DDraceNetwork - developer

ddnet-maps

[ddnet-maps:master] 1 new commit

c4a5f83 A Just Every Fly - ddnet-maps

Jupeyy

[ddnet/ddnet-data-svg] Pull request opened: #32 Add game.svg by mind & Voxel

i'll leave this draft until the new shield icons on the ddnet repo's game.png is decided

def-

[ddnet-scripts:master] 1 new commit

c5bfbe4 fix - def-

def-

[ddnet-web:master] 1 new commit

cd2e1c9 GER1 - def-

would it be possible to add a NetStruct to data types? (e.g. similar to NetIntAny, NetArray, ...)

or, perhaps not a good idea to use anything recursive

def-

[ddnet-web:master] 1 new commit

676e125 tur1 info - def-

cookstown

[ddnet/ddnet] Issue closed: #5071 (feature request) +release for binding actions on key releases

We seem to have an issue with how we linked libnotify

12:43

cmake succeeds when it's not installed

yes

12:43

we dont check it

We should, right?

would be nicer yes

Ninecloud2077

[ddnet/ddnet] Issue opened: #5081 Translation For Server and Cilent

I want to translate a more language for DDNet Cilent But I don't know what should I do Then I want to translate for Server But I can't find any json file So how to translate for ddnet? thx

@Jupstar ✪ the given error for vulkan things not being found is also very meh

what does it say?

https://paste.pr0.tips/t7

ahh i guess https://github.com/ddnet/ddnet/pull/5076/commits/f595318b830653ca13f462e6949fbf80eb7b6860 fixes this

I can't believe I've been using my computer at 800mhz, things compile so quickly now :3

in 20 years i cant believe ive been using my computer at 800ghz, things compile so quickly now xd

1

too bad frequency isn't increasing anymore

13:40

13:41

power wall hits hard

Chairn

power wall hits hard

true story

13:42

i want more single thread performance xd

13:43

now they throw more l1 cache to it and it works xxd

Chairn

Click to see attachment 🖼️

where is that from, i want more stats like that

13:44

performance related

13:44

also for gpus xd

hum, Hennessy & Patterson is a good source

13:44

its the bible for system architect (edited)

13:45

lol, i typed biblio instead of bible, phd really starts to consume me

too bad they stats always end in 2010

13:46

13:47

modern cpu are nowadays more dense in terms of power than a nuclear reactor

nice

13:47

and intel probs already beats the sun at this point xd

hence the power wall, you can't cool them down efficiently

13:47

also dark silicon problem if you have ever heard of it

intel cryo cooler

cool by activly using electric energy

Jupstar ✪

cool by activly using electric energy

(edited)

Chairn

Click to see attachment 🖼️

are you also facing such problems in your memory designs?

partly, but not completely

13:50

hence my phd about computing in memory directly because there is way less power density for the moment

xd

Jupstar ✪

cool by activly using electric energy

so obviously, this is not possible

faster memory = faster effective CPU, i am fine with that

actually, we would need to slow memory a lil bit, depend on the technology

Chairn

too bad frequency isn't increasing anymore

Well each cycle is still becoming more efficient, so comparing clock speeds isn't always really fair

Chairn

so obviously, this is not possible

if you get close to super conduction?

and we can't compute anything there, mostly parallel stuff for AI or BLAS stuff

Learath2

Well each cycle is still becoming more efficient, so comparing clock speeds isn't always really fair

it is because power scales with f²

Jupstar ✪

if you get close to super conduction?

heat must go somewhere, superconduction don't apply to semiconductor

with other materials?

then it's a drastic technology change, we have nothing close in that matter

13:53

maybe optronic might be a game changer

13:53

but we still have a long way to go

Chairn

it is because power scales with f²

Well yes, so it might actually be good that f isn't increasing anymore

yes i want my alien hardware

spintronic is also interesting in the way it works

13:54

in spintronic, the path (copper wire) is actually the gate

13:54

its quite weird, fucking quantum physic

If we can extract more performance while not having to increase the frequency it's a win, and if raw performance starts to stagnate we might finally get back to the golden age of optimization when people actually cared about writing nice code

13:55

win win

Learath2

If we can extract more performance while not having to increase the frequency it's a win, and if raw performance starts to stagnate we might finally get back to the golden age of optimization when people actually cared about writing nice code

tru, it already happens a lot

13:55

script languages get c bindings for slow paths

well, you cna't have more performance without more frequency or more parallelism

13:56

hardware operators have already been quite optimized

but there is still room

so we really need a technology shift

well you can pipeline more, we most definitely can parallelise more

pipeline more --> more stalls when branching

13:56

15 stage is around the norm

--> more people care about writing nice branchless code --> I'm happy

in 2005/2006 they had up to 40 stages and reverted because it was highly inefficient

13:57

you cant write branchless code

13:58

well, you can with predicated/conditional hardware operator

13:58

not the norm in x86

13:58

but that's what is used for GPU

but we still have a bit of headroom for transistor sizes and for the overall gaps between the transistors(for logical units) & other components(memory controler etc) dont we?

interesting tab about simd extension:

13:59

well, intel is still at 10nm, so they can go down up to 2 or 3nm, but they already took 10 years to industrialize their 10nm process

3 silicon atoms would be around 0.6-1nanometer

14:00

ez

14:00

all we need

TSMC does sell 5 and 7nm now, but technology is slightly different and less efficient (performance wise) than intel's finfet technology

Jupstar ✪

3 silicon atoms would be around 0.6-1nanometer

I think quantum fuckiness prevents us from going that low

silicon is just the base die

i dunno i think the quantum effects are only the space between transistors not the transistors itself?

you need germanium, carbon and some more as well

but tbh, who the fuck knows xd

I'm not especially well read in semi conductor technology, but if you go that small the uncertainty in the position of the electron starts being an issue

and the fabrication process would be a mess

14:02

they already have huge problems to get a good yield from their wafer

14:02

like 50% are thrown out due to defects

i pay my 2million bucks for 1 trillion fps

damn, 1ps per frame

Fwiw, I'm not especially concerned about not getting more performance, I think we have plenty enough we are just wasting right now, if we don't get improvements for a while we'll just have to get smarter about our use of cycles

i bet as soon as x86 is the reason for no more performance it dies faster than the industry can imagine rn xd

risc-v ftw !!

but is risc really good in your argumentation?