones I did

Today hopefully (I’ll try)

[sweet nimbus]

we also need a conclusion which I can whip up after transferring the latest edits

culling against a HiZ
this used to say Hi-z, but idk which one is preferred

if I wrote it I probably would have said "hi-z buffer"

[acoustic bobcat]

Is that term defined anywhere

It should be, as long as it is it probably doesn't matter

Google both and see which one is more commonly used

See which one is used in GPU Gems

[sweet nimbus]

big brain

some other article uses Hi-Z

[sweet nimbus]

when you do so, make sure to upload to the main gpu zen article, not the staging one

I renamed the staging one to indicate that no one should change it anymore

[vestal snow]

Ok that sounds good

[sweet nimbus]

fig:depth-analysis needs to have "uv" changed to "UV" three times (putting this here because the comment is about to get deleted)

[vestal snow]

fixed that one

we need to change figures to use similar font to text right?

[sweet nimbus]

yeah that'd be ideal

[vestal snow]

ok I reuploaded everything I have

I'm having trouble compiling the article to debug view it though

[sweet nimbus]

yeah I can only compile it in fast mode 🥲

figures don't render though 😦

[sweet nimbus]

I'm done editing the main article. we still need to reference the figures in the appendix in a yet-to-be-written conclusion though

the diff between the Real™️ article and the staging version should be minimal now, the few differences that remain are edits I didn't agree with

[sweet nimbus]

@vestal snow I forgot to mention, but you can plug your new figures into the staging article to see how they would look in the real thing

since it's actually possible to full-compile that one

unrelated: I'm not sure we actually need a conclusion. I can find other places to reference the appendix figures

I'll let whoever reads this vote on it, as I'm ambivalent myself

[sweet nimbus]

I'm done addressing all the comments. @mystic lark and/or @wooden jolt if you guys have some time, could you skim the official article (not staging) to make sure there are no obvious issues

[mystic lark]

I'll do it around 5pm (~8 hours)

[sweet nimbus]

I think I'll email them that the article is ready after you review it

[mystic lark]

Did they mention a deadline for the code?

[sweet nimbus]

nope

[vestal snow]

Yeah I’ll add the latest figures to staging soon

[mystic lark]

I assume something till next week then

[sweet nimbus]

yeah idk the timeline still

"within a few days" is the one I'm working with

[mystic lark]

Hmm I see I see

[sweet nimbus]

reality is probably a few months

[mystic lark]

Don't tell me that or I'll procrastinate on it again

I think I'll create a GPU Zen branch of Tido and try to finalize it there

[sweet nimbus]

the timeline for the code is probably a lot looser since that's all digital

[hazy steppe]

my frog probably meant april 2025

[wooden jolt]

ayeaye sir

[wooden jolt]

how do you turn off compilation again

it keeps timeout'ing

[sweet nimbus]

uhm select fast compile

it shouldn't time out then

[wooden jolt]

here's hoping

[sweet nimbus]

but figures will be gone

also you can try the visual editor which I think shows figures

but they probably aren't laid out properly

[wooden jolt]

ok I proofread the thing

it's solid so far, but there's a few comments I'd like to uncomment and maybe expand a bit more on

also an editor asked for a conclusion, idk if we want one

[sweet nimbus]

yeah I made an executive decision to not have one

there's at least one other article that doesn't have one either

[wooden jolt]

gotcha boss

what about the future works section

we shouldn't make big changes right?

[sweet nimbus]

ye

[wooden jolt]

alright

[mystic lark]

Oh I forgor about this

Today then sorry

[sweet nimbus]

same

[hazy steppe]

if you want i can keep track of yalls and make sure you get this stuff done in time : )

[mystic lark]

I appreciate any form of digital whipping

[ruby trellis]

did anyone address the one single issue i had with the paper

[sweet nimbus]

I forgor what it was in the sea of issues

[ruby trellis]

same

[prime ice]

i feel invited

[mystic lark]

only VSM related, I will not suffer onslaught of "rewrite task graph with me Saky" messages

[sweet nimbus]

@mystic lark @vestal snow when you guys are done I'm going to email the editors and tell them we're ready

[vestal snow]

Ok, I’ll upload the images to the staging area and see how they look today

[vestal snow]

I moved them over. They're in the main article and the do not edit article

they seem to look ok

[sweet nimbus]

Epic. I'll take a look in a bit

[vestal snow]

cool let me know if any figures need changes!

[sweet nimbus]

@vestal snow did you increase the font size on Setup phase flow chart.png? it still seems kinda small

[vestal snow]

is that the one with the read/write lines? I think that's actually a saky diagram

[sweet nimbus]

ah my bad, I just assumed you did all of them

@mystic lark (can you increase the font size in the aforementioned figure pls)

btw the editor moved some figures around in the main article 👀

[vestal snow]

oh dang, major shifts?

[sweet nimbus]

it doesn't seem too major

they also changed some minor formatting stuff with some of the figures

[mystic lark]

Uh oh that figure is still in?

[sweet nimbus]

is it supposed to be gone?

[mystic lark]

It was kinda a mock and I thought it was supposed to be remade

But I can try to look if I have the original somewhere

Or remake it

[vestal snow]

Oh I didn’t know that

In that case one of us can remake it. I can help if you want

[sweet nimbus]

Yee

[mystic lark]

Forgor to respond, I can do it, but also won't protest if you do it iykwim

[sweet nimbus]

either way we need to get this done asap

[hazy steppe]

@mystic lark do it now

[mystic lark]

I'm uni bound

Should I do it on my phone?

😈

[hazy steppe]

you dont have a laptop with your shit?

[mystic lark]

No laptop:(

It broke and me lazy to get new

[hazy steppe]

!remindme 6hr remind saky to do the thing

[lost jay]

New Reminder | ID:73289117

Alright deccer, I'll remind you in 6 hours about:

remind saky to do the thing

[humble shadow]

A reminder to remind you to remind saky

[sweet nimbus]

!remindme 6hr remind deccer to remind saky to do the thing

[lost jay]

New Reminder | ID:73289199

Alright gpgpu, I'll remind you in 6 hours about:

remind deccer to remind saky to do the thing

[mystic lark]

!remindme 5hr remind me that I will soon get reminded and should mentally prepare

[lost jay]

New Reminder | ID:73289664

Alright m_saky, I'll remind you in 5 hours about:

remind me that I will soon get reminded and should mentally prepare

[hazy steppe]

inb4 18hrs later
Saky: Oh shit i fell asleep, will do it asap.
inb4inb4 12hrs later later
Saky: Shit, I forgor

[mystic lark]

At this moment I can neither confirm nor deny these accusations

[hazy steppe]

time is up 🙂

@mystic lark

i bet you saw it coming hehe

[mystic lark]

https://tenor.com/view/sad-frown-rain-cat-gif-17035614

Was hoping Jaker wouldn't remind you

What is JS crafting the images in? Power point?

[hazy steppe]

jakar didnt remind me either, he eeping

@vestal snow saky needs help

[mystic lark]

okay I'm taking a shower and if JS is not here by then I'll just cook in powerpoint

[vestal snow]

Yeh it’s PowerPoint

It’s fake photoshop . I’ll be around to help

[mystic lark]

What font do you use?

[vestal snow]

looks like Latin Modern Roman
edit: yeah it's that one

[mystic lark]

wtf fancy font

I have modern love though

I have modern love though

I need to restart to add new font to powerpoint

ahhhh

[vestal snow]

oh maybe that's what I had to do lol

[hazy steppe]

only powerpoint, no need to restart the system

[vestal snow]

I think I googled something like "what's the closest font to overleaf default" and it was latin modern

[mystic lark]

I did that didn't show up

[hazy steppe]

then logout and login?

[mystic lark]

And some random forum post from their thingies said you need to restart pc

[hazy steppe]

weird

[mystic lark]

still not there

HAHA

but system shows it as installed

I love you so much Microsoft

[vestal snow]

hmm that's super weird

[mystic lark]

I am not crazy

[hazy steppe]

and Modern Roman?

[mystic lark]

😔

I'll just make the image and figure this out later

[hazy steppe]

why are you using a script like font anyway?

[mystic lark]

@vestal snow how do I export the image?

ah figured it out

this good?

[sweet nimbus]

How big is that text compared to the original

[sweet nimbus]

Epic

Looks good

[mystic lark]

okay I uploadeded it

[sweet nimbus]

To the main article?

[mystic lark]

ya

[sweet nimbus]

Can you reply to the comment from the editor

To tell them it's fixed

[vestal snow]

looks nice!

[mystic lark]

how do I view the comments again?

ah nvm got it

okay replied

seems like the fonts exported a bit weird into the png

is that really how I'm supposed to do that?

[vestal snow]

what happened to them?

I think all I've been doing is select all > save as picture

I don't think I screenshotted any

[mystic lark]

eh I guess it's fine

[sweet nimbus]

As long as it's readable and the font is roughly the same as the book font, it's good

[mystic lark]

okay I'll leave it as is

[hazy steppe]

do they not provide the font?

i have so many questions

[sweet nimbus]

hmm it might be a ttf in the book actually, idk

[humble shadow]

The arrow tips would benefit of being bigger. Not sure how well this will read once printed.

[mystic lark]

better?

(I can only make the arrows more fat unfortunatelly)

[humble shadow]

much better yeah

[mystic lark]

oki, I uploaded this version instead

[sweet nimbus]

Sweet, are all the comments addressed?

If so, I can tell them it's good to go

[mystic lark]

I'm not sure, I didn't know I should 🇨🇿

I can do it quickly tomorrow at work

[sweet nimbus]

I can check after I shower. It'll only take a minute

[sweet nimbus]

I sent the email

[wooden jolt]

lovely

[weary hollow]

https://www.youtube.com/watch?v=h0c83RnEMSo

[mystic lark]

Is good?

[weary hollow]

yes

the real project is some unreleased title

[wooden jolt]

you know I was reading the device generated commands spec last day

and I have a devilish idea in mind

beyond vsm is possible?

[weary hollow]

https://tenor.com/view/elaborate-gif-27637853

[prime ice]

tell

[wooden jolt]

dynamic scissoring

you organize pages into a quadtree and draw them top to bottom into a hw depth target

unfortunately it can't be done right now because you there is no command token for vkCmdSetScissor

but it is planned for the future

[mystic lark]

https://tenor.com/view/thats-what-she-said-what-she-said-dirty-joke-joke-laugh-gif-15661968

Sorry

https://tenor.com/view/out-disappear-bye-vanished-gif-4932063

[ruby trellis]

a vkCmdSetScissorIndirect would also be neat lol

[wooden jolt]

incredible

[prime ice]

i have never set scissor

how does it help, isnt it just clipping?

oh thats big tho

[prime ice]

crrinnge

ooooh with the scissor you can change where you write to

[wooden jolt]

yeeep

[prime ice]

so you can select rhe physical page

ooooooooh

[wooden jolt]

it's the address translation trick

[prime ice]

big

nice

ok we bully khronos

to qdd it asap

[hazy steppe]

can you not prototype with mesa? 🙂

[prime ice]

i dont touch lunix but maybe lvstri will

[hazy steppe]

and then vulkanized 2025 "how we coerced John into using scissors properly"

[prime ice]

wait a sec tho

uuuuuuh

this only works if you draw per phys page

[wooden jolt]

that's what the quadtree idea is for

[prime ice]

hmmmm

[wooden jolt]

the quadtree just bins scissor regions to meshlets

[prime ice]

ok, but the quad tree tiles still need translation then

hmmmmmmmm

like on the lower quadtree levels youd have to have a lot of bins for all the tiles

[wooden jolt]

hmm

the brain is working

[prime ice]

virtual memory

[wooden jolt]

me thinking

[hazy steppe]

VSM = Virtual Scissoring Mekanism

[mystic lark]

I was thinking about this previously without the quadtree idea

too smoothbrained for that, I still dont' really get it honestly

but draw per page discouraged me

[honest berry]

https://x.com/miketuritzin/status/1844434711244120393

Mike Turitzin (@miketuritzin) on X

This article on virtual shadow maps is insanely detailed (and good).

I'm not sure how similar the implementations are, but this technique is also used in Unreal Engine.

https://t.co/pdhT2thhk2

Twitter · Oct 10, 2024, 5:47 PM

[hazy steppe]

people still using twitter

[honest berry]

My colleague just sent me this link. Yes, now they come to us.

[hazy steppe]

we welcome all frogs from all corners of the world 🙂

[mystic lark]

Meh

They kinda do what you'd expect from the title

The dynamic thingie is cool

I was more wondering if you could make this work with http://gpuzen.blogspot.com/2019/05/gpu-zen-2-parallax-corrected-cached.html?m=1

Might be way cooler with way less visible artifacts

Tbh kinda interesting that UE didn't do what they talk about way sooner

[weary hollow]

That would be a win win. I forgot about the parallax corrected shadows

[mystic lark]

I think it also might be interesting if the wary the time slice period based on the clip level

Etc etc

(maybe they do I kinda lost interest after 10 minutes and rushed through it)

[weary hollow]

I think they do that

[weary hollow]

You spend more time than me watching it

[mystic lark]

Yeah I thought it was gonna be a bit more

The perf improvements in the end kinda cope imo

[weary hollow]

If you look into their unreal editor they have more time slices or whatever thing is called

[mystic lark]

Hmmm yeah

[prime ice]

wasnt it 50%?

[mystic lark]

It's time sliced

Makes no sense to talk about "perf benefits" when you just times slice

Imo

Since the speedup should always approach the factor of into how many frames you decide to time slice

"speedup"*

[hazy steppe]

@mystic lark https://graphicsprogramming.github.io/blog/generating-phase-functions-luts-with-mieplot please 🇨🇿

[sweet nimbus]

I posted it in #1166523139468034068 too

[mystic lark]

Oh I saw, I will take a look tomorrow!

Thank you mister Jaker

[hazy steppe]

2 times is ze charm

!remindme 1d saky mieplot review

[lost jay]

New Reminder | ID:74131947

Alright deccer, I'll remind you in 1 day about:

saky mieplot review

[sweet nimbus]

Hehe, keep him on his toes

[hazy steppe]

can you update the link in timderdoodle?

[sweet nimbus]

Ye I was just about to

There we go

[dark crater]

oh thats really useful thanks for the link to that phase thingy

my volumetrics lack proper phase calculation for the scatters

[sweet nimbus]

note that you need multiple scattering for the rest of the fog to not look dark

hg spreads the light everywhere so it doesn't look too dark

[mystic lark]

or just add a constant factor to each step

[dark crater]

i already give some ambient terms to the fog and everything is mostly completely done by eye

so that should give me the multiscattering look i think

[sweet nimbus]

Ayo the gang's here
http://gpuzen.blogspot.com/?m=1

[mystic lark]

We made it chat

[wooden jolt]

let's fucking go

[hazy steppe]

about time

darius is also on the server 🙂

[sweet nimbus]

GPU Zen 3 is gonna be huge, physically

[wooden jolt]

the vsm article is 99% of the mass of the book

due to the massive balls of all of us

[hazy steppe]

🙂

it also comes with a diorama of ze bistro

which you can unfold in the middle of the book somehow

and use your torch light to cast some of them shadows

[remote quest]

8   Real-Time Ray Tracing of Large Voxel Scenes

this better be good

[hazy steppe]

i need to do a world tour it seems because i want my fisikal copy signed by yalls 🙂

[humble shadow]

Ho that would be nice, I would want the same 😄

[sweet nimbus]

btw I just checked overleaf and it looks like the book is no longer shared with us or it got privated

I can't remember how much we edited the article after I renamed our staging article to "DON'T USE ANYMORE", so I might have to ask wolfgang if we want the latest revision to post elsewhere

[remote quest]

[sweet nimbus]

ok I deleted it (after you dl'd it) because I gotta see if the other authors are down to let everyone see it hehe

[mystic lark]

I think our last revision is on their OL no?

[sweet nimbus]

what's OL

overleaf

[mystic lark]

Overleaf :D

[sweet nimbus]

anyway yeah it's on their thingy, which we can't access if we want to post the article somewhere else

[mystic lark]

I think you copied it when they asked us last time to fix the thingies?

[mystic lark]

OH this is what you meant

I forgor

[sweet nimbus]

we did the final editing and revisions directly in their overleaf

[mystic lark]

I thought you meant that you are not sure if their version has all the changes

[sweet nimbus]

oh ye, I'm not worried about that

my issue is much more trivial

[mystic lark]

okok sorry, I just got confusled

[hazy steppe]

smells like you want to publish it on ze blog if wolfi says yes?

[humble shadow]

Confirmed: https://bsky.app/profile/wolfie17.bsky.social/post/3lbn74oblkc2k

Wolfgang Engel (@wolfie17.bsky.social)

The new one is much bigger! Background is Olivia at her happy place :-)

Bluesky · Nov 23, 2024, 7:11 PM

(I was going to say "hope the cost won't be too high", but work is going to reimburse me anyway)

[dark crater]

damn thats a chonker

[hazy steppe]

street shoes in the house is something i will never understand though

[eager moat]

maybe is cold

[dark crater]

@hazy steppe we do that in spain

its common in dryer countries

[hazy steppe]

no

just switch to house shoes

[humble shadow]

Ho wait, it's actually out, I can buy it on Amazon

Lezgoo

[hazy steppe]

12 Virtual Shadow Maps by Matej Sakmary, Jake Ryan, Justin Hall, and Alessio Lustri
: )

the frog behind chapter 14 is also on the server

[viscid shore]

Another book yay

I will get it

Purchased

[mystic lark]

interesting

they didnt' ask for the code yet

lol

[dark crater]

3 euros for pdf version???

based

[hazy steppe]

communication seems to be top notch

[honest berry]

Congrats everyone. Just bought a paper copy and eager to have good shadows.

[humble shadow]

I thought it was only for 2 chapters at that price at first 😄

[viscid shore]

Nice

there's a lot of good stuff in this book I should check out the others? this is v3?

[sweet nimbus]

I have gpu zen 2 and there are a lot of good articles in that too. idk about the first one

[honest berry]

It’s that day.

[viscid shore]

sorry to doxx btw

[mystic lark]

Wonder if we'll get one

Or if I have to buy it

[sweet nimbus]

lol the thing said authors get a free copy but idc either way

[viscid shore]

the book was not cheap

it's textbook prices

I bought 1 and 2 too and now I have to declare bankruptcy

[sweet nimbus]

damn

probably because it's huge

[humble shadow]

Just got mail, it's quite a beast

Print quality is okay, as usual images are dark (Cyberpunk chapter is quite dark).

[dark crater]

PDF releases the 30

[hallow oasis]

Congrats! Books contents is out of my league but still super happy for everyone here 🙂

[worn escarp]

congrats

[severe orbit]

Will this book be applicable to a Vulkan beginner like me?

Or really any of the GPU Zen books?

I'm very interested in the fluid sim chapters of GPU Zen 2

I made a GPU accelerated SPH simulation in OpenGL once

[sweet nimbus]

These books are aimed at an intermediate to advanced audience, but that doesn't mean a beginner can't find anything useful in them

[severe orbit]

It was fun AND cool 😎

[humble shadow]

Yeah, I think a few of the article are accessible, some other are quite complex. I definitely can't understand the whole book, for now at least.

This kind of book is nice to look at again in a few years.

[dark crater]

dammit the amazon version is DRMd to hell

is there a way to get a PDF

need to get kindle reader app to even read it :/

[sweet nimbus]

idk

[dark crater]

cant put it on the ebook reader i have

couse amazon drm so amazon stuff only

yeah i can only read it on the phone until some based ruskies or chinese pirate the PDf and bring the superior version

sucks

[meager relic]

you guys are crazy this 3rd version of gpu zen books is lit, lots of techniques to learn about!

[dark crater]

the gpu driven section is great

sort of a much better version of the vkguide article

[meager relic]

book is about modern rendering in a nutshell imho

scary to think what will be in the 4th edition (:

[hearty beacon]

gonna get this for christmas

[glossy pawn]

Yep, there’s a way. I converted my Kindle copy using Epubor Ultimate Converter. I’m not sure if it’s entirely legal, I only use this program to convert my legitimately acquired digital books. Make sure to use version 3.0.186, as newer versions seem to struggle with converting Kindle books.

[vestal snow]

That’s awesome!! So glad it’s finally out

@sweet nimbus are you posting it on your site? Is that still a thing or did they change the rules?

[sweet nimbus]

I think a more appropriate place would be the official gp blog, and yeah afaik they're still cool with it

[prime yoke]

The kindle version of GPU Zen 3 is 3$ on Amazon 😉

[mystic lark]

Btw I meant to ask, uni is asking me for two things:

1. if I can send them the PDF version of the article (do we even have that?)
2. they are asking for a percentage of my contribution to the article (apparently no one really checks that unless you do a PhD etc etc which I don't plan on, but yeah...)

[sweet nimbus]

you'd have to ask Wolfgang for a copy of the actual article, but we still have access to the staging article on overleaf which is 99.9% the same

also I need to cancel my subscription lol

[mystic lark]

Ahh okay okay, I'll just send that

It's not like it really matters I don't think, it's just for the internal database

[severe orbit]

Our German friend wasn't happy with the print quality

[sweet nimbus]

@mystic lark @wooden jolt @vestal snow I forwarded an email to you guys about shipping for gpu zen tshirts

[hearty beacon]

it just arrived

[mystic lark]

@sweet nimbus @wooden jolt @vestal snow - I have received a question from my supervisor regarding the contribution percentages, more specifically if the article is 100% as a whole, how would we split it into parts regarding the contribution, my suggestion was 40% me, 30% Jaker, 20% Jstephano and 10% LVSTRI, but of course that is just my view, so if you disagree we can ofc disagree. Once we agree on something, do you think it would be possible to get an email from each of you agreeing with the distribution (once have one we all agree on ofc)?

This is because of a grant that I received from my university - it needs to be put into our internal uni system and they require the contributions of individual authors

sorry I'm bothering you with this

[sweet nimbus]

arguably you did more than 50% because you also supplied the demo

[mystic lark]

well not yet 😅 noone asked me about it yet

[sweet nimbus]

lel

yeah kinda weird that we were never asked about it. now I feel bad because you worked so hard on it

but the code is still there and people are free to link it, so it's not like it was wasted

[mystic lark]

No worries, I'm making it into other thingies so I would have done the work anyways

[sweet nimbus]

anyway, your percentages seem fine

[mystic lark]

okay! do you all have my email? I can dm it otherwise

actually I'll dm my uni email, just to be sure

[sweet nimbus]

I do. what do you want me to send you?

[mystic lark]

I think something like should be enough

I agree with following author share for GPU Zen 3 - Virtual Shadow Maps article:
Matej Sakmary - 40%
Jake Ryan - 30%
Justin Hall - 20%
Alessio Lustri - 10%

[sweet nimbus]

normally I'd wait to get consensus from the others, but I don't think it matters that much 😄

[mystic lark]

From what I've gathered it is just a formality yeah

It might matter in case I (or some of us) decide to pursue PHD, this contribution would then decide how relevant this publication is to your PHD publication requirements etc... (at least here that is)

[sweet nimbus]

alright, I'll change it so that you get 99.99% and everyone else fights for the scraps

jk I'm just using your numbers

[remote quest]

no need to fight, i'll gladly take the 0.01%

[sweet nimbus]

email sent

[mystic lark]

Duel over the 0.01 contribution

[vestal snow]

Sounds good to me, I'll send an email

[hazy steppe]

https://tenor.com/view/rosycheeks-approved-i-approve-well-done-good-job-gif-16996150

[upbeat cargo]

Am I correct in understanding that because VSM cascades are snapped to the grid, flickering shadows (due to the camera moving) are eliminated?

[sweet nimbus]

yeah

you can achieve that with regular CSM too but it would be more wasteful in terms of memory usage

[upbeat cargo]

yeah. I was thinking it's one of those "free" fringe benefits

VSM seems to be not for the beginner, and I'm still beginner, but I'm gonna try to figure something out anyway

I think the main problem is going from screen space to page space

[sweet nimbus]

you know how to go from screen space to world space with the depth buffer and projection matrix

then you can go from world space to light space with the respective matrix for the light/cascade

then light (clip) space to page space is a shrimple linear map

the same math you'd use for shadow map UVs come to think of it

[upbeat cargo]

it's actually the screen space to world space that's the tricky bit for me: I haven't done that yet (I store the full position in my g-buffer (on my TODO list for fixing))

I understand the concept, but I haven't worked out the details

but using the depth buffer for that is actually the easy part 🙂

but yeah, it's probably not that hard

[sweet nimbus]

like many things in programming, it's a million little problems

[upbeat cargo]

yeah

[mystic lark]

it's at least a good learning experience, I for sure learned a ton implementing VSMisms

[upbeat cargo]

yeah, I bet

[mystic lark]

going from screen space to page space is the same transform as doing a shadow map test btw

[upbeat cargo]

right now, trying to make some progress with some actual game stuff, though

(camera stuff atm)

[honest berry]

I was just thinking this morning that I should implement virtual shadow maps.

[mystic lark]

https://tenor.com/view/raiden-cope-raiden-all-cope-raiden-no-hope-raiden-all-cope-but-no-hope-raiden-coping-gif-26286406

[weary hollow]

Same here but I want to delay it as much as I can

[radiant grove]

if you delay enough you can just have RT shadows

[weary hollow]

Thats my idea

[upbeat cargo]

probably already known here, but I found this: https://cescg.org/wp-content/uploads/2025/04/Sakmary-Resolution-Matched-Virtual-Shadow-Maps.pdf

[sweet nimbus]

that's @mystic lark btw (who worked on the gpu zen article)

[upbeat cargo]

yeah

saw that

that, and [13] are why I posted the link

[sweet nimbus]

epic

I have been cited

[buoyant lily]

Hey kiddos

do you have any benchmarks or know of any benchmarks

that show how much slower sampling from a Sparse Residency image is

assuming all pages are bound and never unbound

[mystic lark]

I never actually tried the hw sparse thingy :(

@wooden jolt did hw sparsizms fwir

[buoyant lily]

I did some googling

and couldn't find anything except for the info that sparse texture turns off Anistropy on AMD and limits it to 4 on Intel

[mystic lark]

I'd imagine sampling would be the same speed no?

Or do you think there is another indirection?

[buoyant lily]

well from my talks with driver peeps it seems like the whole Sparse stuff forces slightly more book-keeping from the GPU

cause on nvidia the actual sparse thingie is 512bytes

so they're like 16x8 IRL on a 4byte texel texture

and there's essentially shittons of pointers to update within the hardware standard 128x128 page footprint when you rebind