Tech Lab forum - General Discussions!

[fluid quest]

when I was buying mine even second hand was 800+

[jade scaffold]

I say this as someone who owned a brand new epyc platform

[fluid quest]

my cpu cost half my mobo

[jade scaffold]

750 for the mobo CPU was like 1.4 or something

[fluid quest]

what mobo ooc?

[mossy gull]

Does it?
Most people I know swap the motherboard as soon they swap the CPU, they don't understand that X670 works with the 9800X3D

[jade scaffold]

The standard ATX super micro one

[fluid quest]

for epyc 4000 or 9000?

[jade scaffold]

9004

[fluid quest]

huh

[jade scaffold]

9174F specifically

[fluid quest]

it could also be country import prices

cause i'm in australia

[jade scaffold]

Mobos genuinely the cheapest part of servers

[fluid quest]

and prices are a mess here

[jade scaffold]

I also have a Xeon mobo

[fluid quest]

H13?

[jade scaffold]

Which is old by today's standards now but I got it 3 years ago for about 400

CPU was only 200 for the 20 core

[jade scaffold]

Yes

[fluid quest]

yeah they're cheap in US

my issue was nobody sells them in aus

[jade scaffold]

I'm in the UK

[jade scaffold]

There's not much reason to ship them there tbf

[fluid quest]

most servers are prebuilts here

hence why I had to go 7002 for mine

[jade scaffold]

So they are everywhere really

[fluid quest]

I relied on the second hand market, so everything was pretty much way overpriced

a H11SSL was 800

H12 was 1200

cpus were half the price or more

[fluid quest]

that H13 mobo is 1100 here now

[jade scaffold]

Don't have the epyc anymore

Still have the Xeon

[fluid quest]

god you're making me jealous lmao

[jade scaffold]

Still chugging away

[fluid quest]

my 7402p looks tiny compared to that

[mossy gull]

That's Epyc, sorry, had to make the joke.

[jade scaffold]

Same width but a bit more square

[fluid quest]

I was gonna go 9004 ES cause you can get those cheap here for some reason

but mobos were way too pricey

[jade scaffold]

I want one of Intel's new Xeons thoes things are MASSIVE

[fluid quest]

and I didn't wanna deal with ECC ddr5

maybe one day I'll have the money to buy a server brand new

[fluid quest]

but seeing as my whole build was like just over 1100 USD

i'm chilling

[jade scaffold]

Look at that socket

[fluid quest]

geez

homelab?

[jade scaffold]

That Xeon is about the same size as eotc 7000s the socket under it dwarfs it

[jade scaffold]

Yea but I run mostly game servers

[fluid quest]

nice

[jade scaffold]

Wanting to do more AI stuffs tbh

[fluid quest]

I'm working on my own AI project

but not using the server for it rn

don't have the cash to drop on an enterprise nvidia card

[jade scaffold]

I can use my main GPU for the moment

That's where the sparks meant to be good

[fluid quest]

and honestly AI stuffs crazy overrated rn

[jade scaffold]

100gb of vram and 1 PFLOP of FP4 for 3k

[fluid quest]

geez

[jade scaffold]

I love ai it has some truly amazing uses and potential

But I think image generation and chat bots aren't the interesting way to go

[fluid quest]

I love AI/ML not LLMs and image generation

LLMs are glorified autocorrect

take input text, look at massive library of overly complicated data to see what a person would say, say it, repeat

could be made way more efficient

and there's not much reason for AI companies to improve it

they're making enough bank off the practically bruteforce method right now

what I do wanna see (and am working on) is a hybrid AI

making use of a bunch of smaller neural nets + traditional computing

and having the overall AI be live learning rather than pretrained

[mossy gull]

Raspberry Pi AI🙃

[fluid quest]

neural net uses traditional sqlite database as memory and data storage rather than inefficiently encoding it into a neural net

kinda an odd approach but it would be hella strong if it works

anyways, why would companies make an efficient LLM when they can just throw nvidia more money

[mossy gull]

If you can make it more efficient, you can do it even better on stronger hardware

[fluid quest]

efficiency means effort, AI companies can buy better hardware to make their improvements without having to optimise (takes time and work and manhours)

and anyways LLMs by design can't be optimised much in their current state

[mossy gull]

So instead you want to waste more money on electricity instead of optimisations?

[fluid quest]

oh no I want it to be optimised

i'm just saying it won't be

it's cheaper for ai companies to not optimise

[mossy gull]

Lets say this, you say now you can't, but people see that as a challenge and will try anyways

[fluid quest]

oh yeah I'm trying too

I say you can't because in the current way LLMs work you're cramming what should be split up in something that's designed for just language

[mossy gull]

Once I have a better job or 1 where I have more time, I am gonna try to make my own aerodynamic electric motorbike

[fluid quest]

a way more efficient design would be to have the language part separated out from the data part

or, more realistically, have multiple models all of which have different specialised purposes

it's like the classic problem of scaling a monolithic service compared to a bunch of microservices

LLMs are basically a monolith rn, the scaling method is throw more hardware at it

with microservices you can achieve way more efficiency if you design it right

like rather than having to teach a language model how to do math, it just knows when it receives math it wants to compute it runs it directly on the computer

it doesn't need to learn every single piece of information ever and store it in a single model

a well designed LLM can understand context and pull information like that from the internet and other sources without needing to store it

and suddenly you no longer need hundreds of gigs of VRAM and a ton of storage

[mossy gull]

I don't understand machine learning, like I suck at programming.
So to me when making something more efficient is also making it better.

[fluid quest]

right now i'd see LLMs like trying to use a hammer and only a hammer to build a house, a hammer can be used to build a house but doing it with just a hammer is incredibly inefficient

you need other tools for any level of efficiency

and for some reason it's cheaper for companies to build nuclear power plants to run their unoptimised messes instead of getting more tools

mostly because getting more tools takes research time, and AI thrives off trends and hype

[mossy gull]

But we can't build unlimited amounts of powerplants, and that's an issue.

I know how to make electricity in a more efficient way, but never got the funding to do so.

[fluid quest]

oh for sure, except the US gov threw a few billion at it

and investors throw tons at it

it's a hype bubble rn

it'll burst soon if they don't optimise but they don't care

i'm interested though, what's your method?

[mossy gull]

It's already bursting as entire places have black outs

[sacred seal]

I have done some playing around with mixed AI models and hard storage types back when we were dealing with 4k context windows. It does sort of work, but it is much slower and more resource intensive then just one better model.

[fluid quest]

it has more optimisation potential though

[sacred seal]

Yes and no. By the time you optimize it, your optimized model will be outperformed by the latest high end model

[fluid quest]

for now sure, but a mixed model has a higher efficiency and performance ceiling

[sacred seal]

LLMs are just advancing too fast right now based off the brute force method to really bother with optimizing

[fluid quest]

once the current model of things inevitably crumbles, mixed models will take over

I mean anyone who understands how llms work can clearly see the brute force method is not at all efficient

[mossy gull]

I wonder how it'll run on lets say a quantum computer

[fluid quest]

issue with quantum computers is bit space

[sacred seal]

Dont think LLMs are quantum readable

[fluid quest]

we're looking at <1000 qubits rn

meanwhile LLMs, excluding the fact that they'd need to be somehow ported, would need minimum a few billion

[sacred seal]

Well there is also the issue of being able to pull the result out of the computer. The types of data a quantum computer can compute are limited

[fluid quest]

yeah

quantum computers are good at probabilities

and certain problems

they're super specialised

imo AI's current state is underdeveloped and I'm not gonna actively use it till we figure something other than brute force out

[sacred seal]

I was reading an article awhile back that described how they had hardened encryption against quantum hacking. Really interesting stuff, using multi dimentional vectors

[fluid quest]

oh yeah quantum proof encryption exists

it's super cool

only issue is it's not implemented anywhere important

ssl/tls/https and ssh all use encryption methods which can be cracked with a quantum computer

reasonably in ~10-20 years

and as we know standards take that time just to get made, let alone get fully implemented

so I can definitely see a dark age of quantum encryption cracking

especially targeting older devices

[sacred seal]

I think we will be more like a y2k type issue. By the time anyone malicious can access a quantum computer, anything important will have already moved away from old encryption

[mossy gull]

We just need to use 1 of the bestand most efficient Quantum computers more, our own brains

[sacred seal]

Brains are not quantum computers...

[mossy gull]

Neither are super computers

Quantum computers come really close to how our brain works though

[sacred seal]

No? They are entirely different beasts

I would say a regular computer is closer to a brain then a quantum one

Quantum computers are extemely limited in the types of tasks they can run.

[mossy gull]

Like our brain, quantum computers don't just send 1 and 0 back and forth, they have everything in between.

Yeah QCs aren't strong currently, but I think if we put more development into them they'll eventually pull ahead.

[sacred seal]

That is a very different phenomena

[sonic meadow]

the brain is a biological computer. more analog than digital. but nothing like a quantum computer

[sacred seal]

You are comparing analog signal to quantum super positions

[sonic meadow]

quantum computers are good at a fundamentally different kind of task from classical computing. there isnt any "pulling ahead" to be done

[sacred seal]

There us nothing technically stopping us from making a trinary computer instead of a binary one. The reason we don't is they are less reliable

Well relyable is not quite accurate...

[sonic meadow]

they have been made. not been developed much though

[mossy gull]

Okay, closest to a human brain is apparently the Neuromorphic computer, whatever that means

https://www.intel.com/content/www/us/en/research/neuromorphic-computing.html

[sacred seal]

I think the literal brain chips are probably closer

[mossy gull]

I don't trust brain chips, but that's me, and from what I know is they just look at changes in certain brain waves

[sacred seal]

https://futurism.com/neoscope/mini-brains-human-tissue-living-computer

[mossy gull]

Unless I think of the wrong ones

[sacred seal]

I'm talking about literal brain tissue as a processor, not putting a chip in a brain

[mossy gull]

And that's what, less than 1% of a brain power?

Still impressive, now, I want a new brain so I don't feel pain 24/7

[sonic meadow]

brain "power" is... complicated, and directly translatable to conventional processing

[cinder lagoon]

usually pulling the plug while booting triggers a normal boot without fast (there's options to override that tho!)

[mossy gull]

Interesting
https://youtu.be/SA3EpbOc4fU?si=gaQ068d7F6FJGXUU

[fluid quest]

analog computers are good, but they're more like FPGAs and ASICs, they need to be made for a specific task

I think the best way to do it is to design your analog computer product with a digital sim, then make it on an actual chip

[cinder lagoon]

Not much benefit in trinary computers with the current technology since silicon transistors can only really be used in on direction so for the other direction i.E. The third state -1 you'd need another transistor.
At that point it's just more reasonable to use binary and just add more power by using those same transistors in that direction.

Could be really interesting if we had the tech to use trinary in the same space / component as binary.
Binary "emulation" would also be practically integrated so no need to completely re-engineer everything, just new stuff could be made more performant by utilizing the third state

[cerulean helm]

So, recently purchased a new to me car. A police package 2015 chevy impala. And I have discovered that one of the packages that came with this option (one of 2 police packages) completely disables the automatic headlights and they only work on manual. This is mildly irritating. So, gonna see if I can pull some favors at a local dealership to see if the GM tech 2 scanner can re enable the auto headlights.

[sonic meadow]

My car is positively ancient (2005), what about them is supposed to be automatic?

[cerulean helm]

The car has a light sensor that turns on the headlights/running lights depending on light levels. All automatic. My 2006 Impala (non police has this). This feature is disabled depending on the packages that were enabled/installed on the police Impalas.

[sonic meadow]

Ahh, mine just has ignition switched lights so I leave them on all the time

[mossy gull]

Manual headlight are still a thing in 2025?

[sonic meadow]

Well their car was a 2015

And only didn't have them because it was a police vehicle

[mossy gull]

2015 and manual headlights?

[sonic meadow]

And I have discovered that one of the packages that came with this option (one of 2 police packages) completely disables the automatic headlights and they only work on manual.

please... read

[mossy gull]

I mean I personally find it absurd that a 10 year old car doesn't have automatic headlights, like majority of cars even EU based police cars have automatic headlight that can be turned manually.

[sonic meadow]

its age is unimportant, its the fact that it was a police vehicle

police in some locales obviously require a lack of automatic headlights for whatever reason. or did when the trim was made. given that its a chevy, this means it was an american police force. they often get trims made by the american car companies that supply them, whereas many european police forces by off the shelf trims and modify them themselves

why? policey things.

[mossy gull]

I forgot USA police likes to creep up on people 😅
Just joking.
I just find it weird though.

[surreal moss]

Pretty sure it has automatic headlights. The package for police disabled the feature though. Probably as part of integration with the controller for the rest of the lights that were installed. I wouldn’t be surprised if this happens with EU emergency vehicles as well

[cerulean helm]

Yes, the headlight switch gives you the option to turn them on manually. Or you use the auto setting. This makes it so if the ambient light sensor ever fails you can still turn the headlights on. However, since my vehicle has the police package. The auto selection on the switch is not functional. I believe it can be enabled I just need a GM scan tool to do it.

[cerulean helm]

Exactly. On the RPO code list its the 9G8 option.

[sacred seal]

Looking into it, it seems like it may not actually have the sensor itself

Apparently they sometimes ship it with a dashboard without a sensor

Probably worth checking visually to see if it has the sensor before taking it into the shop

[cerulean helm]

That I have no idea on. Police package feature set pdf doesn't say how it was disabled. Just that it is. Though I found a GM service bulletin from the NHTSA that said its turned off via software and may or may not be able to be enabled.

Police package guide:
https://www.gmupfitter.com/wp-content/uploads/2021/05/2015_impala_limited_police_specification_guide_4_10.pdf

Official service bulletin:
https://static.nhtsa.gov/odi/tsbs/2015/MC-10113590-9999.pdf

This requires a dealership to do it though.

[sacred seal]

I was just digging through random messages, someone mentioned that his different model ex police car was missing the sensor module

[cerulean helm]

Yeah, I'm pretty sure the light sensor isn't missing as my dash lights auto dim depending on light levels. Guess I need to call the dealership.

[mossy gull]

I just find it weird😅

[mossy gull]

That's odd, maybe it's a software thing on the switch and a replacement can actually enable it.

[cerulean helm]

Service bulletin says it should be software enabled/disabled. So, gotta eather call the dealership and pay who knows what or find someone with a GM scan tool.

Well shit, called the dealer, dropped my name and relationship with the dealer to pull a favor or 2. And they are unable to software update the vehicle to enable to auto headlights as they can only do factory work.

Thats something that has to be authorized by GM itself

[mossy gull]

Oh fuck

That's gonna be fun

[cerulean helm]

Dead end really. The TSB was pretty clear how the authorization goes. Dealer can do it, but yeah. I am not a gov entity, lol. So, no go.

[sacred seal]

Not a car guy, but I am guessing flashing the module wholesale or replacing the board with a civilan model is out of the question?

[cerulean helm]

The BCM can be flashed as per the image I put up. It requires authorization via a gov agency as it was a former gov vehicle. The dealer won't/can't without that authorization. And changing the module would create way more problems. So dead end for me unfortunately.

You have to call GM's techline to request the software calibration. The dealer doesn't have acess to it themselves.

[sacred seal]

Well that's the official route. I know people have modded thier cars "os" on thier own, but I don't know what all goes into that

https://www.impalassforum.com/threads/how-to-scan-datalog-and-reflash-your-b-body-in-2025.1323187/

Another option that the shop may be more willing to do is flash the whole module to the newest factory settings.

As this is no longer a police vehicle, I don't see chev objecting to resetting it to stock settings.

[cerulean helm]

Yeah, look at the year on that. Thats OBD1 era cars. 1994-95. Not even the same car.

And any shop HAS to call GM's Techline support for the configuration software. Its not just available off the shelf. Independent shops don't have access to that for gov vehicles. A GM TEch 2 scanner is needed, but useless without the config software.

[sacred seal]

I just linked the post as an example, I am finding references to people doing the same thing on modern hardware, but no actual guides

There is a difference between configuring and reflashing.

From what I am reading

Again, not a car guy

Some stuff is going over my head

If you want to configure windows, you need the administrator account password. If you want to reinstall windows, you just need a usb

[cerulean helm]

You change the entire BCM it changes a shit ton more than just the headlights. And even then you can't reflash it to a different BCM model number. You again are looking at OBD1 GM vehicles that used Eproms to flash stuff. Thats a dead technology.

[sacred seal]

No, I am looking at articles referencing 2006 and 2010 chevs

[cerulean helm]

To put it simply, changing out the BCM bricks the car.

Not the same vehicle. Not even the same drivetrain.

14-15 Impala Limited Police were their own thing.

[sacred seal]

https://www.youtube.com/watch?v=KUX7zjCUt9E&ab_channel=APMotorsports

Not sure if this is accurate, but it seems everything after 2005 can be programed in the same way

[cerulean helm]

No, not the same. The Impala Limited itself was not a dealer sellable vehicle. It was only ever sold to fleets or governments. Thats it. To reflash a BCM to a non police Limited requires a "sample" VIN thats not a police package. A subscription to GM's SPS2 service system for the software.
But, because the ECM also has the VIN programmed to it and now won't match the BCM, you have major problems.

Way to much work, way worse things to go wrong. And would cost way more money than if the GM dealer was willing to change the config for the 9G8 headlight option.

[sacred seal]

That depends on how the car was shipped. Is it a stock "OS" where the feature was manually disabled before delployment, or did it ship with a custom OS with the feature disabled by default

I'd guess the former, but I have no evidence ether way.

[cerulean helm]

This is how they are ordered by gov agencies:
https://www.gmupfitter.com/wp-content/uploads/2021/05/2015_impala_limited_police_specification_guide_4_10.pdf

[sacred seal]

Yes, but that does not necessarily mean that is how they are off the assembly line. Do they take a regular impala and convert it to a police package, or is it built from the ground up to be a police impala

Given that it has the switch and sensor in it, I am guessing that it is flashed with regular settings, and then they turn it off after

[cerulean helm]

They were ordered and built. Because of how many options they had, such as mine doesn't have all the external light mounts or side lights. Plus the aux power options, interior antennae and such.

Regardless, as I stated the config and software is given out via GM itself to reconfigure these cars. You have to call to get it.

[tender plank]

Finally got my virtualization server up and running.

[surreal moss]

I spy proxmox

[tender plank]

Yep. Can't trust vmware anymore so I've been working on getting away from it.

[surreal moss]

Coming from a vmware house Prox has been really neat to mess with

[cerulean helm]

My Homelab uses Proxmox for the handful of VMs I have on it. For someone who had never even used it before it was pretty straight forward and simple.

[tender plank]

Yeah, pretty easy to pick up so far. I have experience with esxi and vSphere which carries some over.

[fluid quest]

proxmox is really good

has a lot of the features one would want from esxi and vSphere with none of the price tag

[rancid plank]

Goals:

• transportable mATX / ITX pc for university
• Able to handle my engineer workflow (aerospace engineer) (extensive 3D CAD [NX, Fusion 360, SolidWorks], MatLab & C++, as well as dealing with a ton of browser tabs)
• moderate gaming (modded Minecraft, modded Kerbal Space Program, CounterStrike 2, etc.)
• Willing to dig significantly into my savings for this one, upper limit for budget is $2000

Current spec'd build: -----

BuildCores Parts List: Link

Case: Cooler Master Q300L V2 MicroATX Mini Tower Black Tempered Glass Side Panel - $58.99 (Amazon)
CPU: AMD Ryzen 7 9800X3D 4.7 GHz 8-Core AM5 - $451.50 (Amazon)
Motherboard: Asus B850 TUF GAMING B850M-PLUS WIFI AM5 DDR5 Micro ATX - $219.99 (Amazon)
GPU: Asus PRIME OC GeForce RTX 5070 12 GB - $579.99 (Newegg)
RAM: G.Skill Trident Z5 Neo RGB Black DDR5-8000 CL40 48GB (2x24GB) - $219.99 (Newegg)
CPU Cooler: Corsair iCUE LINK H100i Water 240mm RGB Black - $129.95 (Walmart)
Storage: Samsung 990 Pro w/Heatsink 2TB SSD M.2-2280 PCIe 4.0 NVMe - $159.99 (Amazon)
Power Supply: MSI MAG A850GL PCIE5 850W Fully Modular 80+ Gold Certified - $121.99 (Amazon)
Case Fan: Corsair iCUE LINK QX120 RGB 120mm RGB PWM Black - $54.99 (Amazon)

Total Price: $1997.38

Thoughts? Do all the parts play nice with each other and are reliable / non-chintzy? I previously bought my main rig from Ironside PCs all the way back in 2019, so I have no PC building experience .. although I am technically inclined, or at least, I think I am!

This has been a bucket-list action point for a while now

nearing closer to actually pulling the trigged

seeing as my laptop has been chugging for a while now with my workflow

and how buttery smooth everything is on my main rig

[sacred seal]

You wont have any footprint improvements going with a matx as opposed to atx. The only difference between the two is that atx is taller, which is a non issue in most cases
2z24gb is a really odd config
I would not advise a water cooler for a first time builder. They are not as reliable as air coolers and they do not offer much in the way of increased performance.
SSD is overpriced

Overall its a reasonable build, expecially for a first time builder. It should do what you want without issue

[jade scaffold]

It's cases are the way to go for small form factor

Skyreach 4 mini
Or the upcoming skyreach 5 mini if you CBA with the pain that is ultra sff

The 9800x3D you can limit to like 45w and it'll be plenty enough for the 5070

So you can get a really small CPU cooler and still have it run quiet and cool

[sacred seal]

Here is how I would build out a rig at that price point
https://pcpartpicker.com/list/tFGYZc

[jade scaffold]

Goals portability though isn't it

[sacred seal]

Transportability with a dorm room in mind. As far as physical footprint, this is not sinificantly worse then what was initially proposed

If size is a priority though, you can of course come up with something much much smaller

I don't have a tone of experince with the software described, but with solidworld vram is a limiting factor. So I somehow managed to get a 5070ti in there instead of a 5070

[jade scaffold]

Could fit most 5070tis into a fractal ridge

[sacred seal]

Sorry, talking budget wise

Not space wise

My bad

[jade scaffold]

Not sure what the S5M can fit might only be 2.5 slot might also be 3 slot dint know loads about it yet

[mossy gull]

While the x800X3D are great chips for gaming, they'll be less for 3D programs where maybe the U7-265K shines better with its extra cores

https://pcpartpicker.com/list/6QdPkf

The U7-265K has a 20% faster single core and a 40% faste multi-core if you're not using the X3D features over the 7800X3D

10% faster single core and 30% multi core against the 9800X3D

It's a decently small build for almost 100.- less, and you get Thunderbolt port as standard if you ever need that, probably not😅

It also gets less hot

[sacred seal]

Wow, intel sure cut the price of the 265 down. It's not even a year old

I would not buy a 750 watt psu in this day and age though. No upgradability

Even a 850 watt is on the low end

[mossy gull]

They did, they compete with themselves 😅
But even a cheaper motherboard than the 1 I listed will save you more than 250.-

[mossy gull]

Is it?
For a build like this it should be enough

The 265K can use 250w, and the 5070 Ti anywhere between 250w and 350w, I am not sure though

[sacred seal]

No upgradablility. It's good enough for the current hardware, but your not going to be able to put in a beefier gpu and cpu

Psus are also more efficient at a lower load level, but it's not huge

[mossy gull]

What idiot is gonna release a 350w CPU, not looking at you AMD Threadripper

[sacred seal]

Transient draw can exceed base

Ran into that issue with my last psu

[mossy gull]

Fair enough

[sacred seal]

Worked fine 99 percent of the time, but certain loads would trip it

[mossy gull]

At least it ain't a 450w PSU I once found in a 265K/5070Ti pre-build

[sacred seal]

I'm surprised that would even boot

[mossy gull]

So did I, but they had constant USB disconnections and constant black screens during intensive loads

Got them an 1KW PSU and all problems were instantly gone

[jade scaffold]

PSU should be able to handle transient spikes any PSU that can't is bad

[sacred seal]

When the PSU is already 100% loaded?

[jade scaffold]

Yes

[mossy gull]

Yep

It's a requirement of the ATX3.1 spec

[sacred seal]

Nvidia was doing what, up to 150 watt of transient?

[jade scaffold]

They should be able to handle I'm excess of 1.5x their rated wattage for the nano second long transient spikes

[mossy gull]

I believe they should be able to temporarily spike to double their rated power

[jade scaffold]

Transient spikes are usually 2x the sustained power

So 3090s would do like 700w power spikes

[sacred seal]

I know at the launch of the 2000 series (If I am remember correct) there were a lot of issues with underpowered PSUs crashing

[jade scaffold]

That was 3000 series

[mossy gull]

Poor EVGA, we miss you buddy

[jade scaffold]

Atx 2.0 psus don't know how to handle transient spikes

[sacred seal]

Ah, atx 3 only came out 2022

[jade scaffold]

Even if they were well enough built to handle them alot would detect them as shorts and trip OCP

Where as modern PSUs the ocp is alot less aggressive

[mossy gull]

Tell that to the people who combine an ancient EVGA 1KW PSU with a 5090.

[sonic meadow]

for that psu it isnt a transient spike, just load

[mossy gull]

"BuT iT iS 80+ PlAtInUm"

[jade scaffold]

But any atx 3.1psu should manage for a brief moment double their rated wattage

[sacred seal]

I don't know that I would trust a PSU to handle those kinds of transients, but its good that its at least supposed to spec for it

[jade scaffold]

5090 can do 1.2kw on transient spikes

[mossy gull]

Oh I know, my brother had his 5090 trip OCP on his 1.5KW ATX2 PSU😅

[jade scaffold]

Only had 1 PSU blow up sofar and that was because bequiets dark power 13s suck massive balls

[mossy gull]

Which is weird because the 12M is fine

[jade scaffold]

Yea

I think it's something Todo with the overclocking mode it has

[mossy gull]

😐

[jade scaffold]

It drops like a big bus bar over the rails I think that might be where the issue occurs

[mossy gull]

🤔

https://wccftech.com/amd-radeon-rx-9070-xt-surprisingly-beats-out-nvidia-geforce-rtx-5080-in-vulkan-dx12-benchmarks/

[jade scaffold]

Yea it wins in very specific benchmarks

But looses in practicality

[fluid quest]

to the 5080 yeah

5070 ti not so much

i'd take a 9070 xt over a 5070 ti any day

at least I know it'll be supported with constant (good) driver updates

unlike nvidia's constant reduction in performance

although RDNA 5 looks really good so far

seems like it's gonna be a similar switch as the one from excavator to zen

new arch name, possibly new branding too

[jade scaffold]

Nvidia haven't been reducing performance what lmao

[fluid quest]

in their drivers I mean

[jade scaffold]

Performance has gone up since launch because the launch drivers sucked

[fluid quest]

on the windows side especially

[jade scaffold]

Just not true tbh

[jade scaffold]

There is no RDNA 5

[fluid quest]

wdym

I mean it's UDNA

but

same same

[jade scaffold]

It's not it's architecturally a total rework

[fluid quest]

I know

i'm aware

[jade scaffold]

Hopefully a very good one at that chiplet architecture and all being possible with it

[fluid quest]

but a lot of outlets are calling it RDNA 5

or RDNA 5/UDNA

[jade scaffold]

Cool they are wrong lmao

[fluid quest]

yeah

[jade scaffold]

AMD might compete with Nvidia again next gen thank fucking god

[fluid quest]

yup

my next gpu is for sure gonna be amd

[jade scaffold]

Mines gonna be who evers is faster

[fluid quest]

I don't care tbh

nvidia's linux drivers are subpar at best

[jade scaffold]

Hoping for multi die gpus from both

[jade scaffold]

Very true

[fluid quest]

i'm using them rn, basically always have to use beta builds for a decent experience

and it's honestly peaked at 30 series

cause 40 and 50 have issues

and 10 is out of support

[jade scaffold]

It's sad because their enterprise drivers are fantastic

[fluid quest]

oh yeah

cause they just go whichever makes most $$$ for least effort

[jade scaffold]

I'm looking at either getting the DGX spark or the framework desktop mobo

[fluid quest]

if the general trend is linux for enterprise, they write enterprise linux drivers

[jade scaffold]

The enterprise chips are also the best

[fluid quest]

i'm probably gonna skip the first gen desktop ARM chips

as much as it'd be great

I want something a bit more traditional

[jade scaffold]

Translation layers are really good now

[fluid quest]

oh yeah

[jade scaffold]

The one windows has is actually fantastic

[fluid quest]

I use arm for one of my servers

and it's matured for desktop too

[jade scaffold]

Main issue with the snapdragon windows chips is their GPU drivers

[fluid quest]

just kinda wanna wait it out on the more hardware side

[jade scaffold]

Hardware wise arms crushing it

[fluid quest]

the day I can get a socketed arm chip in an ATX mobo with standard components

i'll fold

heck even a socketed SoC with pcie out

[jade scaffold]

X925 cores are on par with the Ryzen 9950Xs cores

Yea a socketed none sever arm chip would be nice

[fluid quest]

I just love upgrade paths

I mean my pc's almost 5 years old now and i'm not replacing it for another 5

i'll likely VRAM mod my 3070 before getting a new card just to breathe more life into it

[jade scaffold]

I upgrade every 2 GPU gens usually

[fluid quest]

sadly my budget does not allow for that

[jade scaffold]

Do really need me a new CPU

Mine is not fast enough for my GPU

[fluid quest]

I live off the idea of like

if I can play what I want at frames i'm happy with

I don't need to upgrade

the only upgrade I brought recently was storage and RAM

mainly cause I kept maxxing both out

[jade scaffold]

I'm targeting high settings 5120x1440p 120hz base frame rate

[fluid quest]

2x8 + 2x16 gb now for 48gb total, and 2tb + 1tb storage

[fluid quest]

yeah i'm happy with 1440p 60fps or even 30fps

I find that to be good enough for me in a lot of games

[jade scaffold]

I'llive with 60 if it's stable I won't even pay mind to anything less than

[fluid quest]

i generally take consistent lower fps over inconsistent high fps

I'm playing dune awakening on high-ultra settings at 40fps

1440p

I could turn on dlss for 80

and fsr4 framegen for 120

but it makes the quality so much worse to me there's no point

[jade scaffold]

Fsr4 is amd exclusive

[fluid quest]

nah

[jade scaffold]

DLSS4 is fantastic

[fluid quest]

fsr4 framegen shows as an option

for dune awakening

even on an nvidia gpu

[jade scaffold]

Yes it is it uses AMDs ai engine and atleast as current does not work on other cards outside of rx9000

That's just FSR3

[fluid quest]

huh

odd

alright

in any case

i'm not a fan of framegen and AI upscaling

[jade scaffold]

I quite like it

But some games implement it really badly

Like in cyber punk and WH40k Darktide DLSS upscaling looks really good it's almost artifactless

But in hunt showdown you get MASSIVE ghosting and smeering which is weird

[fluid quest]

tbh I think I'm just not a fan of how it almost incentivises devs to optimise less

not all games sure

[jade scaffold]

Oh no totally agree there

[fluid quest]

but there are games where even the 5090 can't hit 60fps without it

[jade scaffold]

But I think DLSS itself is a fantastic thing

[jade scaffold]

That's usually only if you turn setting on the game tells you not to

[fluid quest]

it's good, but it's being used in the wrong places

DLSS/FSR should be used to boost lower end devices, not as a requirement for all devices

[jade scaffold]

Like cyberpunk max settings psyhtracing 30fps on a 5090

You aren't meant to run the game like that thoes settings are for showing off tech

[fluid quest]

oh yeah not that

there was a few games

lemme find one hold on

[jade scaffold]

Or even to get really high FPS in AAA games

[fluid quest]

the silent hill 2 remake

on release

ran sub 60fps on the 4090

[jade scaffold]

Yea

[mossy gull]

Pathtracing is simply extremely heavy, no matter if you have a 5090 or a stack of RTX 6000 Pro.

[jade scaffold]

Do keep in mind that every 4090 and 5090 benchmark will be done at 4k usually because otherwise they just choke on the CPU

[fluid quest]

I give pathtracing a pass cause that's pretty much peak lighting, doesn't get much better than that

i'm talking mainly about games with their recommended specs being 60fps with dlss

and then asking for a 40 series card

or even 50

[mossy gull]

Oh yes it can

[jade scaffold]

Not really

[fluid quest]

in games I mean

[jade scaffold]

Pathtracing is how they do movies

[fluid quest]

anything above pathtracing you're going into minutes per frame

or hours

or days

[mossy gull]

Depends on your hardware

[jade scaffold]

There's nothing above really other than just more paths to trace

[fluid quest]

even on high end hardware

[fluid quest]

that too

[mossy gull]

My brother tried to run path tracing with I believe 12 passes in Cyberpunk, only for him to play a literal slideshow😅

[jade scaffold]

Movies dont tend to use denoisers atleast not super heavily they just throw rays at things till they look good

But nvidias DLSS denoiser is incredible

[fluid quest]

generally no matter how much hardware you throw at it, the performance is days per frame for movies and stuff

cause they add more hardware for the purpose of increasing samples

and passes

[jade scaffold]

Ray reconstruction is insanely good makes any raytracing just look 2x better with 0 cost

[mossy gull]

I can't wait for the day those names mostly disappear and it being just a basic thing for games and GPUs to have😅

[jade scaffold]

What

[fluid quest]

blender cycles is just pathtracing

and that often takes at best minutes per frame for a high quality frame

[jade scaffold]

We are now getting to the point games are being purely raytraced

[fluid quest]

which arguably is not a good thing

as much as I love pathtracing and raytracing

[mossy gull]

I can hear the 1080Ti owners crying already

[jade scaffold]

It works out good in the long run

[fluid quest]

there's optimisation to be done

and benefits to be had in traditional lighting techniques

[jade scaffold]

Tbf raytracing optimisation is already really good

[fluid quest]

I mean, traditional lighting with raytraced shadows

solid medium

I think there's a lot to be done

cause the current path is less "improve software" and more "throw more hardware at it"

[jade scaffold]

At current NVidias GPUs atleast don't take a huge performance hit from raytracing

[fluid quest]

and the best developments and advancements came from limits in hardware

[mossy gull]

Fun fact:
Capcom uses Pathtracing so they can place the raster shadows in the correct location and so it behaves as natural as possible.

[jade scaffold]

Software is as good as it gets for raytracing

[fluid quest]

debatable

I think for current raytracing yeah

[jade scaffold]

Like the math behind raytracing has been really well understood for 40 years

[fluid quest]

but there are other techniques that are worth looking at

[jade scaffold]

It's not a new thing

Hardware just wasn't yet good enough to push it

[fluid quest]

yea

I think it's a matter of at current we're pushing raytracing too early

I still think of RT as in the tech demo phase, not the default setting phase

and I do not like that a lot of games are releasing as RT only

[jade scaffold]

All improvements are going to come from hardware and raytracing hardware is some of the fastest developing hardware arround

[mossy gull]

That sentence reminded me of this
https://youtu.be/Q9t49KFfo24

[jade scaffold]

I mean on a 5060ti or above in most games I wouldn't bother turning RT off tbh

[fluid quest]

oh yeah

even on my 3070 I keep it on for most games

[jade scaffold]

I think we are mostly there

[mossy gull]

Then you live in the past, it's also getting good on AMD, not to mention I personally always have it turned on.

[jade scaffold]

One thing I'm excited for is DLSS's future

[fluid quest]

I'm honestly just salty that my gpu's barely last 5 years

[jade scaffold]

Pretty soon generated frames will have functionally lower latency than real frames

[mossy gull]

Okay, 1 nice thing, Dragon's Dogma 2 uses RT for the shadows and some reflections I believe, I notice little to no performance loss by turning it on or off.

The game is faster VRAM and CPU starved than GPU starved.

[fluid quest]

I think tbh

a lot of people see games using RT

and think that's why they perform so bad

[jade scaffold]

Some games are starting to run faster with RT on than off

[fluid quest]

when it's usually model and texture optimisation that's the issue

[mossy gull]

Dragon's Dogma 2😅

[jade scaffold]

Yea I think it's mostly optimization in like Polly count and post processing

[fluid quest]

basically any lumen game cause "rt off" is just "use software rt instead"

[mossy gull]

Fuck Lumen, I never seen it work well.

We need RT, like most GPUs, heck any modern GPU supports it, even that little APU in the Steam Deck.

[fluid quest]

too many games with no real RT off option

[jade scaffold]

Pretty much anything new comming out is gonna be RT only now I think

[fluid quest]

I do find it funny that nvidia's pushing for all this and then shipping gpus that have 8gb vram

the 1080 ti had 11

that's like

a 7 or 8 year old card

or smth

[mossy gull]

Nvidia is extremely stubborn, but even they start to listen

[jade scaffold]

Yea the 5060 should have atleast 12

[fluid quest]

same with the 5050 arguably

and amd shouldn't have released an 8gb 9060 xt

[jade scaffold]

Hell id take 8gb of fast ram and 4gb of slow ram just so you can use the actual ai features

[mossy gull]

3060 12GB, 4060 8GB, 5060 8GB.

Fun fact: because the extra VRAM the 3060 did outperform the 4060 in some VRAM heavy games.

[jade scaffold]

Like Nvidia have made their GPUs compress memory really really well this gen

As much as 90% compression of some types of data

But still

[mossy gull]

I still wouldn't build on VRAM compression, it's still in baby shoes

[jade scaffold]

12gb 5060

16gb on 5060ti and 5070

20gb on 5070ti and 5080

[jade scaffold]

It's driver based atleast

Doesn't require to work per game like DLSS currently does

[mossy gull]

True, but I wouldn't count on it still, still too much in baby shoes

[fluid quest]

AMDs got really good hardware compression on the ps5

[mossy gull]

I am gonna be honest, I don't like when they force DLSS or FSR when there's no option for DLAA or FSR Native.

[fluid quest]

or goddamnit let me pick FXAA

or MSAA

[jade scaffold]

There isn't a single game that forces DLSS on

Fxaa looks like hyper mega ass

[fluid quest]

or just any option that isn't TAA or TSR or DLSS or FSR

just give the option

[mossy gull]

Dune Awakening, so far I haven't found a way to fully turn upscaling off.

[fluid quest]

it's built into the rendering engine 99% of the time

[jade scaffold]

All AA that isn't running at higher than native sucks

[fluid quest]

dlss, unlock scaling factor, set to 100%

that's what I did

[jade scaffold]

Yea

[mossy gull]

I rather have DLAA, XeSS/FSR Native over TAA, TAA sucks ass.

[jade scaffold]

They are all TAA

[mossy gull]

I play on XeSS Native, I have a 9070XT so no DLSS for me.

[mossy gull]

But they don't look as bad as TAA

[jade scaffold]

What's interesting is that taa and DLSS quality varies between games

[mossy gull]

I wonder how this will run

Original post, CD Projekt Cyberpunk 2077:

The Mac App Store version of Cyberpunk 2077: Ultimate Edition just got a sleek upgrade to path tracing, bringing it in line with the versions available on GOG.com, Steam, and the Epic Games Store 💥

Crank up the visuals and dive into the neon-lit streets of Night City! ✨

P.S. For all snoopy runners out there — a download is required but it won't change the version of the game.

[jade scaffold]

Like I said darktide or cyberpunk look almost indistinguishable with DLSS on hunt showdown turns to soup

[mossy gull]

It's the same with Ratchet and Clank, it looks absolutely amazing with FSR, but Mechwarrior looks likes absolute ass with it.

[jade scaffold]

But yea hugely excited for reflex 2 to work with DLSS frame gen that'll be insane

[mossy gull]

I am curious how it's gonna pan out the coming few years

AMD is playing catch up, Intel's B580 is doing pretty good for the price, Nvidia is pretty overpriced, maybe we'll see ARM or Qualcomm pass by😅

Heck, maybe Apple steals the crown of most powerful "GPU" for games

[jade scaffold]

Apples GPU is pretty good

I think it's like 3090 equivalent in performance and about 5090 efficiency

I think apple are currently winning in single thread performance though hilariously

[mossy gull]

I know, I seen the M4 Max do some path tracing

It's really really good for what it is

[jade scaffold]

Mostly helped by it's massive memory bandwidth Vs competitor Apus

[mossy gull]

Their only competition with that is AMD😅
Nvidia is currently nowhere to be found and Intel hasn't had an APU for quite some time

[jade scaffold]

I mean intel have an apu

Nvidia have one

Apples is just better

With NVidias placing 2nd place amd placing 3rd interms of performance and efficiency

Intel's apu is a competitor to the like basic little M series chip Qualcomms one and AMDs hx370

With Intel there placing 2nd place in performance and efficiency behind apple

Unless we talking CPU performance where it does loose to everyone in multi threaded and matches apple single threaded

Most interesting x86 chip of the last 5 years honestly

[mossy gull]

I mean consumer, I haven't seen much from Intel in the APU space and Nvidia isn't in the consumer market at all

[jade scaffold]

Lunarlake is consumer

N1X will be from Nvidia

[rancid plank]

Foot print isn’t an issue, anything that is dimensionally smaller than a ATX is ideal, I measured this one and it was fine to fit in the pelican, but anything smaller than that without compromising performance would be ideal

[mossy gull]

Wait, which Lunar Lake is an APU

[jade scaffold]

All of lunar lake

Lunar lake only really has 4 skus

A 8cpu+8gpu with 16gb of ram or 32gb of ram
Or
8cpu+7gpu with 16 or 32gb of ram

[mossy gull]

Ah, fair enough

[cerulean helm]

This, this is really cool. Can't wait to see how they develop this.
https://www.youtube.com/watch?v=T17bpGItqXw

[mossy gull]

https://www.guru3d.com/story/intel-nova-core-specifications/

[cerulean helm]

Well, jumped into undervolting on the gpu. And, I'm pretty impressed, and shocked. Haven't tested it through a ton of games but so far its steady. I've dropped anywhere from 15-20c in temp with no performance loss. Kinda crazy.

[cerulean helm]

Well, ran Cyberpunk for about 30 min for a good gpu test with the undervolt. Running 1400p with RT on the games psycho setting and everything else on high with DLSS on quality I saw no performance drop in fps. The gpu bately hit 60c and most of the time was in the low and mid 50s. Just crazy.

[fluid quest]

jesus core count on a desktop chip

I mean it doesn't touch server chips but also they're server chips

though knowing intel they'll fuck something up

[mossy gull]

They use same architecture as on the 1st and 2nd gen Ultra CPUs, I think it's much better than the old structure we have.
Not only are U5, U7, and U9 mostly the same, you can easily scale them up and down.

[jade scaffold]

Nova lake is a totally new architecture

Arrow lake is getting a refresh this year though

[sacred seal]

A pelican case is not really useful for transporting a computer. The main concern is not external impacts, but rather internal ones, specifically the GPU and CPU cooler.

When I brought my full sized computer to Japan, I pulled the GPU and mechanical hard drive out of the system and transported them seperately

[rancid plank]

so you brought an ATX build to JAPAN?? jeez! yeah, I have a big luggage, backpack duffel bag, and small luggage (25" pelican) so transporting all of that + a PC is hard
but if you made it all the way to Japan ... so regular ATX builds are possible to transport feasibly?

hmmm

a normal ATX build would open up the possibilities a great deal

[sacred seal]

Yep. Brought my PC and 2 27" monitors. Case fit nicely in my checked bag, and I just filled it in with my clothing to act as padding.

Put the GPU and hard drive in my carry on

Though I did ship it without the glass side panel.

Could probably have brought it wedged between the two monitors, but it was not worth the weight

But yeah, in my opinion there is no functional difference between matx and atx when it comes to transportability.

ITX is a different beast, but the small cases are typically harder to work with and there is a decent cost increase.

[sonic meadow]

Grrr. My ups isn't starting anymore. I'm wondering if the battery is dead 🤔 .
Came down to it reporting overload+replace battery, and power cycling it has it continuously try to test the battery before power cycling itself 😦

Admittedly the ups is probably 20 years old now, but the batteries are much newer

[sonic meadow]

Batteries are reading as 12.0v, which is a tad low for lead acid bricks

[mossy gull]

I thought they keep using the chiplet based architecture they've been using for all Ultra CPUs?

[jade scaffold]

Yea it's still chiplet based but it's not the same architecture

Same as how zen 2 and zen 5 are both chiplet based but they aren't the same architecture

[mossy gull]

Yeah, I meant the chiplet based architecture😅
I know they will be different, we probably wont get another Sandy-Bridge for the consumers

[cinder lagoon]

Should not prevent a UPS from starting, just displaying a fault

[sonic meadow]

Maybe. Would need to actually have a look at the documentation 😆

[fluid quest]

i've found a GladOS voice that works with the TTS to mic app i'm making

this is so peak

[sonic meadow]

Regardless, I've got the pair of batteries charging on an external lead acid charger and will have another look when I am back on site on Tuesday

[mossy gull]

I didn't expect this from Corsair

https://youtube.com/shorts/VH4VCST77io

[cinder lagoon]

Ughghh thing about the switch dock is - it's really just a USB-C dock but with encrypted communication.
It makes me so mad.

[mossy gull]

Ofcoarse, so we can't use our better designed docks

[jade scaffold]

Hey do you lot think a thick 280mm radiator will be enough for 800w ?

[cinder lagoon]

...how thick?

chonky boy? yea, about adequate.
slim boy? hell naw

at least not at low water temps

[jade scaffold]

30mm rad 40mm thick fans

[cinder lagoon]

static pressure is important, yeah.

30mm sounds OK-ish. would choose thicker if available or feasible

i think there's up to 36mm

[jade scaffold]

I intend on pretty high coolant temps I'm aiming to keep components at 80C or lower 45C+ fluid temps would be fine

I don't know the clearance yet of the case all I know for sure is it can fit the 4090 FE

So at a minimum 60mm

[cinder lagoon]

hmmm I see

[jade scaffold]

I'm hoping for 70mm

What would be best to use as coolant just the usual water based or would something else offer me better performance

This is the case the S5M it's not out yet they are currently figuring out production

[cinder lagoon]

water has the best heat transfer capabilities of the coolants

[jade scaffold]

Oh I miss spike before it's 280mm I'm trying

Not 240

[jade scaffold]

Planning on using this fan mostly because of that insane static pressure also apparently reasonably quiet at low rpms

[balmy flicker]

IDK if this exists but I am looking for an "I home" clock radio dock that is not locked to the apple eco system or depend on bluetooth/wireless charging.
basicly I want to pair a USB speaker, charger and clock that syncs with a phone in to one self contained brick

I can find only Bluetooth and wireless or clock radios that at best take a 3.5 aux (can be dealt with USB aux splitters exist but still)

[cinder lagoon]

Wtf is an "I home" clock ? XD

So you want an

• Apple compatible
• Android compatible
• dock
• that has an integrated clock radio
• that acts as an alarm clock being synced to the phone's alarms
• and plays music from either the phone or FM/AM?

Right?

Did I miss something?

[balmy flicker]

IDK about the raido part but yah hard list I know

honestly I just want the clock

phone alarms can be done with just the phone

[cinder lagoon]

I think you might be partly out of luck.
Either there's some bluetooth docks/radios that require an extra app to sync with the phone (and at that point why should it have wired anyways) or they only act as an audio device which frankly would be enough because the alarms will play through the device

[balmy flicker]

I want a as a single unit

clock
USB speaker
fast charger/charging dock

I found some stuff that is close but they are almost all Bluetooth.

[cinder lagoon]

I believe the closest device would be a display ALEXA

[balmy flicker]

I have an I home but the dock is getting flakey

which is makes it kinda useless

[cinder lagoon]

Yeah bluetooth allows a hardware-independant communication with the external device and also audio - why is bluetooth not an option?

[balmy flicker]

the point is to chage my phone and BT is inefficient and causes my I phone (SE) to get surprisingly hot when used

[cinder lagoon]

Hmmm yea bluetooth was a little bit of a problem a couple years ago

[balmy flicker]

now that might be an issue with how phones don't always keep the screen blank

granted they BT speaker I am using pairs automatically but I also use earbuds in the hose and done want to fight a dock for the pair

[cinder lagoon]

If the dock is getting flakey - i don't think there are many alternative current products.
Stuff is moving away from wired audio because the cretins don't care about sound quality and it's literally cheaper than engineering a wired solution.

I believe your options are finding another unit like it or repairing this one

[balmy flicker]

most the units like it are wireless to boot now and that is my main issue I have never had wireless charging on my phone and want wired charging at a minimum

[cinder lagoon]

Yeah I totally get it

What about manufacturing a single-unit case for a multi-device solution?

[balmy flicker]

could be fun I am sure USB speakers still exist and If I didtch the clock all I need is a powered USB hub and said speaker assuming the phone is smart enough to see it as an audio output

[cinder lagoon]

(I know, I'm kind of addicted to finding problems for 3D-printable solutions now)

[balmy flicker]

I have a hub that has a 3.5 out

see I dont have a printer yet and I also am starting to do that

also this was the closest to what I want that I found clock radio/fast charger

[cinder lagoon]

Well, the clock can be had from cheap 7-segment displays and a wifi micro controller

[balmy flicker]

hmm the issue is syncing it with the phone (man is that is nice feature for DST)

[cinder lagoon]

Does the clock need to sync to the phone or is an internet time server enough lol :D

[balmy flicker]

could be a fun project add a battery bank in for power outage "durability"
both would work

[cinder lagoon]

Yeah

[sacred seal]

Computer survived bieng shipped back from Japan, but I think I need a new case now

[surreal moss]

Rip

[jade scaffold]

Just bend it back with your teeth

[cinder lagoon]

yea hang on lemme just get my hardened steel grills lol

actually looks fixable with a rubber mallet

[mossy gull]

Also with 3 clamps and 2 pieces of flat wooden boards

[sacred seal]

I don't think a rubber mallet would work, just because its an akward angle, but the clamp idea might be worth a try

[mossy gull]

Could this company kick Nvidia, AMD and Intel's ass in the consumer market any time soon?
https://bolt.graphics/

[jade scaffold]

No lmao

They can't even touch them in the data centre outside of a very tailored benchmark

Functionally it's only any good at raytracing so maybe it'd make a good RT accelerator

[mossy gull]

That's now though.

And the fact their RT seems to be pretty good they have a decent chance, especially in the consumer market if they allow us to upgrade "VRAM".

They seem to do much better than the Chinese who can barely if not at all run DX11 games.

[jade scaffold]

They don't allow you to upgrade vram

That extra memory is super slow completely fucking useless for rendering

APUs are too fast for DDR5X 8533

[mossy gull]

Lower latency than going over the PCIe lanes to your RAM

[jade scaffold]

Anything that requires any kind of memory bandwidth will eat shit on anything more powerful than a 1050ti

[jade scaffold]

It's not

CXL can do ddr5 over pcie gen 5x16 with only a 4% latency hit

[mossy gull]

Tell that to people who still rock PCIe 3.0 and get a almost 15% performance hit because they lack VRAM

[jade scaffold]

Regardless DDR5 useless for GPUs

[mossy gull]

Those people still have DDR4 though

[jade scaffold]

Intel run ddr5X 8533 ON THE SUBSTRATE. And it's not enough for lunar lakes iGPU

Additionally sodium has MASSIVE Latency and clock speed issues basically capping out at 5600mhz cl40

[mossy gull]

I know enough people still running 2nd gen AMD Ryzen CPUs or older with PCIe 3.0 because "it still works fine" or "it's too expensive".

So even DDR5 "upgrade" on a GPU would help their performances, even slightly.

[jade scaffold]

It wouldn't though

You'd still get dog shit performance and constant crashing

Upgradable gpu memory isn't a thing because signal integrity on GDDR is a fucking NIGHTMARE

And absolute best DDR5 isn't even enough to keep up with anything more than a 1050ti

[jade scaffold]

Additionally worth noting its not been built yet and it's performance is Entirely Theoretical

Same way the 5090 in in theory 2.2x faster in all regards to the 5080 but functionally only ends up 1.6x faster because diminishing returns and all that

This card from bolt also won't be cheap the cost of the material for the die alone would be atleast $350 never mind the actual cost to manufacture

They also don't actually release till 2027 when AMD and Nvidia released their next gen cards

Next gen both Nvidia and AMD will have 0 issue increasing their performance by double and as RT and AI cores are new they develop extremely fast chances are RT performance next gen could for the high end land at 8x depending on how big GPUs get

[mossy gull]

Not even less overhead?

Like PCIe 3.0 DDR4 would be faster for a PCIe 5.0 GPU over on-board SO-DIMM?

[jade scaffold]

No

There's be technically better performance but it'd still be fucking unusable

[mossy gull]

That I find hard to believe really as I seen 5060 8GB getting over 15% performance hit as soon it has to swap memory with the RAM on PCIe 3.0

[jade scaffold]

Also again someone who can't even afford the 300 USD upgrade to a ddr5 platform can't afford a card thats core costs 350 in materials alone

[jade scaffold]

No you haven't

You get massive stuttering and then crashing

That's what you get if you run fully out of VRAM

You at first loose performance due to compression then if you run out you eat total shit

Putting ddr5 on a card also cripples it's normal memory bandwidth if it's a hybrid card

[mossy gull]

Uh no, that really depends on the game.
I seen stutters and higher RAM usage but no crashes when a GPU runs out of VRAM.

[jade scaffold]

That's only if you aren't pulling much from ram

If you start pulling loads from ram it dies

Because you choke out your CPU you choke out you SSD and you choke out the pcie bus

Regardless you'd see the exact same issue on a 5090 their card even with 128gb/s memory bandwidth

Which sodium can't even manage 90 in dual channel so like

[mossy gull]

I think what Bolt wants to achieve is more or less like an "upgradable" buffer.
Where it moves less important textures to the SO-DIMM and keep the most important on the VRAM itself.

[jade scaffold]

Which doesn't work for gaming

It's fine for rendering a movie

Which btw is the actual use case if this thing

[mossy gull]

It doesn't work now for gaming, keep in mind that the industry is constantly changing.
Like it has been only since the Steam Deck launched that we get proper iGPU gaming.

I remember the days we barely could game on 720p on an iGPU, and now we can run Cyberpunk 2077 on 1080p medium settings no problem(tested on my R7 8840U and 32GB RAM)

[jade scaffold]

It won't EVER work for gaming

This isn't arguable

Gaming NEEDS low latency

And that's because your 8840u couldn't beat a 1050ti in a fight

We aren't talking low end here

[mossy gull]

That I don't believe.
There's many times people said it won't ever work and here we are where games actually can utilise stuff from a PC they couldn't before.

[jade scaffold]

This isn't a bottom of the barrel device

[jade scaffold]

Dude if I NEED X data my only options are wait for it or already have it loaded

If I wait everything hangs on it

So everything stays loaded in gpu memory that's needed

Some things you can pull later as you need them but thoes things are critical to stuff not turning into SOUP

The only advantage to having that extra memory is if you use it to run something less memory intensive

For example you could run frame generation off of 4gb of gddr6

So you could have 8gb of fast gddr7 and 4gb of slower gddr6 for example

Maybe it could be used to reduce pop in too though that's not an issue with anything remotely well made

This bolt card is made for render farms. It's made for like rendering a Pixar movie something where memory latency is almost irrelevant

Unlike the 5090 in gaming which really does need an enormous amount of memory bandwidth just to do traditional raster like 800gb/s just for raster never mind raytracing and AI ontop of that which is why it has that colossal memory bandwidth it can easily do 2.4tb/s overclocked

[mossy gull]

Even though Bolt actually wants to make a gaming capable card?

They said they plan on releasing Dev kits to certified devs

[jade scaffold]

No they haven't

It's a card for development it'll be for baking scenes and the such

But 40x seems arbitrary the card has less FP64 and less memory than B200 does

This is what's actually exciting

That being said 12x 5090 kinda pathetic

This one is pretty exciting

If it's actually true like the 10x faster at pathtracing statement is likely a lie and is contradicted by another statement

[fluid quest]

i'm ngl some games from like 10-20 years ago look better and run better than modern games now

case and point, need for speed 2015

accurate shadows and reflections, no smearing, hits high fps on max settings at 4k

if games took rendering methods from then and enhanced them with modern tech from now rather than stuffing temporal BS and software RT into everything, maybe games would actually look photorealistic

[mossy gull]

That's because they don't use TAA and used Raytracing to properly replicate raster shadows.

[fluid quest]

iirc nfs 2015 uses no RT

it's an xbox one/ps4 title

[mossy gull]

TAA is the worst type of anti-aliasing, I have compared different types of AA methods and TAA is the worst, even FXAA looks better.

[mossy gull]

They use RT to replicate raster shadows, they don't have it active in-game.

[fluid quest]

SMAA and MSAA are peak, SMAA is a good middle ground, MSAA is for those rocking high end hardware

[mossy gull]

@fluid quest some devs use Raytracing to properly replicate raster shadows and reflections but don't use it active in game.

A nice example is Dragon's Dogma 2, where they used Path Tracing to properly replicate Raytracing and Raster shadows, but disabled it for the player as it's very buggy and heavy to play with.

[mossy gull]

Depending on the game, MAA also can look pretty good

Although I have no experience with Nvidia's AA stuff

[jade scaffold]

Depends on how it's tuned tbh taa can look acceptable but it can also look straight up like fucking soup

[mossy gull]

Majority of the time it looks absolutely shite, heck FXAA in Fallout looks better, and that game looks like shit

[jade scaffold]

Like I said it depends how it's done

TAA and FSR look hyper ass when not implemented properly

Even dlaa isn't fantastic and that leagues better than either

My goto method if I want balling AA is use DLSS to run the entire pcs resolution at higher than the monitors resolution and then downscaling that really really looks good

[mossy gull]

You basically run at "Super Resolution".
I do that on my 7900XTX system, my 9070XT doesn't have enough VRAM for that in the games I play

[jade scaffold]

Yea DLSS is the best way to do that tbh gets you the best result at the least cost

[mossy gull]

Maybe I should try FSR4 Native + Super Resolution 😅

Anti-aliasing + running higher than native res

[mossy gull]

Why not DLAA + Super Resolution?

[jade scaffold]

DLAA like FSR AA is just TAA

They have loads of smeering issues especially FSR based AA is the worst I have ever seen I honestly would rather take 0 AA atall over it

[mossy gull]

Wait, then why do we have FSR Native and XeSS Native?
Are both just shite?

[jade scaffold]

Yes

[mossy gull]

Then why do they even exist😐

[jade scaffold]

Even NVidias native DLAA sucks

Ment to be better then TAA and have almost 0 or rather actually 0 in NVidias case effect on performance

[cinder lagoon]

https://discord.com/channels/222052888531173386/1403108560472375389

killed my $20 investment from 10 years ago 😬

[timber barn]

Juicy https://youtu.be/N5xhOqlvRh4

[fluid quest]

the actual best method for that is MSAA, except a lot of games stopped implementing it

but thanks to modern deferred rendering, MSAA isn't that usable

in fact that's the issue with most modern games

they use deferred rendering which lets you use things like SSAO and SSR and a ton of other things, especially with RT and GI, but it has the downside of missing out on MSAA and SMAA, both of which are majorly better than temporal based systems.

MSAA relies on forward sampling, otherwise you're just supersampling and downscaling

which is really not viable

real SSAA is very expensive, basically running pure 8k with no upscaling

deferred rendering is also part of the reason for modern VRAM usage, though realistically it can be optimised a ton

it was only a major issue in terms of VRAM when the average person was running <1gb vram

[jade scaffold]

No msaa looks ass

[fluid quest]

MSAA 8x?

[jade scaffold]

Yea it can hurt texture quality's

[fluid quest]

MSAA is just super sampling but with proper optimisation

I mean it can hurt texture quality, but that can be fixed with other methods

that's an issue with downscaling in general

same thing occurs with any supersampling

[jade scaffold]

Yea it can

But msaa looses out 100% of the time to ai super sampling in performance and visuals quality

[fluid quest]

ai supersampling is still temporal

so you have issues of smearing

which arguably looks worse

[jade scaffold]

The issue with NVidias DLAA is it's just trying to smooth over an image it has which looks good on a still image and is temporal

[fluid quest]

i'd say msaa looks better in motion, while any ai/temporal AA looks horrible while moving

[jade scaffold]

Actual super sampling doesn't really need the temporal aspect which is why DLSS 4 looks so good

And it's why using NVidias upscaling looks way better as an AA method

Would genuinely take 1080p upscaled to 1440p with the transformer model over taa fxaa or even msaa if for only for the fact it'd run better than any of them with a visual impairment so insignificant that in any actual movement you can't tell at all

Never mind upscaling up from native then squashing it back down to native which looks just as good as msaa with a fraction of the performance cost

[fluid quest]

isn't it still temporal though

so you get a ton of artifacting

[jade scaffold]

It is but you don't get tonne of artifacting

[jagged star]

Yeah it is that's why it still looks blurry in motion

[jade scaffold]

DLSS4 practically killed most of the artifacting

[fluid quest]

counterpoint, now you need a 50 series gpu

[jagged star]

And has fiz but you can't remove temporal AA in modern games anymore because graphics are designed around it

[jade scaffold]

No you don't

DLSS4 works on any GPU 2000 or after

You don't get frame gen

[fluid quest]

even ray reconstruction?

[jade scaffold]

But you get the newer version of upscaling

Yes

It's only frame gen you don't get

[fluid quest]

bruh

alr

I honestly also just

don't understand modern AA

like

why is it all vendor-dependent

[jade scaffold]

You get 2x frame gen on 4000 4x on 5000 they might be able to port 2x to 3000

[jagged star]

Transformer model is only 4k and 5k

[jade scaffold]

No it's not

[jagged star]

Without performance hit

And frame gen is 4k+

[jagged star]

And 4x frame gen is 5k +

[jade scaffold]

No you don't get any performance hit from transformer model either

Infact transformer model is vastly more efficient

Which is why they said if they can optimise it further they might port 2x FG to 3000

The transformer model also scales better with more training so it'll only get even better over the next few years

[jade scaffold]

Oh okay mb must have gotten it mixed up

Not that it matters because DLSS4 claims that it's performance mode has altleast equivalent quality to dlss3 quality mode

So better performance at equivalent quality would be the more accurate statement

[jagged star]

That's cap

But it does make quality usable at 1440p

No reason to have it off unless your 1080p anymore

[jade scaffold]

It's not I think GN did a visual comparison to test the claim

[fluid quest]

I always run DLAA, not DLSS

[jade scaffold]

DlAA looks worse than DLSS

[jagged star]

I don't got the horsepower for that

[fluid quest]

my 3070 wants a break lmao

[jade scaffold]

Way more smeering Vs using dlss4 on dlaa

[jade scaffold]

Even at 1080p now it's starting to get useable old DLSS I'd never recommend upscaling from anything lower than 1080p but dlss4 is so good I've seen people go from 540p to 1080p and it somehow looks weirdly good

[jagged star]

cant do it still looks like taa blur imo

never understood how everyone kept saying it looks like native when 2.0 came out

untill i realised they mean against native +taa

[jade scaffold]

Yea with less data DLSS is less accurate

The higher the base resolution the more accurate it is

[cerulean helm]

Just picked up a mini PC Beelink S13 N150 with 16gb of ram and a 500gb SSD for $168. Should be here tomorrow. Way easier on the electric bill to use to mess around with as a homelab than My ancient Proliant Dl380 G7 with dual X5670 cpus.

[cerulean helm]

I'm sorry what?
https://www.youtube.com/watch?v=ltgyS8oJC8g

[fluid quest]

geez, I thought my system was old with 2nd gen epyc, but dangg

that's a 15 year old cpu

mine's less than half that

6 year old

honestly a 180 watt tdp isn't that bad

thought it was higher

(for my cpu)

and hey I have an upgrade path kinda

i'll go up to a 7742 if I can ever afford it once ebay prices drop

that thing's a multicore beast

but the cpu's still like a grand

meanwhile I picked up mine for like 200 bucks

[cerulean helm]

It idles, just idles at 140-160W. Thats just with Proxmox installed with 2 ssd drives in the bays.

[fluid quest]

i'm gonna be honest i've never checked my server wattage and i'm too scared to, I'd rather live in ignorant bliss

[cerulean helm]

My current electric bill with this hot ass summer without turning on that DL380 at all last month was $300. I don't even want to think what it would be running that thing.

[fluid quest]

yeah my bills in the middle of winter even with solar are like 250/mo

for some reason i've yet to figure out our power usage at midnight is like 1 kWh

or even higher

even at like 3 am we're using 500 Wh

desparately need to hunt for the cause

even before getting the server it was that high

no clue why

[cinder lagoon]

Feel that - my Haswell CPU got 4x 4,6GHz with multithreading tho :P

(I'm sure any reasonably modern laptop would take it apart in a drag race)

[cerulean helm]

Update on the Jetkvm. By far the easiest, simplieslt network kvm to setup. Just installed my 2nd one on my new mini pc. Takes less than 2 minutes for everything to just start working.

For $70. Its a steal for how powerful it is.

Okay, this Beelink S13 has Windows 11 pro pre installed. Was not expecting that at all.

[cerulean helm]

Kinda annoying as I am planning on installing Proxmox and a a couple VMs.

[cinder lagoon]

I mean.. you do have the licence now

Sync it to your microsoft account or write it down somewhere, it's gonna come in handy somewhere

[cerulean helm]

Fair point. Still updating the Beelink and the Jetkvm software so, it will be a minute before Proxmox is even installed.

[cinder lagoon]

Speaking of Proxmox - that's like the main free hypervisor, right?
I wanna restart my "Server" but I'm not sure what OS to choose.
TrueNAS?
Proxmox?
Whats that other one called?
It's mainly intended as a NAS but also game server.

[cerulean helm]

Proxmox, I have installed on my DL380 G7 server, and have a couple VMs on it. A windows, Debian, and I think a different Linux fork. You would install TrueNas via a VM. But, honestly if its a mssion critical NAS, I would just install the NAS OS and nothing else.

[cinder lagoon]

Not at all lol

Just media storage - doesn't it cause problems with virtual hard drives?

Like I'd want direct hardware access for monitoring etc - probably would need to pass through a SATA controller then

[cerulean helm]

No idea. I'm very new to Proxmox.

[cerulean helm]

Well, the Beelink S13 mini is up and running, I cannot believe hopw tiny this thing is, lol.

[cinder lagoon]

Well, it's a mini-desktop, about as big as the minisforums

[cerulean helm]

Well. I have royally fucked my ZFS pool in truenas. I deleted an unused folder that had been hidden, found, added to the smb share but couldn't delete because permissions for whatever reason. So I purged them via the shell. And now Truenas can't even load the ZFS pool

I don't even understand how this happened.

But sweet if I lost ALL my data. Like everything, all my backups from the last 10 years.

[fluid quest]

proxmox is great, if you wanna run any NAS stuff run it in VMs, gives you a ton of choice

[fluid quest]

you can pass through individual drives

I do that for my current setup, I have a few HDDs for NAS storage, + a few SSDs for VMs

pass through the HDDs, use a virtual disk for OS storage

[fluid quest]

proxmox is the main hypervisor for homelabs, not much reason not to use it as it gives a lot of the features needed in a lab (mainly missing out on some stuff for enterprise-level clustering), and it's FOSS unlike most other hypervisors

[cerulean helm]

Hot damn. Repaired the ZFS Pool and got all my data back. Holy shit that was the most stressed out I have ever been.

[cerulean helm]

Still missing 2tb os storage in the zfs pool but uhh. After what I just went through, I may just leave that well enough alone of trying to fix.

[surreal moss]

If it's primarily a NAS first then I would run TrueNAS.

[fluid quest]

eeh i'd do truenas in a vm on top of proxmox still

[surreal moss]

Proxmox is good for VMs and can do NAS stuff and TrueNAS is good at NAS stuff and can run VMs

[fluid quest]

especially if it's their only server

that way you get the benefits of both

[surreal moss]

Why do TrueNAS in a VM though?

[fluid quest]

you get all the benefits of proxmox as a hypervisor, with all the benefits of truenas as a NAS

[surreal moss]

hmmmm

that's... an interesting take

[fluid quest]

especially if you pass through your drives rather than using vdisks

which you should anyways

for a NAS vm

[surreal moss]

Sure, but that's layering for not a lot of gain imo

I'd just pick one and go with it

They have enough feature overlap that I don't think it's worth doing them both like that

[fluid quest]

there's a ton of gain, you get LXCs, virtual networks, proper VM ballooning and memory management

[surreal moss]

In ProxMox, yeah. But for a few game servers one what is essentially a NAS it's overkill

[fluid quest]

who said game servers?

I have like 20 vms on my server

truenas would suck for that

[surreal moss]

Chaos did

[fluid quest]

and anyways, even with that

I personally have all of my game servers on individual vms

[surreal moss]

Congrats?

I would use Prox for that

[fluid quest]

makes memory management and port management way easier

[cinder lagoon]

If you learned ANYTHING from this:

ALWAYS keep an offline backup somewhere.

[cinder lagoon]

Does this also pass through stuff like SMART data or just the blockdevice?

[fluid quest]

passes through everything, as if it was a bare metal drive

[cinder lagoon]

And in this case: do as I say, not as I do lol

[cinder lagoon]

Ouhh ok nice

[sonic meadow]

raid isnt backup, and one copy of something is zero copies

(im adding to what you are saying, not contesting it)

[cinder lagoon]

Well yea kinda but an OFFLINE backup that's connected once a month would have saved my ass so many times

Ahh

[cinder lagoon]

For me it's

• raid against hardware failure
• offline backup against software (or human) failure
• off-site backup against catastrophy, if your data is valuable enough

Full 3-2-1 is not really viable for privat use

[sonic meadow]

local pc
external/portable hard drive
cloud (google drive, whatever apple has, onedrive)

[cinder lagoon]

Yeah cloud storage do be expensive tho and if it's only used as a backup - give a friend a hard drive every half year or so

[sonic meadow]

(those ones I listed are all free, though do have capacity limits I think)

[cinder lagoon]

Yeah

• gdrive 15GB
• iCloud only 1TB per apple device or somthing
• onedrive 15GB in free tier bc microsoft wants your data and lets you "backup your pc" (3 mostly unused folders by default)

If you hava a homelab - you've GOTST do have more than 15GB xD

(and probably no apple device)

[sonic meadow]

oh definitely, but for personal data backups are likely to be fine (documents and photos)

[cinder lagoon]

Important documents - yes.
Important photos and videos - nope - got over 500GB unwillingly encrypted

[sonic meadow]

I didnt say anything about videos

deliberately didnt in fact

[cinder lagoon]

Well the videos were like 8MB a piece back then but yea

[cerulean helm]

This is what I am currently building right now. A Beelink mini S13 with a DAS hooked up to it, and will once a month back up my main NAS and then shut back down. Will probably use Proxmox and host TrueNas in a VM on the Beelink.

[cerulean helm]

Easy part is being done with the Proxmox install. I still need the DAS and drives which won't be for a bit as the DAS is relatively not expensive, but the drives... ouch.

[cinder lagoon]

Yeah high capacity drives are expensive

Wondering wether there's an easy solution for getting a good workflow for

• Validate file integrity
• start/connect the offline system
• back up
• disconnect/shut down

[sacred seal]

I have too much data to economically back up, so I just directly store it. No fancy file systems, just normally on spinning rust. That way if somethingg goes horribly wrong I can bring it to a recovery service and have them pull the data off

I always hear of these complicated redundant systems nuking themselves and rendering all the data unrecoverable. It is pretty hard to physically kill a mech drive to the point where it can't be repaired

[cinder lagoon]

Thats no good in case of user error or natural catastrophy tho

A raid1 would be still advisable

But I know you have a 12TB hard drive in your daily driver soooo...

[sacred seal]

14tb, but yeah. Honestly I don't think raid 1 would be a good solution because the raid array messing up can be a source of data loss. Even if I was to delete everything off of my hard drive I could still recover most of the important files

And because I have so much data and my drive is not that fast (and is noisy), I have some protection against malware

[cinder lagoon]

At least there's that

[cinder lagoon]

Stuff gets difficult if you accidentally overwrite stuff

[sacred seal]

Oh, of course. But that is what the recycling bin is for. If I accidentally delete something I can restore it within 30 days. That is something that raid 1 would not protect me from.

Raid is great if you need the uptime

You can rebuild your array quickly

[cinder lagoon]

No but professional data recovery is more expensive than another drive

[sacred seal]

Yes, but drives don't normally die overnight

You can normally pull most of the data off of a dying drive

My drive will die at some point, but I have contingencies in place for when that happens. It's a server drive and I mostly use it for bulk storage so I am hopping that is not for a long time and I will probably have replaced my drive with something newer and bigger by then.
A raid array can just decide to do something silly and then a bunch of data is just inaccessible.

Unless you need the uptime, I really think raid and similar tech is more risk then benifit.

[cerulean helm]

Looking for a decent DAS as well. I could use a 2 bay but would need very high capacity as it would have to be a mirror setup. A 4 drive would be better, with one drive as a hot swap spare. But that would be in raidz1, not sure if thats a good idea or not as this is going to be my failsafe oh shit storage backup.

[cerulean helm]

No kidding. Lost my old Synology NAS the other year and a clean room wanted $4400 to get the data off 2 drives. Just 2. They are currently sitting in a box in my closet.

[surreal moss]

If I’m worried about data integrity I’d prefer to have a RAID configuration anyway. Something that allows a drive to fail and still have things functioning. And then whatever else on top of it. I haven’t had any issues with it to make me gun shy

[cinder lagoon]

Be sure to use an actual recovery firm then - I've heard too many horror stories from smartphone "repair shops" that opened up a perfectly recoverable drive and voided all data

[sacred seal]

Oh, deffinately

[sacred seal]

Behold, the computer of Theseus.

Other then the motherboard and case, every part has been replaced.

[fluid quest]

makes sense, the amount of work that goes into data recovery is crazy

[jade scaffold]

Looking at buying some Austrian audio Hi-X50s

[sacred seal]

This is a new one...

Happens when I first launch chome, goes away if I refresh

[mossy gull]

Welcome to Google

[sonic meadow]

so apparently i never allocated almost 100gb of the ssd in my laptop

[sonic meadow]

Got an old case with 2x120mm on the side panel... Kinda want to put a radiator there. Just ignoring that it would probably interfere with any modern gpu

[mossy gull]

Depends on how you place the GPU

[sonic meadow]

Grrr, photo is rotated. That's the front panel at the bottom, and back panel at the top of the image

You can see the GPU currently in there on the left

[cinder lagoon]

This is actually good.
SSD controllers often read the MBR/GPT to find unallocated regions and use them as overprovisioning space to extend the lifetime of the drive.

[jagged star]

its probs by design as ssds will slow down if fully full

[cinder lagoon]

Who told you that?
Thats absolute bogus.
was marginally true-ish with HDDs, not with SSDs tho.

[sonic meadow]

Considering I did it myself, nahhhhhh no design just stupid 😆

What probably happened was I shrunk it to be able to clone it cleanly, then never expanded it back to full size