🔌│tech

[cursive summit]

For an ATX build obviously you can go with a high end Noctua and it'll be great

[final flame]

Aios are trying to emulate a custom loop but they simply do not have the volume of liquid a normal loop has which forces them to run the fans higher to compensate

[cursive summit]

But a lot of these ITX builds have very little clearance for an even remotely medium sized air cooler

Water is the way to go

[final flame]

For those fine, but the average atx case air cooling is far better

[cursive summit]

I mean, I just did so it very much works

[sharp badger]

Depends on the CPU with going with Aircooling though?

[cursive summit]

Agreed

[sharp badger]

Sure Aircoolers can handle hot chips like 13900K but AIOs do help with workloads

[cursive summit]

I would never do water myself if not for the circumstances

Person just had a fairly limited budget and really needed a portable build

[final flame]

Yea, especially if you run your PC 24/7

[cursive summit]

ATX is just not portable at all

ITX is rather easy to carry if needed

Also they are cute

Ryzen 1 was just 6 years ago

And 8 core top end chip for less than 1000$ was a really big deal

The fact that 16 cores now is just normal is pretty amazing

Intel would still try charging 1 grand for a 6900K 8c/16t

[final flame]

Yea

My i7 9700k was 400$ when I got it

That's only 8 cores which was quite a bit for the time

In general everything has gotten cheaper for higher performance

[cursive summit]

It's kind of amazing that back in 2017 people were amazed that 8 cores was able to be had for just $329

Just two years later Zen 2 introduced 16 core CPUs

For 749$

Like, the industry is so much better now

[deft marlin]

It has fans that are pre installed, which I assume protrude toward the rear

And encroach on GPU space :p

[gusty kiln]

[true quest]

https://tenor.com/view/yes-simon-cowell-glowing-eyes-gif-16217022

[pale sigil]

oops

[cursive summit]

https://www.extremetech.com/gaming/amd-announces-new-gpu-detective-crash-analysis-tool

Got excited for a tiny second

RDNA2 & 3 cards only

I guess I will never know why my RX 5700 keeps crashing

[hasty thunder]

i can help with it tho

[hasty thunder]

what model?

[cursive summit]

Reference

Tbf that's all I know about it, it's just a reference blower card

[hasty thunder]

yeah... no wonder, the blower reference cards tend to crash due to their own heat, choking them basically

[cursive summit]

It's not overheating

Not according to the temp sensors

[hasty thunder]

then lower the vCore by 0.25V

when exactly do the crashes occur?

[cursive summit]

It's fine, it crashes seldom so I'm okay with it

Pretty sure it's just RDNA1 being annoying and not a temp issue

This card had crashes since day 1, it has only gotten better since then

Driver updates made it a lot better

And I think I only ever had crashes in directx titles

Can't remember any vulkan crashes

[hasty thunder]

k, then it's def a driver issue

[halcyon willow]

nice

just what i needed

[raw zephyr]

another reason to stick with nvidia

[cursive summit]

Nobody cares

[halcyon willow]

least aggressive amd fanboy

[cursive summit]

I'd rather take driver crashes but pay 40% less for the same performance

[raw zephyr]

i'd rather pay 40% more for reliable software and nvenc

[cursive summit]

Plus driver crashes are uniquely RDNA1 problem, 2 and 3 should be fine

[halcyon willow]

💀

[cursive summit]

nvenc isn't as great anymore as you think they are

[raw zephyr]

rofl

[cursive summit]

And AMD software is generally pretty good

[raw zephyr]

rofl

[cursive summit]

Try to disprove either of those statements

[halcyon willow]

DLSS 3, Frame Generation looking at this rn

AMD software is the worst steaming pile of dogshit ive ever used

[raw zephyr]

amf is far behind nvenc, it takes a really fat hit on low bitrate scenarios like streaming

[hasty thunder]

"reliable" xD wtf, i've only had issues with their Game ready bullshit

[cursive summit]

You are years behind on that information https://www.tomshardware.com/news/amd-amf-encoder-quality-boost

Seriously fuck bing

[raw zephyr]

then don't use it? i always used nvcleanstall

[cursive summit]

I hate that shit

I tried to degoogle and bing is absolute trash

[hasty thunder]

uh... AV1 is a thing too, then there's FSR 2.2 as well as AMD's own Frame Gen soon

[raw zephyr]

4 years old eposvox

[cursive summit]

Tbf if you care about things like nvenc, you are probably a creator and should choose Nvidia regardless

AMD is becoming strictly a gamer brand

[halcyon willow]

id rather have a 4090

[hasty thunder]

not to mention that AMD manages to get their shit all together in a single software suite unlike NoVideo. Overclocking? Gotta rely on Afterburner. Wanna record shit? NV Experience

[raw zephyr]

i had the chance to compare latest amf and latest nvenc, i had to go 720p60 8k bitrate to have a watchable twitch stream

[halcyon willow]

and be able to fit cards in SFF

[cursive summit]

What are you talking about

I watch twitch streams all the time on the 5700 no problem?

[raw zephyr]

i mean encoding

[cursive summit]

Ah, so streaming

[halcyon willow]

as in someone watching them...

[hasty thunder]

as said before: AV1, fuckers

[raw zephyr]

av1 is not gonna save you, it's not ready yet

av1 nvenc is gonna come on top most likely, just like it does right now with h264 and hevc

[hasty thunder]

Twitch isn't ready for AV1, that's correct

[raw zephyr]

i didn't really test

and where do most people stream?

[cursive summit]

Kick

[hasty thunder]

still, i don't see any reason to pay a premium of freaking 40% for basically the same performance

[cursive summit]

Actually

https://www.tomshardware.com/news/av1-live-streaming-is-finally-coming-to-youtube

[raw zephyr]

i do

[halcyon willow]

tell me what graphics card rivals the 4090

[cursive summit]

Crossfired 7900XTX's

Basically same price anyways

[halcyon willow]

don't get me wrong, I'm still going to use the 7800X3D

[cursive summit]

4090 is literally not a gamer card

[halcyon willow]

but Nvidia is on another level with their GPUs

[hasty thunder]

fucking christ... The xx90's are basically fucking titans, get that shit in your head already

[halcyon willow]

gotta pay to play

[cursive summit]

4080 is where the GeForce line ends

4090 is yeah, a Titan

Borderline A-series card for workstations

[hasty thunder]

the most logical and reasonable GPU a normal gamer would buy is a xx80 Ti, that's where the line for normal people ends usually. The xx90's are just halo products made for enthusiasts, not gamers

[halcyon willow]

3080

[cursive summit]

Especially with how much tech NV is cramming into the 40 series, those cards are getting hoarded by professionals

AI training, rendering and whatever

NV has to decouple gaming and workstation aspects imo

Resurrect the GTX brand

[hasty thunder]

and i still hate NV for the decision to revive the xx90's as single GPU PCBs

they were always dual GPU PCBs before

[old marlin]

not to mention that experience is the most vile shit

I actively choose to not use it

because it is so shit

[old marlin]

dual gpus suck

and always sucked

[hasty thunder]

The GTX brand's basically gone with Raytracing. They should address the 90 cards as what they actually are. Titan!

[cursive summit]

Tbf NV was supposed to launch a Titan or something this generation, got canned

[old marlin]

titans are workstation cards

[cursive summit]

I wonder if AMD's Titan will come out or not

[cursive summit]

So, 4090s

[old marlin]

well 4090s became worstation cards

nvidia realized there is not much of a reason to buy a titan

if it is just a totl gaming gpu with an extra gig or 2 of vram

[hasty thunder]

i mean look at the Titan XP i.e., technically 1080 Ti's with 12 GB VRAM instead of 11

[old marlin]

so they just combined workstation card + totl gaming gpu + entry level ai card

[old marlin]

you mean the titan Xp?

[cursive summit]

Isn't that what they said

[old marlin]

no

there is an XP and an Xp

[hasty thunder]

oh ffs

[old marlin]

well technically there are 2 titan X cards

one of them is called XP

then they made the titan Xp

[hasty thunder]

i mean the Pascal based one

not the Maxwell

[old marlin]

both of them are pascal

[hasty thunder]

Maxwell sucked

[old marlin]

and iirc both of them are the same chip as well?

can't remember

the XP sucks though

1080ti was a kick in the teeth for people that bought it

since it was faster

[old marlin]

maxwell was good wdym

pascal was just stupid good

especially the 1080 and 1080ti

[hasty thunder]

anyways, the only reason why the Titans were more expensive wasn't the extra VRAM you got, but also significant performance boosts in terms of productivity

[old marlin]

esp the ti was massive overkill in the market

[old marlin]

no it was just a gig of vram

they were the same card otherwise

with maxwell

like 1:1 the same

the 1080tis were faster in productivity since they clocked higher

as long as you didn't need the extra gig

[hasty thunder]

ngl, i've ordered like 5 different 1080 Ti's back then, because they were so damn cheap for that stupidly insane performance they offered

[old marlin]

the titans also are nto as insane as quadros in reliability

they just made 0 sense

and nvidia also didn't really know who they are for

nobody bought them

[hasty thunder]

honestly, the xx90's right now aren't for a specific target group either

[old marlin]

they are for multiple

they are so expensive that it is fine for them to also cover workstation/entry level AI card

[hasty thunder]

that's what the titans were for too

[old marlin]

no

the titans were marketed strictly as non-gaming gpus

[hasty thunder]

despite the fact that everyone knew they were exactly that, but just for those who wanted the extra edge in some cases

even Chat GPT gets to the details and points out that the Titan had more Cuda cores, specifically made for professionals

just like the 3090 and 4090 rn. These are cards for enthusiasts and professionals

[final flame]

does it really matter if it's labled as a damn gaming card or not
while it may not be it's intended use it doesn't matter if it's still more powerful and capable of everything a gaming card is capable
the architecture is the same

[hasty thunder]

For me it definitely does matter, because it pisses me off that NV downgraded the Titans to mere xx90's cards, killing 2 birds with 1 stone by wiping the Titan series out of existence and dual GPUs at the same time

[final flame]

if you were to comparing it to a quadro then it's more understandable but in the end the geforce is the "gaming" cards

[old marlin]

time to game on an A100

[hasty thunder]

and the fact that the xx90 cards give people a wrong picture of things

[halcyon willow]

who the fuck even uses dual gpus though

unless youre doing some serious rendering or something

[final flame]

Titan is just the damn name, why does it matter if they just renamed it to "90"
even when titans still existed people who were gamers still used them

[halcyon willow]

in that case i would think budget wouldnt matter for you lol

[final flame]

if someone already wanted the best they would've gotten a titan anyways
people don't buy the highest end because they want price per dollar, they buy it because they want the absolute best

[old marlin]

at that point you are not using sli tho

at most nvlink

and not the shitty version used in the last gen of sli

[hasty thunder]

Ok, i don't think you get the picture here. The 90's designation was reserved for dual GPU cards. Rebranding their Titan cards as 90's basically removed the illusive Titan brand entirely, killed off the 90's designation for dual GPUs, because NV doesn't seem keen to the idea of dual GPUs anymore (despite the fact that AMD showed them how it's done right), as well as giving the casual consumer a wrong picture of things by introducing a lineup into the gaming market that wasn't initially intended to be there

[cursive summit]

Dual GPUs are hella cool

[old marlin]

dual gpus are shit

[cursive summit]

And effective

[old marlin]

just live with it

not for gaming

[cursive summit]

295X2 was good

[hasty thunder]

uh... it actually depends

there were 2 variants of the 590 iirc

one was basically a dual 580

the other one was a weird thing

[old marlin]

even then

sli sucks

and always sucked

shit scaling

shit frametimes

[hasty thunder]

they also had 2 different methods of utilizing the cores. The weird one rendered the whole stuff on both GPUs at every time while the dual 580 rendered each frame on each GPU seperately

[old marlin]

that was just sli afaik

[cursive summit]

Which 580 are we talking about, RX or GTX

[hasty thunder]

GTX

[hasty thunder]

basically it was SLI on a single card, yep

[cursive summit]

RX 580s in crossfire was quite a popular configuration quite a few years back

[hasty thunder]

people would still stick to SLI and Xfire if developers would actually utilize it

that reminds me of another reason why i hated the 3090 so much

i thought it's a dual GPU, because it had the 90's designation and no SLI/NVlink connector

[cursive summit]

It was sure priced like one

[hasty thunder]

exactly

it made sense to me, because double VRAM, power connector, lack of SLI/NVlink

so i assumed the've revived dual GPUs in a significant new way

[cursive summit]

They discontinued the little NVlink thing after 2000 series

Rather unfortunate

[old marlin]

no like different sli modes

in general

with alternating frames and half frames

[cursive summit]

Although, this exists and I'm all for it https://www.amd.com/en/graphics/workstations-radeon-pro-vega-ii

[old marlin]

with both of them sucking

[hasty thunder]

i remember their announcement of "the next Evolution of SLI" with a NVlink, just to kill it again with the next generation

[old marlin]

sli nvlink was shit anyways

[hasty thunder]

the idea behind it wasn't so bad actually

and in fact you saw some big gains after a few updates

[final flame]

well the idea behind anything dual is good until you realize how difficult it is to get 2 different dies to work with eachother is

bandwith and timings, those are the main limitations

the reason why multi die cpus work is because they're really close together on a physical level

[old marlin]

you mean when they did checkerboard rendering

so you didn't have omega cucked frametimes

[ashen spindle]

Well we will get the real endgame of that in hopefully 2 gpu gens, multi die on same package

At least hopefully since canning big rdna4 is to get rdna5 in time so we gotta hope

[old marlin]

well we hopefully have better rendering division techniches by then

cause even checkerboard isn't ideal

[ashen spindle]

No the idea is you don’t need that at all

[old marlin]

[ashen spindle]

With fast interconnects so for the system it would look land function like one die

[old marlin]

ig?

maybe

we'll see ig

[ashen spindle]

The apple pro chips already do that

[old marlin]

sli and crossfire weren't it at least

atm I doubt we have a need to go back to multi gpu

since the chips are so stupid expensive

[ashen spindle]

I don’t think we ever will

Just that high end will be like mi300 and such, tiled solutions that are as expensive as two old flagships

But at least with performance that makes the price logical

[old marlin]

what

what does mellanox have to do with nvlink

[dull crystal]

Greetings people. I don't want to highjkack the conversation but I have a question and I don't know whereelse to ask. After using a 60% keyboard for 2 years I've moved to a TKL for a couple of months and I noticed how much I use the Function keys (blender + some other softwares that I need that use these extra keys). Not that I can't do the things I do with a 60% but I have some coordination problems (which recently got much worse) so it's easier for me to press one key than two at a time. Is there any chance of a 75% or a TKL in the horizon?

[ashen spindle]

There is something bigger than a 60% coming but nothing known more than a teaser yet

Pins in the general chat have the image

[dull crystal]

Thanks

[old marlin]

aka prob in like a year

[cursive summit]

Few months until campaign I imagine

Delivery prolly around same time as Switch 2

[old marlin]

will be just ready for hl3

[crimson wigeon]

Alongside deckard

[final flame]

yeouch
@gusty kiln

[latent mural]

Worth

[pale sigil]

When the heck is that last tax

[marsh glade]

State, county, and city

[reef patrol]

Murrica' 😄

[latent mural]

Better than paying VAT

[pale sigil]

You ain't wrong

[reef patrol]

https://www.youtube.com/shorts/9CYokP6YIKw

Yeah...no.

[pale sigil]

My first time in America basically

[old marlin]

us tax system is so stupid

[reef patrol]

https://www.youtube.com/watch?v=tXZ2hxoKmic
I love how weird pant is to people. 😄

[vagrant pagoda]

Have you tried not being poor?

[pale sigil]

My soul kept being crushed every time I saw something I wanted.

[vagrant pagoda]

If you're not poor the extra 10% dosent matter

[pale sigil]

Idk chief

[old marlin]

it's not about that

[pale sigil]

I wasn't poor, I was in school with pocket money

[old marlin]

it is about not being transparent

[pale sigil]

When we went on holiday I saw a plushie I liked in my budget but bam tax

Dreams crushed

[solemn remnant]

:(

[vagrant pagoda]

See that's what poor people say

Rich people don't complain about sales tax

[old marlin]

see that's what shitheads say

it's not about the amount

[vagrant pagoda]

See you're missing the point

America is designed to inconvenience poor people as much as possible

And post tax helps with that

[pale sigil]

Try the UK

Even Aldi isn't our savour

[vagrant pagoda]

Because we also do a really poor job at teaching math to low income people

So not only can they not correctly calculate the tax in their head they also don't have enough money to cover it if they fuck up

[latent mural]

Broke? Just earn more money

Simple

[vagrant pagoda]

Perfect double whammy

[pale sigil]

I need

[wheat plume]

The Kill Myself Server

[old marlin]

Meme material thumbnail

[cursive summit]

You have made that joke before you know

[fair gull]

not that its really funny imo

[cursive summit]

It's not

Just leave the term alone, it's a key management service

[cursive summit]

https://videocardz.com/newz/asrock-arc-a770-gpu-with-16gb-vram-drops-to-299-8gb-model-at-249

Really not sure if Intel is worth it

Performance seems to be getting better and better

But the non-AAA title performance is a big question

[astral haven]

🤷 its pretty solidly improving insanely fast

[ashen spindle]

it's still the most cost effective way to get N6 silicon

[astral haven]

on the other hand amd is about to massively improve in reliability
https://www.techpowerup.com/312562/amd-releases-radeon-gpu-detective-rgd-1-0

https://www.techpowerup.com/312564/intel-graphics-announces-directx-11-performance-uplifts-and-frame-time-reductions
but yea intel is working hard on fixing dx11

[astral haven]

nvidia has had their tool since 2018, which cant have hurt their reliability...

[ashen spindle]

yeah if the price range is what you look for the intel options are an option now

even if/when they were to stop client gpus again they also wont die so fast that you cant get warranty claims so it's safe there lel

[cursive summit]

https://www.techspot.com/news/99853-ibm-40-workers-have-learn-new-skills-over.html

https://twitter.com/TechEpiphany/status/1693298485917864362

TechEpiphany (@TechEpiphany)

🔥 CPU Retail Sales Week 33 2023 (mf)

ℹ️ Sales:
#AMD: 3460 units sold, 76.46%, ASP: 282
#Intel: 1065, 23.54%, ASP: 313

ℹ️ Revenue:
AMD revenue: 974319, 74.52%
Intel: 333063, 25.48%

ℹ️ By socket:
AM4: 2040
AM5: 1130
1700: 980
1200: 20
sWRX8: 10
1151: 10

CPU sales…

Twitter · Aug 20, 2023, 4:26 PM

Wdym phone

That's a PS5

It's alright, we've all been there

[pale sigil]

Y'all being sarcastic?

[cursive summit]

What, no?

Why would we be

Sarcastic is being silly

I am not a silly bitch

I might be a bitch, I will NOT tolerate being called silly

[pale sigil]

It's an Xbox one s

[cursive summit]

Ahhhh

It's one of those Apple iPad models

I never remember any of them, their model numbers are hella random

Sorry to disappoint you miss, I will not let you down again

[pale sigil]

I liked the one s design

And series s

Eh I still prefer it over the PS5 design tbh

@cursive summit any events happening today?

[cursive summit]

Your sorry ass appeared for the weekly "Pleasant visits his server" event

[pale sigil]

@dawn epoch didn't you have roles

[cursive summit]

Other than that, no

[steep creek]

i had this when i was younger, it was due to my allergy. gone now though 😁

though to be fair i wasnt very consistent with the medications

[astral haven]

If you are adult, you previously were adolescent and young before that
and allergies arent ageist afaik

[steep creek]

not sure exactly the correct term, i guess adult. im 21 now. but my eyes dont itch anymorw which im very thankful for haha. just mild cold like symptoms which the medicine is very effective against

[gusty kiln]

Nice, when is it coming

[final flame]

Their lead time is 1-2 weeks for customs

The actual connectors I want from aliexpress take about 2 weeks to arrive, so looking at probably 3 weeks

[gusty kiln]

not too bad

[final flame]

This cable is gonna be real nice once I get it

Too bad I don't think cables change the sound

[gusty kiln]

kek

https://www.youtube.com/watch?v=QWbyF1fMGwY

probably the only time a cable made a difference

[final flame]

Yea but not a appreciable difference

But if were going on placebo, I prefer the sound from the Hart audio cable I already have compared to the stock cable

Sounds more "detailed" although I don't know how much of that is true

Especially as its hard to ab test without a proper setup

[astral haven]

measure the resistance and capacitance, lower is better

my chinesium cables are like .1 ohms which is pretty darn high and can make a difference with 10 ohm iems (in theory)

[final flame]

Resistance is known to change the bass frequencies, same reason why tube amps have warm sound they have a high resistance output

[gusty kiln]

https://www.reddit.com/r/headphones/comments/15x3zle/canjam_london_2023_heddphone_refresh_is_sleek_and/

A new contender

[final flame]

Looks like if the drop panda and a susvara had a child

[gusty kiln]

crinacle liked the first heddphone

hopefully similar

[astral haven]

i would like someone to make something with the ethos of the drop panda but none of the implementation issues

[gusty kiln]

the first heddphone was 700+ grams

[final flame]

Damn

[gusty kiln]

wonder how much lighter this one is

[cursive summit]

https://youtu.be/mGinUBppUhc

Hm, interesting

Having been planning an SFF PC this would have been actually useful

Scythe not as good as I expected

Noctua also lacking

Thermalright though seems to carry their weight, considering those coolers are rather cheap

Then again testing could be flawed, it's not Gamers Nexus

[astral haven]

you say that as if gamers nexus has flawless testing...

[cursive summit]

What's up

Why am I being mentioned

Ah..

I was wondering why I have not heard of this before

It's an AIO

[astral haven]

if you have blue kitkats you really shouldnt eat them at all...
unless ofc you are in japan

[cursive summit]

.... why is there an AIO in low profile cooler comparison

[astral haven]

cause technically its low profile

[cursive summit]

nh-u12a is not a low profile cooler either by any means

[astral haven]

its a good point of comparison

techpowerup uses it for their standard cpu temp tests

[cursive summit]

Seems that Noctua nh-l12s is the first of the actual low profile coolers

And Thermalright is just behind it

That thermalright looks really good

Around 40$

I'll have to remember that, seems like a solid package

https://wccftech.com/msi-boosts-gaming-performance-by-up-to-12-with-new-high-efficiency-memory-mode-on-am5-motherboards/

Tf is this

Memory try it?

Is that a challenge

[astral haven]

auto overclocks

[steep creek]

lol

strange naming decision for it

[astral haven]

Asus calls it MemOK and MemTweakIt

[steep creek]

to me mem tweak it sounds the best, but personally would prefer something more direct like memory overclock

[hasty thunder]

interesting... so, i was watching a germany YTer reviewing mice to the utmost details, including pads and such

since i've experienced an issue with my performance recently i decided to remove my pad from the hotel table

turns out it was the entire issue of my performance dip

so yeah, do not buy the QCK by Steelseries

[gusty kiln]

qck hasn't been in any conversation for like the past 5 years

maybe except super boomer CS players

[ashen spindle]

hey at least I was able to just use it when I got it years ago

[crisp granite]

Nah

[ashen spindle]

this new one I had to air out for weeks

[crisp granite]

They’re

like matcha itself tbh

Quite the acquired taste

[ashen spindle]

think I dosed this zowie pad in vinegar, baking soda, hot water, cold water and all before I could put it in my room lmfao

[gusty kiln]

lol wtf

this locus pad I got didn't smell at all out of the box

[gusty kiln]

which zowie?

[ashen spindle]

zowie gsr I think

[hasty thunder]

maybe it did and you didn't notice, because you have the same smell?

[gusty kiln]

time for GSR II

https://zowie.benq.com/en-us/mouse-pad/g-sr-ii.html

[steep creek]

lol i just had a matcha about 3 hours ago, very nice for study sessions

[ashen spindle]

I mean its fine now, and I guess its the G SR SE since its blue with the pattern

[gusty kiln]

maybe it smelled if I put my nose right up against it, but it didn't fill the room with it

ah

the red one is more durable apparently for when you upgrade

GSR-SE Rouge

[ashen spindle]

if I need a new one I will never get this piece of shit again lol

[gusty kiln]

kek

[ashen spindle]

im not spending another week on getting smell out of it before using it

[fickle ore]

I made some graphs nobody cares about 😢

[gusty kiln]

the B650E Taichi Lite should be a best seller when it comes out

[astral haven]

my question is did they actually get better, or just did the others trip over having to deal with AMD seriously?

ooh speculation of 3x-10x optical zoom on s24 ultra
that would be hella hype

https://9to5google.com/2022/12/27/lg-optical-zoom-sensor/
was very dissapointed when samsung didnt go for something like this

[pale sigil]

that was fast

[thick elk]

I thought that would be the case but I am playing dying light 2 now and before that horizon zero dawn.

My first game was basically the meme that is vampire survivor

true, still most of the time the battery life is sufficient for commute to work + commute back + lucnh break

[halcyon moss]

Hah, tuya? Known for selling whitelabel smart home stuff

[marsh glade]

It's deep into the summer doldrums, I'm not really expecting much excitement at all

Though, sure does look like NVDA wants to attack new highs and earnings is perfect cover to do so

Consolidate tomorrow, then on Wednesday morning's manufacturing PMI release, drop below Tuesday's low, turn around at 10:10am and work its way back up to 480

Then after the bell gap up and rip off the faces of anyone short 📈

[fair gull]

ive seena few of their PCs

and how they come etc

id personally say if you have money to blow to pay someone else to build a pc for you

go ahead

[hoary tendon]

wtf

what is he using them all for?

[fair gull]

almost

they dont ship those things in wooden crates i think

but close to wooden crates

but yes super clean cable management

super clean builds

installs are also very clean if you select preinstalled os

and the accessory box is also nicely layed out etc.

and if i remember correctly they had livestreams of your pc being build

idk if they still do but a customer of my old job showed me and it was dope af

[cursive summit]

Ooo

https://www.techpowerup.com/312631/nvidia-bios-signature-lock-broken-vbios-modding-and-crossflash-enabled-by-groundbreaking-new-tools

This opens up a lot of possibilities

https://www.overclock.net/threads/nvflashk-flash-any-bios-to-nvidia-gpus-safe-board-id-bypass-up-to-4xxx-series-including-founders-edition-cards.1807438/

[astral haven]

sweet!

only slightly less sweet than the first time it was posted ofc @cursive summit

[cursive summit]

https://www.amd.com/en/support/kb/release-notes/rn-rad-win-23-8-1

• Improvements to high idle power when using select multi-display setups with mixed high-resolution and high refresh rate displays on Radeon™ RX 7000 series GPUs with variable refresh rate enabled. Further optimizations are being investigated to improve idle power on additional display configurations.

Maybe it'll be better now

A lot of peeps avoid RX 7000 just cause of that high power consumption bug

[halcyon willow]

as they should

[astral haven]

how many times have they said that?
also the problem wasnt just with vrr

[cursive summit]

Ah come on hateboy

It's not like it's a massive bug

Annoying one, but it's fine

• Application crash or driver timeout may be observed while playing Baldur's Gate 3 with DirectX® 11 API set on certain systems with Radeon™ RX 7000 series GPUs.

How about RX 5700 too you bastards

I crashed at least 5 times with driver timeout

Although tbf it's prolly just RDNA1 being RDNA1

https://videocardz.com/newz/intels-latest-gpu-drivers-can-overclock-arc-a380-gpus-by-150-mhz-through-a-bios-update

[astral haven]

👀
and every gpu architecture amd has made before and since

[cursive summit]

Lol what's up with the BIOS updates

[astral haven]

welcome to microcode

[cursive summit]

Oo they have nearly done it

Someone's results with new drivers

33W multi-display idle

[astral haven]

be aware its a bug that isnt guaranteed

heck having the control panel open could force the gpu into a lower power state

[ashen spindle]

tbf rdna3.5/4 would be the first that really started where they had actual money and not just sliding near the 1$ stock evaluation

[astral haven]

Revenue vs valuation fam

[ashen spindle]

that doesnt change that they at the point of making those they were severly smaller than now

[astral haven]

They were still larger than IBM
Who makes a heck of a lot larger and less buggy chips

At least if you go by valuation

[ashen spindle]

just easier to say what timeframe I meant by saying the time when the stock was in the basement

[astral haven]

And my point is valuation doesn't matter to make bugless chips

[ashen spindle]

yes I know but at the time of that evaluation they had less money as well is all I'm trying to point out xD

[fickle ore]

this should correlate to better software, but somehow not

to be fair, 2022 looks like the first year they got a lot of ppl

[gusty kiln]

mostly from Xilinx

[ashen spindle]

^

[fickle ore]

stock goes 100x but ppl increase by snail count

[gusty kiln]

gunna be a while until the new hires actually makes a difference

[fickle ore]

well tbf they got enough ppl atm if we compare to intel and intc revenue

yea

[gusty kiln]

nvidia been building their software team forever

[fickle ore]

yea, nvidia aint got that much more employees

would be funny to nv software team vs radeon team kek

[gusty kiln]

also AMD is split between Ryzen, Radeon, and Xilinx

Nvidia is all on GPUs of different types

and ARM

[fickle ore]

yeah, hardly a contest

[astral haven]

And Nvidia has a CPU team too now
Tho they are using neoverse cores so it's not them designing them

[gusty kiln]

ripped off the built-in grips on my RVU, feels a lot better

[cursive summit]

What's an RVU

[halcyon willow]

razer viper ultimate

[ashen spindle]

you can't make people above junior level appear out of thin air if you have more money, its not like there is a stash of senior driver devs waiting somewhere

[cursive summit]

I literally have that mouse

The grips are fine

[fickle ore]

you just put juniors on the driver stack and let them go at it

free jackets?

[cursive summit]

Wansn't Google also a top employer

Well until the new shite CEO appeared

Why didn't you ping me

I love AI

That is actually really interesting

[latent mural]

Average Razer fan

[cursive summit]

Will keep this one

[cursive summit]

Not a fan

Razer has trash software

I just speak from experience

[latent mural]

G Hub and Nvidia software > Synapse and Reflex

[cursive summit]

G hub is just as bloated trash

[latent mural]

Lesser of two evils

[cursive summit]

I don't install any of these control hubs

All spyware

[latent mural]

Heavily agree

AMD software is malware too

All my bros love not having to download software to use something

[cursive summit]

You should think before you speak

[latent mural]

I can say the same for you

[cursive summit]

Ah yeah?

[latent mural]

Yep

[cursive summit]

Give your sources then

[steep creek]

Ahah thank you for sharing. It seems fairly proficient in verilog code as well, which I could not say about bing chat. Impressive stuff, and seems unbiased from my small testing

[latent mural]

Source: trust me bro

[cursive summit]

I want to see sources for "amd software is malware" argument

[latent mural]

Jokes aside, I just don’t like AMD software

I could never like that skinny fuck

[cursive summit]

You stole that one from me

[latent mural]

I don’t ever need to record something longer than 10 minutes so GeForce Experience is good enough for me

I like the filters they have as well for select games

[steep creek]

I also especially like that you don’t have to create an account to use wooting software. Just find it pointless of companies to require

[cursive summit]

How else will they siphon your data

And personalize your ads

They have to sell it somehow

[steep creek]

Fair point lol. I guess facebook and google has their ways of still doing it

[cursive summit]

Slowpoke

Snorlax?

[latent mural]

Slowking

[cursive summit]

Slaking

Participate in white collar hacker competitions

[steep creek]

Bing chat provided very strange results when I asked it about verilog. Sometimes it would answer, but provide very unoptimized and long code, other times it would answer something unrelated lol

[cursive summit]

Slaking issa pokemon

[latent mural]

💀💀

[steep creek]

Perplexity seems more straight forward but I can’t say for now as I’ve only tried it briefly

[cursive summit]

Slaking is better

[halcyon willow]

Slowking better

[cursive summit]

You are just derpy that's all

In a good way

[halcyon willow]

unspoken rizz

[cursive summit]

I don't rizzoli

[halcyon willow]

vex cooking up a nasty rizzoto

[cursive summit]

Nasty? Fuck off

I make a banging one

[halcyon willow]

nasty in a good way kid

[cursive summit]

Shut up toddler

[steep creek]

It’s honestly impressive. It’s very accurate for the specific things I’ve asked it. Results are similar to the ones I put in my lab report

[cursive summit]

Yeah that site is great

[halcyon willow]

thats so mean

[steep creek]

Yep I’ll bookmark this one

[cursive summit]

How do they handle translations?

[steep creek]

Or install their app rather

[cursive summit]

For non-english speakers

[steep creek]

It seems to know Norwegian pretty well lol

[cursive summit]

Also how does it handle shopping

[steep creek]

The wording it used it correct

[cursive summit]

Is it a full or just partial search engine replacement?

Oh nah I'm perfectly fine with english

[steep creek]

It doesn’t have my language in the list, danish is the closest one but seems to understand me regardless

[cursive summit]

I would ask something like "what's the cheapest local shop price for X"

And I dunno if it supports my country

[steep creek]

Yeah it will be interesting to see if it factors in local shops or not

[cursive summit]

Prolly not

[steep creek]

Yah it tries to but I’m gonna say no lol. Just refers me to the price match site

[cursive summit]

Unfortunate

So it basically is just another chatgpt/bing

A lot faster

Bing isn't slow though

Chatgpt is snail pace sadly

[steep creek]

And somewhat outdated as well

[astral haven]

razer software is malware
autoinstalling itself

yall talking about perplexity?
perplexity is chatgpt under the hood just using better prompting

and im pretty sure chatgpt is faster if you pay for it, free version is slow tho

oh wait no its currently only on 3.5, so if you pay for it chatgpt is alot more capable as well as faster

bing chat is 3.5 tho

works shockingly well

🤷 yea this part is definetly subjective

im less scared of microsoft than the EA folks tho

the other problem is that gpt-3.5 and 4 are proven to be shoggoth models
with bard theres at least no proof it is a shoggoth

[vagrant tartan]

What's a shoggoth model?

[astral haven]

a model that can assumes identities and those identities can slip and reveal different identities

[vagrant tartan]

What makes one model capable of that but another incapable?

AI is interesting but from my current knowledge level on the subject it seems like working on it is like working on something inside a box that lets no light out

Oh like the feedback thingy where you have people rate the models responses? And then you factor that into the training

I guess there's way more stuff though

Huh interesting

It's sad to me that it requires such immense computing power to train

Oh the main processes in order to prevent this type of behavior is on the output side?

I'd have thought intuitively that the training input is a better place to try and sanitize

I guess it's an insane amount of training data to sanitize too, so that in itself is a large task

Is modifying just training it in different ways?

Oh wow that's clever

Evolutionary is pretty easy iirc

I mean yeah it's kind of suited for certain types of tasks

[astral haven]

that hasnt been proven yet btw
we currently have no way of proving if it is showing us a mask or is the underlying reality

[vagrant tartan]

Easily rate-able

[astral haven]

the current models are trained purely on predicting the next token in the dataset, then once the training is done, they perform rlhf

https://en.wikipedia.org/wiki/Reinforcement_learning_from_human_feedback

and rlhf's wikipedia article is half what it does and half issues with making it work

[vagrant tartan]

are the models dividable and then mergable?

[astral haven]

nope

we can understand small parts of the models, but take out more than a few% of the part we dont understand and they break for mysterious reasons

also this is only for gpt-2, there hasnt been significant research done on gpt-3 or higher yet

[vagrant tartan]

For some reason i really wanna write some deeplearning thing on top of a mesh network now

[astral haven]

connor leahy would like to have a word with you

his idea is a bunch of small llms networked together (virtually but physically is close enough)

[vagrant tartan]

It's silly though because i don't have either input or output

[vagrant tartan]

This is kinda cool

[astral haven]

his idea is more that you would specialize each llm on a specific task, and you could implement classical algorithms (like a calculator) onto a decision tree

[vagrant tartan]

I was just gonna write that lmao

[astral haven]

ye

[vagrant tartan]

Having specialized models makes a lot of sense

[astral haven]

and unlike the huge inscrutable monoliths, you could at least have some visibility onto the decision structure

[vagrant tartan]

I guess i might mess around with deep learning once I'm done with my current project

Seems pretty fun

[astral haven]

llama 2 just got releasedish

its available on huggingface

[vagrant tartan]

No clue what either of those are but I'd wanna write my own implementation of everything for fun i think

Should be able to get it working semi okay, though very unoptimized

I know i would have to do gpu compute eventually tho which seems annoying

[astral haven]

As long as you have an Nvidia GPU it just works and if you have a Radeon or intel GPU it usually works

[astral haven]

The problem is that it takes tens of millions of dollars worth of computer to train these models

Sam Altman (ceo) claims gpt4 cost $100 million to train

The way that people use these foundation models at home is they fine-tune which you can do on your own yes but is a lot less than tens of millions of dollars or you can do prompt engineering which costs literally nothing just your time and testing

[astral haven]

https://www.youtube.com/watch?v=UTUnVEiO-TU

[fading cloud]

Probably yes

The more you wanna spend the faster you get the model

And gpt-4 was relatively quick released

But training is essentially just vram depended

And vram is expensive

Not to forget the storage you need for such a model

[astral haven]

it took most of a million gpu hours to train gpt-4

on a single gpu thats a couple thousand years

[fading cloud]

Yeah

But simpler things can be done on a consumer GPU

I did train a voice model

[astral haven]

capabilities scale logarithmically with gpu hours

[fading cloud]

Was okay after 5 hours of training, but I restarted my PC and somehow the model sound wrong and glitchy

Not quite sure what broke

[astral haven]

didnt save the progress correctly?

[fading cloud]

If my PC was quieter I would run such things over the night

[fading cloud]

It did save

[astral haven]

was that voice model foundation or fine tuned?

[fading cloud]

But both weights broke

Or the software to use the model broke somehow

Which is more likely

Or windows did a silent NVIDIA driver update

Trained on 3080 10gb

I mean I should honestly try to get a more better voice model clone

I was both amazed/ inpressed on how fast it went and 'scared' what theses tools can do in the wrong hand

[astral haven]

were you fine tuning or did you build from scratch?

[fading cloud]

Like essentially, you can copy every voice as you please. As long you find the person talking

[fading cloud]

Mostly automated

[astral haven]

that doesnt answer the question lel

[fading cloud]

So nothing really fine tuned

Scratch?

To what extent

I got the training data myself, that's about it

[astral haven]

aka was there any blobs in the inputs

[fading cloud]

Blobs?

Scuffed training data?

[astral haven]

like did you start with a generic model

or was it all human legible code

[fading cloud]

I am not quite sure by that.

I mean I fed it the training data, and let it calculate it

But I doubt the weights (aka model) can be read by a human

https://github.com/w-okada/voice-changer/blob/master/README_en.md software to use the model

https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/blob/main/docs/README.en.md program it was trained on

The training data was approx 1:30h long without any major pauses

[astral haven]

"RVC requires other pre-models to infer and train."

[fading cloud]

Oh ok

[astral haven]

did you download something from huggingface?

[fading cloud]

Uhh

I

Uh

[astral haven]

if you went to huggingface its definetly finetuning

[fading cloud]

Believe so? But only to download the program, if I am correct

And opon installing it , it downloaded stuff

But yes most likely it was training with a generic model

I am not that deep into ai training

I currently barely understand how to use the trainer

[astral haven]

tbh it really doesnt matter but its more about total compute time to train from scratch vs fine tune

[fading cloud]

Yeah

It depends on what you wanna do

Recreating Pictures is much harder to train for I'd imagine

[astral haven]

the point is more that this isnt quite correct
you fine tuned a pretrained voice model as far as I can tell

this would be like gpt-4 to chatgpt
very useful yes but not the part that cost 100 million dollars

[fading cloud]

Anyway it is 4 am and I kinda wanna sleep 🙃

[astral haven]

somebody probably paid 10 grand in compute to make the foundation model that you used maybe

you could probably train it yourself given the data and a month

[fading cloud]

I mean the model, did sounded like the guy I copied

After I trained it with my data

[astral haven]

ye

and before it probably sounded either hella generic or didnt work at all

[fading cloud]

That would explain why it was so relatively quick

[fading cloud]

No idea

I wouldn't know where it was

If it would be possible even

The next few years will interesting

With ai etc

[fading cloud]

But I highly doubt it is humanities downfall like many people claim

It will get better yes

But how much, no body can say for sure

[astral haven]

i suggest you listen to this tomorrow
https://www.youtube.com/watch?v=pMoVsM1EWR0

[fading cloud]

But computing will be the limit

I mean you need a good GPU to run must models

[astral haven]

I cant tell you how exactly stockfish will completely and utterly prevent you from winning a game of chess, but I can tell you the outcome

[fading cloud]

That isn't exactly great on a battery powered robot

[astral haven]

yes but people train and test the largest model on cloud infrastructure that we cant turn off for various reasons

[fading cloud]

I mean you still need a GPU to run it

As of now

[astral haven]

we wont get killed by a small ai
we will get killed by a lol parameters go brr ai

[fading cloud]

Hmmm

I doubt that one

[astral haven]

listen to the podcast when you are awake

[fading cloud]

Will do

Oui 1:4h

[astral haven]

leahy doesnt have as clear and precise examples of how deadly ai can be as yudkowsky, but the dude is alot more down to earth

[fading cloud]

That's a bit long

[astral haven]

https://www.youtube.com/watch?v=Yd0yQ9yxSYY
heres the short version
its hard to believe, but its also not been successfully argued against, the long version goes over all the counterarguments

[fading cloud]

I will listen to it, for how long no idea . I am not a prodcasz listener

[astral haven]

you can start with this but its a heck of alot more bleak

[fading cloud]

Thing is yes in theory. But would it? I doubt it

It is just to inefficient to run on a large scale

[astral haven]

he has formal logical proof for it

[fading cloud]

Ye ye I will listen to it

But you heard my hot take on it

[astral haven]

ye and its definetly not a hot take
its very much the popular take

[fading cloud]

Oh

Ok

[astral haven]

the problem is that its also disproven

[fading cloud]

How so

[astral haven]

for that you need to listen to the hour long version

[fading cloud]

I guess it is argued in the vids

[astral haven]

yup

ive only spend 2 weeks searching for counterarguments, i havent researched well enough to actually make the arguments myself

[fading cloud]

Also, there isn't a "the" ai model

[astral haven]

at the moment there is

[fading cloud]

Many in fact all doing different stuff

[astral haven]

yes and those arent a problem
the problem is the 1 model doing everything without us having an understanding of why it does all the things

[fading cloud]

I can't imagine the compute time for "the" all in one model

[astral haven]

100 million dollars or 2-3k gpu hours

[fading cloud]

For a chat bot

[astral haven]

no

chatgpt is just finetuning

[fading cloud]

No?

[astral haven]

im talking about gpt-4

[fading cloud]

gpt -4 is basically a fancy chat bot

[astral haven]

humans are fancy chatbots

[fading cloud]

With more variables

[astral haven]

it doesnt matter how it does all the things
it does many things
we dont know how

we dont know why

[fading cloud]

Sure?

[astral haven]

it would be fine if we understood how or why or could predict what it would do

[fading cloud]

It tries to find patterns in the data, it connects those

Hence it is run on a GPU which is designed to compute matrixes

[astral haven]

if you can prove that statement, you can probably win a turing award

[fading cloud]

I pretty sure that's how it works

Or boils to

[misty saffron]

bro what

[fading cloud]

Or I am extremely misinformed

That can it also be haha

I d like to be proven wrong

That's how one can learn

[misty saffron]

it runs on gpus because those can do maths and move data around very fast, in simple terms

[fading cloud]

Also correct

[astral haven]

thats what it looks like
but we dont actually understand
if we take gpt-2 (not 3 or 4 we cant even get a basic understanding of how they work)
if we remove the parts of gpt-2 that seem to be doing random stuff unrelated to pattern matching, it completely falls apart and fails to match patterns

again the problem isnt the ais tho

[fading cloud]

It's how we use them

[astral haven]

no

its that we use them

[fading cloud]

No?

Uhhh

Interesting point of view

[astral haven]

if the only thing that gpt-4 was used for is to write content for the internet, it would actually be fine (to some extent)

the problem is that we take this thing that predicts text on the internet, and stuff it into a loop and ask it to problem solve
and it does that

[fading cloud]

Because it got that data along side it? No ?

[astral haven]

gpt-3 releases
15 minutes later
https://github.com/Significant-Gravitas/Auto-GPT

[astral haven]

data > something > gradient decent > something > it can do things is the extent of our understanding

[fading cloud]

Hmm

[astral haven]

https://youtu.be/jW2ihBRzLxc?t=276
heres the shortest possible explanation of the issue that i can find
timestamped, about 2 minutes

the what?

[mild dome]

I believe we are still a long way away from making a TRUE AI, but getting a program like Chat-GPT to run society would still be insanely dangerous since its not truelly aware of its actions or the data fed into it.

[astral haven]

the problem is belief is all we have

all the plans in the world dont save you against something that can predict your actions

as if microsoft would let him do that to azure

[mild dome]

Really depends on how a true AI acts doesnt it? Since it would be self aware it might just decide it doesn't like existing and kills itself

If we make a true AI its probably going be be using something like neurons first. Or maybe some kind of artificial neuron (a neural net isn't quite advanced enough)

[astral haven]

the true ai that kills itself isnt financially viable and we will keep trying until we make one that can do things
and if you can prove the second statement you can go collect a turing award

[mild dome]

If you can prove the opposite go collect YOUR turing award

[astral haven]

unfortunately people have already collected awards for that

lel

[mild dome]

Prove it

Like bro just yelling "go write a 100,000 word study and collect your nobel in 10 years" doesn't mean jack shit in an internet argument. Most of us here know like the bare essentials

And as far as I know you don't know anything more then I do until you provide a source to back up your argument

[astral haven]

https://arxiv.org/abs/2105.14111

[mild dome]

Well reading through this, it just seems to be about errors in machine learning algorithms

Not on true intelligence

Like I don't know if you missed my point. I was talking about a truelly conscience self aware living AI

[astral haven]

🤷 thats all human centric rationalization

[mild dome]

This is why I fucking hate that machine learning algorithms were called AI. It's a marketing tactic

It aint

Read up on the "chinese room" thought experiment

It helps provide insight on a very long list of instructions vs understanding

[astral haven]

its also completely misguided anthropocentric bs

[mild dome]

It fucking ain't dude

[astral haven]

ok so you misunderstand the problem then

[mild dome]

You're clutching at straws trying to make these algorithms smarter then they really are

I mean they are super good, but they aren't intelligent

[astral haven]

define intelligence

ill bet that your definition isnt provable

[mild dome]

That's because consciousness is hard to define and I am not an expert. However your definition of intelligence is so loose that a speakeasy counts as intelligent

But here I go
They don't recognise their own actions or thoughts, they don't rationalise, they are designed around purely seeing a pattern and reacting to it as efficiently as possible which to them takes billions of tries. An AI desig ed to paint for instance cannot consciously develop a new style at this point in time. Anything it makes it makes unknowingly

In general we don't recognise animals as intelligent as they don't hit these marks either.

[astral haven]

crows, dolphins, elephants, and chimps all fit that definition

as does autogpt or gpt-4

if you literally just loop gpt, it generates all sorts of wacky shit that noone has ever seen before

[mild dome]

It really doesn't
And it's quite clear I can't convince you

[astral haven]

but heres the actual problem

your definition doesnt matter
once it can perform actions that humanity as a whole are unable to counter, we all die

in all likelyhood, what we get killed by, not a single human would define as intelligent

[mild dome]

That we can agree on

We are going to be killed by a machine learning program designed to efficiently make cans ☠️

Every atom in the solar system will be used to make a can

[astral haven]

that would be one of the better outcomes tbh

the problem is we have no way of directing an ai to make cans
it would likely figure out that theres a specific shape of metal atoms a few nanometers across is what it likes the most and the solar system gets converted into that

[mild dome]

Wasn't there some thought experiment where knowing you are in an AIs simulation leads to it torturing you?

[astral haven]

🤷 does that matter?

if we can build an ai coherent enough to make a simulation, that would actually be pretty fantastic

[astral haven]

This is the problem

and this is proven

https://youtu.be/Ez9uVSKLYUI

We can't even work together as humans properly, why do we expect to be able to build a new intelligence from scratch, and have it coordinate with us?

[astral haven]

holy mackerel thats an incredible amount of time
200 a100 years

[reef patrol]

https://www.youtube.com/watch?v=W3lSjuLAAVQ

Texan weather is scary 😄

[magic rampart]

tfw u get brain eating amoeba and it starves

[severe phoenix]

https://videocardz.com/newz/nvidia-announces-dlss-3-5-with-ray-reconstruction-launches-this-fall

[cursive summit]

Another one

You know, NV keeps making a compelling argument for why I should buy one of their GPUs

... or not

600$ 6800XT goes zoom

[halcyon moss]

$ or € ?

[cursive summit]

€ cause I can't be bothered to go into google to search for the euro sign every single time I talk about prices

The prices on 4070s are horrendous, and you get less VRAM

[halcyon moss]

just press right alt and 5 ?

if you configured your layout to handle that

[cursive summit]

Yeah I don't ever use the right alt

And I have no clue why I should

Uncomfortable

[halcyon moss]

to get €€€€€€

[cursive summit]

I'd still rather google every time

[halcyon moss]

you are a weird one

[cursive summit]

Mate

Right alt, windows key, control and fn are completely useless buttons

They are uncomfortable to press and I don't need them

[crimson wigeon]

Wait what isn't € alt gr + e

[cursive summit]

It's right alt + 4

€

And I would still rather google the sign every time rather than press right alt

I can't even find the source of DLSS 3.5 info

It's not on youtube

Is it gamescom announcements?

Nvidia always presents a compelling argument why buy Nvidia

But man are their prices horrible

[halcyon moss]

depends on the OS keyboard config, for me its right alt and 5

[crimson wigeon]

Alt gr + 5 is ‰

As it should be!

[cursive summit]

On my woot the euro sign is literally on 4

Idk what the layout is on this one

It has the big enter though as it should

[halcyon moss]

again.. depends on the config

[cursive summit]

Option?

[halcyon moss]

cute.

[cursive summit]

Oh yeah that was a part of Mac that confused the fuck out of me

They had a key for the same position for where winkey is

And I kept pressing it for start menu and nothing was happening

Pissed me off

[halcyon moss]

cmd space

minor change tbh

[cursive summit]

It was this damn key I think

I kept pressing it, nothing was happening

[halcyon moss]

its not supposed to in osx

it is a modifier

[cursive summit]

So does macos not have an equivalent of winkey?

Win10 has had that for a long time

[crimson wigeon]

No, winkey + c tries to get you to use teams lul

[cursive summit]

Winkey + V

It's called clipboard

S8U has this shittiest ever autocorrect when it autocorrects normal words into hell knows what

[halcyon moss]

the history paste is nice though

[cursive summit]

Really thinking of turning off autocorrect but I can't type for shit on tablet so I better keep it on

[halcyon moss]

change keyboards?

[cursive summit]

S8U

Tablet

[halcyon moss]

yes

change keyboards

[cursive summit]

What do you mean by that

Install a non-samsung one?

[crimson wigeon]

It's Android you can install different keyboards

[halcyon moss]

use swiftkey or gboard or anything else

[cursive summit]

I think there's a feature for that, never used it

[crimson wigeon]

Well you can do that on iOS nowadays too apparently

[halcyon moss]

its android, of course there is

[cursive summit]

Imagine they make it work on all GPUs

That'd be the true hype

But nah, DLSS brings in sales

[ashen spindle]

Ah nice I saw Ian mention when he was at like 20k words so that should be some good content for later

[cursive summit]

I suppose this is about Ian's take on whole LTT situation?

Saw the video this morning

A lot of peeps are rather unhappy with how Ian took to the whole situation

[ashen spindle]

I would imagine people are unhappy with Ian since he isn’t a fan of either side massively so fanboys from both sides hate it

[astral haven]

people did buy geforce for shadowplay and still do despite nvidia doing their best to gut it

[astral haven]

nailed it lel

[cursive summit]

I saw it here https://www.reddit.com/r/hardware/comments/15xr80u/techtechpotato_the_problem_with_tech_media_ego/

[ashen spindle]

Which works worse for me than radeons replay ironically

[cursive summit]

I haven't watched the video yet tbf

But good context beforehand to keep critical thinking up

[ashen spindle]

If you need context before watching something I think you are lacking critical thinking tbf

[astral haven]

now that one surprises me
given i dont have a single gpu compatible with even remotely current drivers, but thats quite impressive to hear
shame the quality is so low tho

[cursive summit]

https://arstechnica.com/gadgets/2023/08/windows-11-has-made-the-clean-windows-install-an-oxymoron/

[cursive summit]

Wasn't really the point, but alright

[ashen spindle]

Amd worked fine, shadowplay gives me constant stutters every like minute

No matter what setting I try

[astral haven]

interesting

were you recording to the drive you launch games from?

[ashen spindle]

Uhh would need to check but I think both are on C yes

[raw zephyr]

shadowplay is so limiting

[ashen spindle]

I just use it to save clips when I need a replay during raids or such

[cursive summit]

Has shadowplay been at all updated since 2016 or so or is it still riding on its past glory?

[astral haven]

its actually challenging to tune obs to the same quality of the shadowplay clips

[raw zephyr]

how?

just use cq18 and don't bother with anything else

[ashen spindle]

Don’t need the best quality I just need to see what happened but duration, quality etc. all don’t change that I get a notice framedrop every so often, and I am absolutely not gpu limited

[astral haven]

nvidia has been making it less and less functional over time ofc
at this point its purely riding on the insane quality of the encoding hardware
if it werent for nvidia being 2 generations ahead in video processing units, shadowplay wouldnt be relevant

[ashen spindle]

So must be something like clearing cache of old rendering that gives the cpu a hiccup or something

[astral haven]

yea nvidia slams the disk hard, you gotta record to a separate disk, preferably not a ssd (well a tlc or worse ssd)

[astral haven]

rip hdd

[ashen spindle]

I mean slamming the disk shouldn’t matter, the game doesn’t read a lot of info

[astral haven]

until it does and your game has to wait for it to load in

[raw zephyr]

if you are limited by space, you can use cq20 or higher

[astral haven]

its actually insane how much boot drive matters in a modern system

[astral haven]

im limited by upload bandwidth

[ashen spindle]

But it happens constantly in 10+ minute fights where the only content that gets loaded is effects and particles

There isn’t much you could even try to load from disk

[raw zephyr]

1080p240 hevc, p1 preset, 60s clip is around 600mb-1gb depending on the game

[raw zephyr]

i record in hevc so i reencode most of my stuff anyways

[astral haven]

i can upload 800mb in a day

[raw zephyr]

https://github.com/vladaad/discordcompressor

[astral haven]

ooooor instead i could have my videocard do it in realtime and i can have it uploading already while im playing the game

[raw zephyr]

i have no idea what you are trying to achieve

[astral haven]

i just wanna point the program to the hdd, and my hdd i have set to automatically back up to the cloud

ez pz no dealing with extra steps

[raw zephyr]

well if that's the case, you can use p7 hevc or av1, cq24 and have smaller filesizes and probably better looking videos

[astral haven]

and lower fps in game 🤷

[raw zephyr]

i get more fps when using obs when compared to shadowplay, but you need to wipe nvidia overlay

you can use nvcleanstall to only install gpu driver + physx

[astral haven]

hmm well maybe theres been an order of magnitude improvement from 29 to 30

but i dont buy it

its still using nvec and obs still has pretty sizeable overhead

[raw zephyr]

no?

[astral haven]

obs has 2 digit cpu usage at full idle on a quad core (with ht)

[raw zephyr]

1%-5% cpu usage, no idea what you are doing then

[astral haven]

im talking peak fam

doesnt matter how low it goes, only how high

[astral haven]

ok thats more realistic now, how many cores you got