off-topic-tech

[night girder]

For example, data is not always deleted from your SSD, but instead marked as "can be overwritten". This reduces one write. Because it would have been overwritten with empty data. And then overwriten with new data. But now it will just be overwritten with new data. And all this information has to be stored somewhere.

[minor pawn]

Is the Kingston XS1000 good cause i can't find it on the spreadsheet

[dire igloo]

It's on there but no data.
Looking up a review for it tells me it used SMI2320 controller which is DRAMless

[minor pawn]

Wich controlles do have dram?

[dire igloo]

SM2258 and SM2262 can come with DRAM, Phison E18 does but it's unlikely to be seen in an external drive
Just some examples I found

Phison S10

Phison E12S

Hynix Beetle is an external SSD with DRAM

[gleaming shadow]

sooo, this aint normal right?

btw only game thats happing with

[cobalt ivy]

💀i can tell satisfactory loves you so much it decided to become more colorful

[night girder]

I see nothing wrong.

[cobalt ivy]

Yep. Just standard satisfactory love

[dire igloo]

I see nothing.

[gleaming shadow]

no seriously, last times ive seen this shit happening, ment a GPU died..

[cobalt ivy]

💀 as long as it's only game which loves you that way your GPU should be alive

[dire igloo]

Can be driver corruption or smth specifically triggered by that game

Try DDU

[gleaming shadow]

its certain vieuw only,

not permanent looks fine

[night girder]

Two totally different games 😂

[gleaming shadow]

i should check unreal engine game then 😛

[cobalt ivy]

Try something which will use all your graphic and is called Factorio-

[gleaming shadow]

you haz a map i can boot up?

[cobalt ivy]

I have no idea which games even use UE except bodycam and Satisfactory lol

Never really cared about game engine

[cobalt ivy]

💀big

Very big

[gleaming shadow]

also seems fine, and not alot of vram

[cobalt ivy]

Hmmm what GPU you have?

[gleaming shadow]

7900xtx

[cobalt ivy]

Nice

[gleaming shadow]

now you know my stress 😛

[cobalt ivy]

Im running RTX 3060TI myself

[gleaming shadow]

i ahd that, founders upgraded to this

[cobalt ivy]

3 fan edition thankfully

Since i seen some 2 fan versions

But overheating.exe lol

[cobalt ivy]

Mr that factory in factorio is bigger than my biggest base in factorio

[gleaming shadow]

had a founders, no issues with heat

its even bigger then it looks, ive downloaded a map 😛

[cobalt ivy]

Yehhh im overclocking a lot

[night girder]

👆 + clean install game + clear shader cache if possible.

[gleaming shadow]

i did DDU earlyer this week on DX12 i seem to have crashes

[cobalt ivy]

So 3 fan edition just falls from sky for me

[gleaming shadow]

they advised to go to DX11

[cobalt ivy]

Mr i feel like satisfactory might just not like you for some reason... Did you ever kill any

[gleaming shadow]

ofc 😛

[cobalt ivy]

....hmmmmmm

I never killed one but im telling everyone i killed 5 of them....

But i killed a lot of spiders atleast-

[gleaming shadow]

havent been in caves yet

[cobalt ivy]

Tip: if you have tractor they won't attack

If you are in it atleast

[gleaming shadow]

fucking hate spiders, they jump to much xD

where can i find factory save files again?

[cobalt ivy]

💀hmmmm no idea

[cobalt ivy]

True. First thing that killed me in this game was spider

[gleaming shadow]

fall damage xD

[cobalt ivy]

💀wish it was.

My friend fell from edge of map...

Then i threw 5 tractors off map

[gleaming shadow]

oooo i got factory using 3 gb vram now

[cobalt ivy]

Why? Im bad driver

[cobalt ivy]

HOW MUCH?!

THIS IS WAY MORE THAN ANYONE COULD HAVE THROUGH!

[cobalt ivy]

....wait it's mod or im just stupid

[gleaming shadow]

map editor

ive never played factorio

ive downloaded a benchmark map thingy, and didnt understand how it worked

[cobalt ivy]

I did.... Using funny website to get it for free

But no mods since you can't log in is painful

[gleaming shadow]

yea me to, i tried a "Demo" version of the game

i may or maynot have more, o wait i can boot up doom eternal and see

[cobalt ivy]

I try all games on cracks-

Before buying them

[gleaming shadow]

me to, and if i like it enough i get the full version

Doom using more 😄

[cobalt ivy]

Yep.

[cobalt ivy]

Im using FNAF cracks since no multi is there anyways-

[gleaming shadow]

FNAF?

[cobalt ivy]

Five Nights At Freddys

[dire igloo]

Need I remind you that discussing illegal activity like piracy is explicitly outlawed by the rules of this server?

[gleaming shadow]

lets, get back off topic

[gleaming shadow]

Doom metrics

main menu,

[cobalt ivy]

I love that first doom is in reality 2d game which is reason you can't point up and down 💀

Atleast first one,..

[cobalt ivy]

Dang

very high usage💀 totally

[gleaming shadow]

its using 8gb vram!

[cobalt ivy]

Wait what-

dang

[gleaming shadow]

ever tried?

unobtimized....

lags ALOT

[cobalt ivy]

Hmmmm on crack? Yes.

[gleaming shadow]

its was free on epic games

[cobalt ivy]

FS22 was free on epic too

Which was nice

[gleaming shadow]

you dont say

[cobalt ivy]

are you free games hunter?

since i am when im not lazy while waiting on "Wrench" discount on steam

[gleaming shadow]

on EPic yes.

[cobalt ivy]

I hate The Sims 4 DLC prices 💀

[gleaming shadow]

on steam.., well

[cobalt ivy]

Ehhhh

[cobalt ivy]

There are nice discounts

80% on train sim world 4

[gleaming shadow]

you dont say!

[cobalt ivy]

Where did steam show price of acc-

[gleaming shadow]

https://steamdb.info/calculator/

should realy move to offtopic..

[cobalt ivy]

Hmmm not bad 500pln

On lowest prices

[cobalt ivy]

Today price gives me triple that

[cobalt ivy]

Shhhhh

As long as mods don't have problem we are fine here

[gleaming shadow]

on DX11 it seems fine, this happens i think with LoD

[cobalt ivy]

Never really played more games than Enlisted on DX11 lol

[gleaming shadow]

i still play Everquest 2, it got updated to DX11 like 2months ago

game is 20 years old lol

[cobalt ivy]

you Talking to man which has separate device just to play settlers 💀

But none of these games run DX11 i think

[gleaming shadow]

dunno tbh

[cobalt ivy]

Never really cared if DX11 or 12 unless i wanted to use Reshader

[sand saddle]

Fuck me, I’ve joined the Apple ecosystem. Got an Iphone 12, apple watch series 6, airpods pro 2 and airtags now.

[cobalt ivy]

NOOOO

WHYY

NOKIA 3310 WAS BEST CHOICE NO MATTER WHAT

just noticed i had caps on lol

[sand saddle]

It just happened…
Got the APP2 for quite cheap
Got an old Iphone X from my mom/sister when I broke my Samsung.
Got airtags for the battery life.
And watch now cuz uhhhh, perhaps I’m stupid.

[cobalt ivy]

....nokia 3310 better

[languid gulch]

i definitely miss the days when you could drop a phone and not have a mini panic attack

[cobalt ivy]

I miss days before my reddit addict got me 💀

[soft bloom]

cough every game is 2d since displayed on surface of minotor

[soft bloom]

its the best game

[soft bloom]

which gives more horror: the fact of doing that inpublic or expected charge to repair it?

[cobalt ivy]

It isn't what i meant lol

[tough owl]

the inverse of a good pipeline

[jagged snow]

Finally caught my PC resetting bios

[twin dew]

And HMB doesn't work over USB.
Host Memory Buffer is the SSD using main computer RAM as on-PCB DRAM replacement, but that only works over PCIe.

Nah, 64MB is the max the HMB can be AFAIK, 32MB is the norm currently.

HMB drives work mostly like completelty DRAMless drives, but they can put part of the on-flash redirection table into that HMB, lowering writes needed.

[twin dew]

That 1GB per 1TB is what is needed for completely flat redirection table, where each LBA has its own full entry without any compression or space saving tricks.

[languid gulch]

but it's all 1s & 0s!

[twin dew]

Just way too high latency over USB, and no DMA access.

Could be done, but so slow to be worse than useless.

And I did get it was a joke, just had to still explain.

[languid gulch]

it would be fun to see just how far you could adapt from one "thing" to another

like what would it take to have a direct hardware connection to a drive from the 70s

[twin dew]

"Nice"... this share NIC line can work on BOTH of the 10G ports, not just one of them...

I was just hoping to leave the shared port unused to protect from any mistakes.
As having BMC/IPMI accessible to even normal internal LAN is giant attack surface.

I had accidentally re-enabled it in BIOS after disabling it in IPMI management.

And then to reset the IPMI config, as I made first firewall rule to allow all from specific IP, and then another rule to block all IPs...
But the second block is now overriding the first rule too.

Or lets first see if I can use local IPMI tools to just reset the firewall.

Yeah, seems it should be possible.

[night girder]

There was a sdd with 200 MB HMB, might been a typo.

[twin dew]

Or that 64MB has just been the largest being used so far and there isn't real technical limitation.

[night girder]

WD Blue SN580 1 TB SSD

N/A, but 200 MB Host-Memory-Buffer (HMB)

[twin dew]

Spec doesn't seem to set a maximum.
Drive just has "minimum I need" and "maximum I can use" fields it gives to the computer, and if the computer doesn't give back allocation in that range then HMB isn't used by the drive.

Two bytes for the setting, in pages (4kB usually).

So about 256MB max size that can be asked for.

65535 pages.

Ah, and the minimum has 4 bytes for it, and you can set the max to 0, where it is ignored.
So the max for the minimum is gigantic.

[night girder]

Baldur, we talked a while back about Kaspersky. They pulling off the American market. (source: www.bbc.com/news/articles/cyr7ex16p32o)

[twin dew]

Not much else they can do while the case is in court if they didn't get injuction to stop the order from taking effect while the suit is in progress.

[night girder]

It effectively barred downloads of software updates, resales and licensing of the products from 29 September, while new business was to be restricted within 30 days of the announcement.

So, everyone who had Kaspersky in the US is fucked and needs to buy new one

Anyway, I don't run Anit-Virus anyway. Just the standard windows one, but nothing extra.

I have bad experiences with them.

I guess they want to push more people into home grown too like Malwarebytes.

[twin dew]

That ipmitool firewall functionality was just for the KSC interface commands, not the IPMI network firewall.
But got in by enabling IPv6 support for the management LAN in BIOS and connecting via that.

[edgy hazel]

@dire igloo what do you call it when a programmer fixes the front of a boat ?

Bugfix

[wanton orchid]

when you say ipmitool what tool are you talking about exactly ?

I struggled with ipmi tool to connect
and even more to get documentation on the behavior of the connection for such tool

all I know is ssh

[twin dew]

using ipmitool and the like from inside the specific computer, which doesn't need connection settings at all.

KSC is the connection method between the BMC and the local OS.

As I had blocked all LAN traffic to the BMC over IPv4 with firewall rules.

[wanton orchid]

I don't know why it never seem to work for me

[twin dew]

That can be blocked with config.
Or if you mean using IPMITOOL or like from another computer, then it needs several parameters for the connection.

[wanton orchid]

btw when shutting down redfish, the whole board needed reset
I don't know why

I really hope next boards have less clutter implementation

[twin dew]

But the LSI controller silicon sits at little under 80C, so completely "fine".

While doing DD block copy from one HDD behind it to another.

Yeah, 80C for die is completely ok it seems, as the spec is 110C for thermocouple on the sink.

The measured heat sink temperature on either heat sink should not exceed 110 °C (230 °F). If the temperature is greater than 110 °C, try the board in a different slot, provide more airflow, or make both changes

LSI has replied that chip sensor max is 115C in operation.

But probably would be good idea to replace the stock gum with more TPM.

Someone replaced with normal paste and moved from 93C to 58C.

Now that the card has been tested somewhat and I can read the temperature.

[wanton orchid]

what are you talking about exactly

you started talking temp out of nowhere

[twin dew]

Temperature for LSI SAS 2308 based SATA/SAS controller card.

PCIe 3.0x8 to 8x SATA/SAS 6G

[wanton orchid]

ok

btw for hdd do you agree sas does not provide any advantage over sata when smartly used by the storage controller ?

(hdd side ofcourse)

[twin dew]

SAS just gives some extra features, if the HDD also is SAS.
And allows for real port multiplexing.

But the cost of that multiplexing is excessive for home usage.

[wanton orchid]

but what features provided are still in the scope of hdd storage
because all I seen about hdd partitionning and security
it is for retarded consumers that got money to spend on useless things basically
as having acl inside the hdd does not realisticly provide any real security or benefits if not the opposite actually

there exist error reporting and correction extensions for sata

[twin dew]

SAS is purely Enterprise tech.

And mainly for that multiplexing and dual-porting.

Two controllers/computers talking to same drive via two channels, for controller level failure tolerance.

that dual-porting thingie.

[wanton orchid]

so it's only for multiplexing actually I see

[twin dew]

And to use one x4 cable to connect 16+ drives to same controller port for example.

Which is why you got SAS3 with 12G per lane for example, when both SATA3 and SAS2 are just 6G per lane.

[wanton orchid]

so 12G is 4 hdd per lane

[twin dew]

No, 12x4 per connector.

[wanton orchid]

I said per lane, not per connector

if connecotr is x4, then it's x4x4

(16 as you said)

[twin dew]

That was just example number.
That controller that I have has 2 physical ports, each giving 4 lanes, each lane 6G.
And can address little over 1000 drives if wanted.

[wanton orchid]

for few MBps per hdd

[twin dew]

Not realistic use except for some weird archiving setups.
But could.

If you chained enough expander chassis behind those two ports.

But I'm just using it as 8-port SATA controller.

As the consumer ones are pretty unreliable in comparison, as the chips are crappy.

Switched from RAID-mode firmware to IT (Initiator-Target) firmware that just passes the drives as they are.

[wanton orchid]

I knew that but somehow it didnt work back then
maybe I had faulty kernel or faulty bmc config, or both
but I got it working just right now
it works fine (at least to show sensor info and stuff)
thank you I guess

[twin dew]

That card is using the worst kind of glue-like OEM stuff...
Need to see how I can get it to let go without pulling the chip with the heatsink...

[wanton orchid]

rotating ?

[edgy hazel]

https://tenor.com/view/spinning-monkey-monkey-spin-hd-monkey-spin-gif-25391894

[wanton orchid]

usually provide the most nondestructive friction

[twin dew]

Problem with that are the excess thermal stuff that surround the chip and are attached to the heatsink still.
But got it off.

Whitish acetone resistant gunk.

[twin dew]

64 C

Up to 70C now, but need to shut the computer down and put it back in right place.

[twin dew]

Now to read the source for the lsiutil 1.72 I found on the Internet.
As you cannot get it from Broadcom anymore.
And their current utilities don't allow reading the temperature from IT-mode cards, only from RAID-mode cards.

[twin dew]

HDDs staying at around 35C while doing 1MB sequential zero write with DD.
And controller at 73C.

[jagged snow]

Don't forget to check prime day deals y'all

[tough owl]

That's today ☹️

[jagged snow]

Yep

Pixel 7a for 230

[tough owl]

I got a 7p for 280

I love my 7p

I need to look at the sales on tablets with pencil support

[tough owl]

Seagate is selling AI branded hard drives

[twin dew]

For long time for the Skyhawks.
Base Skyhawk is video surveillance storage optimized drive.
The Skyhawk AI adds firmware changes to work better with "AI" analyzis of those recordings at same time as the actual writing is ongoing.

Because base Skyhawk is optimized for writing multiple static speed sequential writes at same time.
The AI part adds significant amount of reads into the mix at same time.

[tough owl]

But it's not AI

[twin dew]

OKk, might not even be that, as there is no overlap in capacities.

So can just be marketing gone wild.

[tough owl]

Marketing gone wild

[twin dew]

Skyhawk AI is optimised for use in video analytics and imaging applications. Designed with advanced features, it is suitable for use with AI NVRs, servers, and appliances with AI analytics and deep learning capabilities.

Non AI:s are listed to work with up to 64 HD Cameres.
The AI:s are listed to work with up to 64 HD Cameras and 32 AI streams.

So part is real, part is marketing.

Because that real-time machine learning analysis changes the load profile from just storing multiple constant video streams.

[night girder]

https://www.youtube.com/watch?v=od9lve4SvqI

[tough owl]

Silly apple

/9to5mac.com/2024/07/16/apple-used-youtube-videos/

I assume this is a direct relation

[night girder]

Yes it is 🙂

and it's against YouTube’s terms and conditions.

[tough owl]

Apple v Google here we go

[charred pewter]

fuckemboth 😄

[night girder]

Not really. Apple smart.

It’s important to emphasize here that Apple didn’t download the data itself, but this was instead performed by EleutherAI. It is this organization which appears to have broken YouTube’s terms and conditions.

[tough owl]

so Apple used a proxy company 💀

[night girder]

a lot used that company.

Nvidia and Anthropic.

[winter path]

Hey does any have ideas for the best laptop for school and light gaming? Under 500$ would be best.

[silk jay]

I am not that up to date with laptops so the best advice i can give is:
-slim to carry it easily
-must have a separate gpu, otherwise you can't game
-make sure the cpu and gpu are cooled separately to ensure longevity of your laptop

[winter path]

Can you recommend a specific company that’s best? Right now Lenovo look like the best quality.

[silk jay]

For low budget Lenovo is indeed a great pick, however i recommend looking at some (older) models of ASUS TUF F15, since they have good cooling. They can be a bit ticker than other laptops, but that is to hide the cooling. I am not sure if the prices are the same for you (i am Dutch)

[night girder]

ASUS TUF GAMING F15 is $739.00 atm on amazon.

[steady spindle]

exuse me, anyone wanna talk while I play satis

am bored

[night girder]

#satisfactory ?

[steady spindle]

yessir

[silk jay]

Depends on GPU

[winter path]

I have a real pc, the main purpose for the laptop will be school. Although I would like to be able to play the optional game. Currently I have a hp that only lasted me 2 years with not great price to performance.

[silk jay]

you mean vc?

[steady spindle]

yeah

[silk jay]

Sure

[steady spindle]

YIPEE

[gilded helm]

If you want a laptop for school, prices on AMD Phoenix and Hawk Point stuff are probably about to drop once Strix comes out. I'd wait for the last days of July or first week of August.

Depending on your definition of light gaming, no dGPU might be fine.

You can watch here if you want something with a dGPU, but a quality unit under $500 will be hard.

gaminglaptop.deals/

[winter path]

I’m hoping to get something during prime day. Currently I’m looking at “Lenovo IdeaPad Business Laptop 2024, 15.6" FHD Display, AMD Ryzen 3 7320U Processor(Up to 4.1GHz), 8GB DDR5 RAM, 512GB SSD, Wi-Fi6, HDMI, Fingerprint, Webcam, Windows 11, Abyss Blue, KyyWee Accessory” (Amazon)

[gilded helm]

Reddit's /laptopdeals is also solid

[winter path]

Any opinion on what’s pasted?

[gilded helm]

Not really, I have a Lenovo and like it, but it's a Legion. Each line is so different that you can't really judge a whole company across their different lineups.

The specs are fine, but if the keyboard feels mushy or the screen is terrible, then you'll be itching to replace it. Stuff like that only shows up if you can find reviews.

[winter path]

Oh ok thank you for your help have a good day

[languid gulch]

i'm still mildly allergic to lenovo because of the whole "sure, we'll gladly put chinese spyware in our stuff"

[sage pier]

Hey balder, Why is my ethernet so inconsistent when the eero pod is connected?

[maiden coyote]

because eero is shit.

[dire igloo]

dGPU is not that important for light gaming actually.
Current gen iGPUs are surprisingly performant

[dire igloo]

Oh, some1 already pointed that out

[twin dew]

But for IGP in gaming use, 16GB of RAM is minimum and 32GB is strongly recommended.

[twin dew]

Solid for basic desktop use, not suitable for gaming or other high performance requiring tasks.

That 8GB of RAM is pretty small today, usable but would probably be the first thing to start limiting.
The CPU is Zen 2 based, not Zen 3 or Zen 4, with 2 CU RDNA2 IGP.
The IGP performance is in line with 10-15 year old low end desktop GPUs, so not really suitable for gaming.

[languid gulch]

yea i mean a 1050 or a 710 is what IGPs are comparable to, so don't expect any miracles

[twin dew]

Nah, 1050 is 478% the performance of that IGP in TechPowerUp listing, but that is just based on numbers and not testing.

The low-end of 1000-series is the 1030, which is 241% performance in the same metric, not 1050.

Near same thing in my 7800X3D (almost same specs, different power limits and double the ROPs etc.) can do 20FPS in Satisfactory with everything on low at full-HD.

The theoretical max of that 610M IGP for FP32 is 486.4 GFLOPS, when 1050 non-ti can do 1.862 TFLOPS and Radeon 6600 non-XT can do
8.928 TFLOPS.

The RDNA3 IGP in 7540U can do 2.560 TFLOPS, the RDNA3 IGP in 8840U can do 8.294 TFLOPS

As comparison for current good IGPs, rest of the 7#4# and 8#4# APUs go between those two.

Those 7#2#U:s were designed as Chromebook CPUs.

And are fine/good for office worker type workload, but not really meant for games etc.

School laptop: probably perfect fit.
Gaming laptop on the side: no.

[languid gulch]

remember the talk a few years ago how monitor demands were gonna be high enough that a workstation & a gaming pc were gonna kinda merge 🤣 good times

[soft bloom]

i never thought that 15W cpu in laptop can feel that hot

and heat up whole room

[languid gulch]

yea, when i jumped my gpu up a couple tiers the room definitely got hotter

[edgy hazel]

Every watt you put into your PC you get out as heat

[languid gulch]

true, i was mainly surprised at how much higher this one idles at

[soft bloom]

you are talking about hundreds of W, here its just 15... idk what is wrong with i7200u

[sterile plinth]

yall i got some big issue

whenver I start my PC, within 5-10 minutes of it booting up all the device drivers will get wiped

restarting fixes it but after 5-10 mins same thing

it will work like normal then sound drivers get wiped network drivers get wiped

literally all of them

when i open the sound control panel it says no audio devices found, internet stops working

literally any service

happened again

the funny thing is I can hear my frined on VC, but there are no audio devices

the moment i reload the app audio is gone, and all new apps i open have no audio

@twin dew help 🙏

[twin dew]

No idea what could be going on.
Reinstall the OS?

[night girder]

😂

Seems more like OS being corrupt/bugged/broken than the drivers magically being uninstalled.

Unless you download something naughty and got yourself a funny virus that focuses on deleting drivers.

[sterile plinth]

One more thing, regarding the constant crashes I said about..

Sadly ram is not the issue, I ran 2 passes on memtest86 all flawless

What else could be causing them?

I get a variety of 5-6 different BSOD error codes a few times a week

[dire igloo]

Get WhoCrashed

[sterile plinth]

What's that?

[dire igloo]

A software that looks at the crash reports, analyzes them and gives you a rough idea of what might be the issue

[sterile plinth]

uhh these should be all

[twin dew]

So Windows kernel getting corrupted.
And the RAM sticks aren't broken.
So either CPU or RAM settings.

Or like that says, broken driver which doesn't cause instant crash.

[sterile plinth]

could it possibly be the overheating

the KMODE_EXCEPTION error is by far the most common, and it says hardware failure

[twin dew]

"Could be either software or hardware" as the first line,.

[sterile plinth]

clearly not software as i have formatted way too many times to count

but how do I go about finding which hardware part is responsible and fixing it?

sure, could be cpu or could be ram

[twin dew]

I said RAM settings.

[sterile plinth]

so, XMP?

[twin dew]

As like said before, Memtest doesn't find all unstability from those.
So yes, XMP.

Set the RAM to all-auto and see if you still get issues.

[sterile plinth]

2questions:

how much performance will i lose from disabling xmp
can I use the ram's warranty if it fails to run in the company-set XMP mode it is advertised with?

[twin dew]

No to the second.
As the XMP is what the RAM manufacturer says the RAM can do in Golden Sample system. not in yours with your CPU and your MB.

But point is to diagnose at this point.
Seeing about the performance loss and final settings comes later, once the fault source has been identified.

[sterile plinth]

so i buy a ram that is called "3600mhz cl16 ram" and it crashes 3 times a day at 3600mhz cl16 manufacturer timing and thats normal?? i hate this

[twin dew]

Yes? As the actual speed speed is determined by the worst of CPU IMC, MB and RAM.
The RAM is specced to do that, but they can give no guarantees on your CPU and MB.

All-auto sets the RAM to what all three can do at stock settings.

So JEDEC voltages, JEDEC timings and the frequency set to the lowest common denominator from the RAM SPD JEDEC frequencies, the CPU official limits for the amount of sticks and ranks, and what the MB maker guarantees.

[sterile plinth]

that would be around 2400mhz?

the ddr4 default

[twin dew]

No, there is JEDEC specs up to 3200MT/s for DDR4.

[sterile plinth]

ah, thats nice, i will try and let you know if i get any more crashes

might as well format too

[twin dew]

But what the memory maker guarantees at stock voltage can be down to 2133MT/s for DDR4 sticks.

[sterile plinth]

since for the 2nd time my entire os has been corrupted from the too many bsods

[twin dew]

So you might not even get that 2400MT/s at stock.

Depends on the SPD contents.

But point is to first stabilize the system and find the cause.
And then go back to as close as previous performance as possible.

[sterile plinth]

Alright,I turned off xmp profile

ok fuck its 2133mhz 😭

[dire igloo]

Happens

It won't be for long tho

This is only to find out if XMP is the issue

[dire igloo]

This.

[sterile plinth]

supposing i never get a crash again then i have to manually change the ram settings?

manually overclocking ram sounds crazy hard

it has like 4 different timings

[twin dew]

Probably just setting XMP and then lowering frequency one step.

[dire igloo]

If your lego constantly breaks, disassemble it step by step until you find what's causing the instability.
Then fix and rebuild from there

[twin dew]

But we will see.

If XMP.

[dire igloo]

Four? Try four dozens - and even that ain't all of em

[sterile plinth]

dont scare me more

[sterile plinth]

it better be

[dire igloo]

Don't worry, you're not on your own.
You have a rich source of great knowledge here (and me)

[sterile plinth]

baldur is like a philosopher he has the answer to everything

(and you :) )

[dire igloo]

Wait. He has the answer to me?
Damn, I'd like to hear it

[edgy hazel]

We need to stop using abbreviations in IT. I cannot read this article because it keeps using "Web Application Proxy"

"The server that hosts WAP must install an update..."

[dire igloo]

(this project was coded by Bibi H.)

[charred pewter]

WAF ... Wife Acceptance Factor ?

[dire igloo]

Warendorf

[tough owl]

Your brain is rotted if you can't read that normally

[charred pewter]

someone was telling me to install a WAF ... and i was just blankly staring

[dire igloo]

Given that that's the German license plate code for a village near me, I'd be fckn confused constantly

[tough owl]

y'all have codes for different areas on license plates?

In America we just have different plate designs(same formfactor) for different stated

[dire igloo]

Every city has their own one to three letter code at the beginning of the license plate

B is Berlin, M is Munich, H is Hanover, etc
WAF is Warendorf

[charred pewter]

interesting

[dire igloo]

There are a few exceptions too.
NRW is Northrhine-Westfalian police
BD is federal government
Y is Bundeswehr (military)

Just to name a couple

[charred pewter]

yeah, in USA, each state has their own 'plate' with the full state name on the top

then a bunch of vanity/styled plates... i get the Humane Society one that has a kitty on the right side, and my number on the left 🙂

[dire igloo]

And that's pretty much everything you have in terms of localizing the origins of a car

[charred pewter]

makes more sense... and yeah, plates in eu look more like proper serial numbers for cars

USA its all manner of crazy froo froo

[dire igloo]

Btw, how a German license plate is constructed:
Blue bar with EU stars and D (to show it's Germany)
One to three letters for the city it's registered with
TÜV stickers that show when the next check is due
One or two letters, customizable
One to four digits, customizable (no leading zero)

So you can make a license plate that says
GS-G-9

But it'd have to be not taken by anyone which is super unlike for GS-G-9 since GSG-9 is a special forces unit of the German Federal Police (Bundespolizei)

[dire igloo]

Government vehicles are different.
They have a letter code for showing they're government (like BD or BW),
then TÜV stickers,
then a number that indicates the actual agency (BD 1 is Bundestag, the federal parliament),
then a number to identify the specific car (usually consecutive)

Blue bar too of course

Example:
BD 3 2 is the 2nd car in the pool of the Bundesrat

[wanton orchid]

here I don't even know if letters have any special meaning
it used to be fully random
I think now you can customize it a bit more but unsure

plates are staying with cars

there is no personal plates anymore

it's only car plates

[tribal kraken]

Also what's worth noting that plates in Germany are personal? You buy car without them? New owner will get he's own plates for used cars? In contrast Finland the car has the same plate it's whole life

[edgy hazel]

yup

but the car is also registered to the plate so you can't just swap the plate to another car

[tough owl]

Plates are registered to cars here. But you can swap plates to a different car if you take it to the dmv

[dire igloo]

Basically plates are tied to car and owner

You have to deregister the car if you're selling it, meaning the plate gets invalidated (or moved to a car you've bought) and the car you sold will have to be registered by the new owner with their license plate of choice.

[gleaming kite]

U for gto Y plates

[night girder]

Bro, I am thinking this for years. Too... many... abbreviations.

[frank elk]

hey guys any idea what this could mean

I am host and it keeps logging of my friend with this message on a specific spot then when he joins he gets killed......

[night girder]

Wrong channel?

[frank elk]

what does that mean?

[gleaming kite]

server channel pls

[frank elk]

ahhh

[twin dew]

This is Off-topic.
#1038092680493801533 for help.

[dire igloo]

I intentionally didn't mention them because idk how they work. BD and BW I do know how they work. No idea about Y

[dire igloo]

wdym? idm it tbh

[gleaming kite]

they a strictly for military vehicles I.e lepards marders boxer Jagers

or military personal cars

[dire igloo]

That's pretty much what I said. Vehicles registered to the Bundeswehr

doesn't matter if it's some general's Mercedes limousine, a tank or a transporter

[gleaming kite]

ah didnt scroll that far up XD

[dire igloo]

#off-topic-tech message

the message I replied to explained that they worked differently: letters, number, number. not letters, letters, number(s)

[gleaming kite]

i know how they work XD war selbst soldat

[dire igloo]

Mein Beileid

[gleaming kite]

wie kommste darauf

[dire igloo]

I'd love to hear you explain Y license plates tho

[gleaming kite]

not much to explain Y means military the number is for the vehicle probs with cert numbers like to certain parts

[dire igloo]

I work IT in a government agency, the amount of shit I have to put up with because of higher up politics is insane (but also super entertaining cuz it doesn't affect me that much)

[gleaming kite]

the money was good enough not to care about the politics XD

[gleaming kite]

also gives the side affect police may pull u over but must then call the "feldJäger"

which is worse trust me XD

[twin dew]

"Nice"... AMD doesn't provide firmware updates to be used with linux early boot CPU firmware update functionality for the consumer CPUs, only for the server CPUs...

[maiden coyote]

how long would you use a psu for? I removed a unit from 2009 from my nas lol

[gilded helm]

Oddly enough, that would get you laughed at today, but in reality that's probably fine for a lot of builds.

[final zephyr]

i get "Out of video memory trying to allocate a rendering resource. Make sure your video card has the minimum required memory, try lowering the resolution and/or closing other applications that are running. Exiting..." when i try to start up the game i updated my drivers and i have a 3080 so i dont know why i whoud get this error

[edgy hazel]

this hat is cute

[twin dew]

Well, got microcode file extracted from BIOS update for the current CPU from "random" github repo, then packaged it for use by that microcode update functionaly with contents of another "random" github repo that was 7 years old.
And it worked.

[edgy hazel]

IT'S 9€!!

[twin dew]

You probably have Intel 13th or 14th gen CPU.
But wrong place.

But basically Intel & MB makers fucked up and 50+% of 13900 and 14900 CPUs are unstable at stock, and significant percentage of 700:s.

That error is even listed in Nvidia GPU driver changelog as being Intel CPU caused.

So I'm up to the firmware included in the AM4 AGESA 1.2.0.B for the Renoir.
The new one in 1.2.0.Ca is too new, but could probably extract that myself.

[final zephyr]

i do

but how to fix

[jagged snow]

You don't

It's a hardware issue

[final zephyr]

so i just cant play satsfactory? then? on my brand new pc?

[jagged snow]

You might potentially be able to fix it by lowering ppt and/or disabling motherboard default OC profile

[jagged snow]

But yes, you got defective hardware just like many others

[night girder]

That's some nasty stuff.

[jagged snow]

Did you see the analysis that GN/HW unboxed did of failure rates?

[night girder]

Is it defectice out of the box, or did MB fry CPU? 🙂

[jagged snow]

Both, most of the blame lies with intel though

[night girder]

Since I got AMD, I don't follow Intel news closely. I can vaguely remember something mentioned in here.

[jagged snow]

Even on the server chipsets with no possibility for oc, up to 50% of them are too unstable for commercial workloads

[night girder]

That sucks 😕

[jagged snow]

Yep

[night girder]

No group lawsuit or anything against Intel?

[jagged snow]

Lots of game studios and game server hosts are switching their 13th and 14th gen hardware to AMD

[jagged snow]

No, people are still hoping they'll fix it

There's no public information that even nails down the cause of failure yet

[night girder]

But if the hardware is damaged, then it's not fixable besides by replacement.

[jagged snow]

Not damaged, defective design

Could be fixed with a microcode update

Whatever the solution, it'll almost definitely involve a performance loss

[night girder]

or by downclocking maybe, but then you lose potential power, which you payed for. Sounds like a lose-lose situation.

[night girder]

wdym defective design?

[jagged snow]

The chip design itself may be the cause of the instability

[night girder]

This didn't sound like design flaw.

[jagged snow]

Afaik baldur wasn't referring to the latest info on the issue there

Because it's not a degradation problem

They're unstable out of the box, not after some period of being run

@night girder https://youtu.be/QzHcrbT5D_Y?si=dfIgqHWn1aVjT3uZ

[night girder]

Twice he said, "grain of salt" so far 😂

[jagged snow]

Yup, and this is pretty much the best information avaliable on the issue

[night girder]

Interesting video, but it misses some deeper data.

for example; how many servers crashed in the last 6 months. And how much % of those crashes are 13/14th gen CPU with W motherboard.

and is that more or less than previous generations. All he does is show a black screen of one server that crashed with a 14th gen.

[jagged snow]

He went into more detail in this video
https://youtu.be/oAE4NWoyMZk?si=C-xJsNUfFX7f1U4C

[jagged snow]

He did say this though
50% of the systems deployed with 13/14th gen on W series chipset experienced instability

13:30 timestamp

[night girder]

for two companies. But yeah, but that's not good.

This is bad for Intel. Good for AMD.

So weird not all CPU's have this issue.

[wanton orchid]

mali 10 series driver merged in 6.10

looking nice

[night girder]

If it's a CPU problem, wouldn't all show issues?

It must be rather a combination of CPU + Motherboard? Or certain factories outputting bad CPU's.

[jagged snow]

Not necessarily

[wanton orchid]

not if this is a hardware problem

[night girder]

How can some CPU's be damaged and others not? 😒 Like if it;'s a design flaw, all cpu's should have the same design flaw.

[wanton orchid]

big volume providers using volume of cpu also uses volume of motherboard, they are like half running on exact same board

[night girder]

Then wtf? How is this even possible.

[wanton orchid]

not if this is a hardware problem
read please

hardware never is copy paste indefinitely

[night girder]

Yes I can read.

[wanton orchid]

there is always batch high and low

that's why binning exists

[night girder]

how can hardware not be a copy paste.

It's made by machines based on schematics.

[wanton orchid]

should I remember you cpus are made with irl material ?

[night girder]

That's the whole thing about machines, they can produce the same material over and over again.

If it's handmade, then I can assume deviations.

[wanton orchid]

why dont you think we didnt squeeze billion of transistors earlier in the human history ?

why do you think wafers are that expensive

when that 0.00001% is make or break

then it make for 50% and break for the other half

anyway it is called bad quality control / marging engineering

a freaking lot of behaviors in (pushed too far) silicon are snowballing

for instance in most diodes it's called "Avalanche" breakdown point

[night girder]

This "defective design" got me confused.

Because if it's a design issue, then all should show weird behavior.

[wanton orchid]

everything is fine, until a single electron wants to make the whole diode snap in 1/1000000 of a second

[jagged snow]

By that I meant that they're essentially pushing the silicon too hard
Similar to vehicles with high performance engines

[night girder]

Wonder if Intel will ever admit what is the real cause. For legal purposes. If it's trully their fuck up, they are in some legal trouble.

[silk terrace]

There's always some randomness in material sciences. A design defect might be just increased potential to crack under high temperature or some such.
Doesn't mean every unit will, but the chance is higher than acceptable.

[night girder]

I get that, but if they did that I still think all should at one point become unstable.

[jagged snow]

And that might be the case

[night girder]

It's just not possible that "by luck" some can't handle it and others cannot. There has to be a reason.

[jagged snow]

Wendel did say the failure rate did appera to increase over time

[jagged snow]

That is entirely possible though

From the factory, the CPUs are all on the very edge of where they will or won't experience issues

So natural variation from chip to chip desides which side of that line they fall on

[wanton orchid]

the reasons are not always under control, but you have control over how much control you get over it
that's marging engineering

in building usually the rule of thumb is 1/4

[jagged snow]

For instrastructure it's usually double

Bridges etc are designed to handle double their rated load

[night girder]

Sorry, that's just shit engineering if you cannot explain why certain behavior is happening. That means you don't fully understand your own product.

What a interesting story though

[jagged snow]

That is very often the case with anything that even comes close to the complexity of a CPU

[night girder]

Let's hope they can learn from it. And maybe we too. If they ever tell us the truth.

[wanton orchid]

you are getting off topic there, your point was why is it not the same for every unit
we answered that

[night girder]

Oh no! I am getting off topic... in the off topic channel 😉

[wanton orchid]

not having control over, does not mean not knowing

[night girder]

You rather have that I just keep arguing for the sake of arguing? 😂

[night girder]

If I am an engineer, and I design a CPU, and 50% of them causes amount of crashes and the other 50% doesn't, and I cannot figure out the cause, then I don't fully understand my own product. Simple as that. Nothing wrong with it. But it means there is an opportunity to learn.

[wanton orchid]

" I cannot figure out the cause" again

[night girder]

Maybe certain materials act different than they expected or something.

[jagged snow]

There are hundreds of engineers working on a single platform, I doubt any of them fully understands every intricacy

[night girder]

True, but together they should no? It's been months according to you all. I didn't bring up this subject btw. You all stated that we are in the dark atm with the cause 😂

[wanton orchid]

I didnt

also WE is not Intel

[jagged snow]

That's true, but they can't act as a single entity. And maybe they know exactly what the issue is, and are trying to figure out a viable solution

[night girder]

That's what I was thinking too. And the legal stuff.

If Intel somehow dropped the ball, there will be cases against them and money loss etc. And reputation damage. Which already might be happening.

[jagged snow]

Yeah, could be either of those
Intel management might just be outright unwilling to replace every 13/14900k they've shipped though

And so they're likely pushing engineers to figure out a cheaper fix

[night girder]

Yes, wouldn't surprise me at all.

[wanton orchid]

I'm tired of the naive "justice"
there are 2 possibility

• they are at fault
• they are not at fault and can prove it

[night girder]

You are stupid if you think Intel will commit suicide by saying: "we are total at fault, we replace everyone one of them".

[wanton orchid]

the "third possibility" is just creating an escape by making them act stupid

[wanton orchid]

I'm saying that if they dont say anything, then they are

but ofcourse people will be naive and go like "oh but what if they are just stupid and nice, everyone makes mistake"

"- and also your parents right ?"

[night girder]

Not saying anything is not possible for them either I think. They have to say something and fix something. That doesn't mean they will ever admit being fully at fault, if that's the case.

It's a big company, with lots of investors and a reputation to uphold.

Intel continues to investigate reports of instability on its 13th and 14th Generation desktop CPUs (K/KF/KS). While it has yet to identify the root cause of the problem, Intel has offered guidance on helping prevent the instability. Intel has also uncovered one power-related bug, and says that it’s working with vendors to distribute a patch for it.

[wanton orchid]

my take is :
some are failing as soon as over 2 weeks
that should have throw an alarm

but some executive probably said that they can't afford to have their feet still grounded

now they are falling

gg wp

[night girder]

I read they are declining RMA requests.

but source is a bit questionable.

HardwareTimes reports that in one incident, they talk about how Intel denied the RMA for a "plagued" Intel Core i9-13900KF processor. The outlet had two units they wanted to RMA, the Core i9-13900KF and Core i7-14700KF, with Intel accepting RMA for the 14700KF, but not the 13900KF processor.

[wanton orchid]

from wendell they are at least not declining Big buyers rmas apparently

[wanton orchid]

ahah, customer service not being aware they are basically the same underthe hood

[night girder]

Not really:

HardwareTimes explained the issues to Intel, with the company replacing the 13900KF with a new CPU, but that also had the same issues within two months, with the event view logger showing an astounding 44,242 errors in four months. Intel denied the RMA request for the second 13900KF processor, even after initially agreeing to a refund.

It's actually Intel denying it it seems.

[wanton orchid]

what is getting me a bit off though is we are only now really hearing about these big issues

these have been reviewed way back in the year already

[jagged snow]

No, people have been talking about it since then, especially inside organizations.

It's just that it was never made public, and many people inside companies denied it

[wanton orchid]

some people will probably caught fire

[night girder]

I wouldn't be surprised, people were a bit hesistant in the beginning to immediately blame the CPU when strange things started to happen. But after a while connected the dots that the common factor in the issues was the 13/14th gen CPU's from Intel.

And once they start to share that information with the outer world, more and more people started to look at the same culprit.

[languid gulch]

yea i can't wait for the public torching of intel with all their lies & denials

it seems obvious now that the last stable CPUs they put out were 12th gen

[gilded helm]

I think Intel is going to get sued over Raptor Lake if this gets worse, and then they will have to do replacements. It will be the stores and OEMs doing the servicing who will sue Intel.

[languid gulch]

oh yea it ain't gonna be the gamers, it's gonna be the BIG sledgehammers of those with datacenters

[tough owl]

I bet a class action lawsuit is being drafted as we speak

[maiden coyote]

I already sold my intel stock, let em have it.

[wanton orchid]

I mean yea, advertising being better than your competitor then selling product that just crash, is going to feed some backlash

[languid gulch]

i liked that wendell revealed that the company he talked to set things so low to get it to be stable that it dropped below 7950X performance & that's why they switched, that's gotta hurt

[jagged snow]

Anyone heard what the performance benefit of SCALE is compared to OpenGL on rdna?

I'm tempted to see if I can compile meshroom CUDA for AMD gpus

The openGL implementation is dog slow

[charred pewter]

whomever did it, shame on you... but now that ive found how to use some ai tools, im having a blast 😄

[twin dew]

Make #1038092680493801533 thread and ping me in it.

[twin dew]

Combination thing on consumer side.
MB makers are/were underclocking the CPUs with wrong settings.
And Intel might be feeding the CPUs too much voltage at high frequencies and causing rapid degradation in the ring-bus inside the CPU.

Last time this was talked about with more "data" on the possible VCore & Ring-Bus degradation side.

Including data from Warframe devs about client crashes and then binning data from Intel CPUs on the max VCore they use to correlate etc.

[edgy hazel]

Warframe mentioned

[twin dew]

But most client cases are probably from those consumer MB maker tweaks away from the "baseline" Intel settings.
And the server side is probably more from actual degradation.

[visual tree]

Got a new AC so I can survive this summer without melting. Hope it's better than Toshiba crap 😅

[languid gulch]

warframe's spinny jump animation is something that every other game should steal

and also it should be just as much of a pain in the ass to learn

[languid gulch]

it'll pay for itself with the first heat wave

when you're no longer laying in bed sprawled out try to not melt

[soft bloom]

untill there's power ...

[twin dew]

And I even got NUT working for two machines \o/ (The networked UPS utility)

[maiden coyote]

@twin dew if you've got a minute, would you mind taking a look at this, this came from a system that complained about a stpe address leaf, and then locked up https://pastebin.com/p0Cq0nL3

[twin dew]

Running VMs under KVM?

VM does something, causes "handle_ept_violation" to be run in the main kernel kvm.
Then main kernel does some kvm_mmu (memory management unit) page fault stuff, to kwm_tdp_mmu_map (two-dimensional paging), and crash in there for some reason, like the last line says.

But no idea what is causing that. I have never before even seen (IIRC) linux kernel crash.

SPTE: Shadow Page Table Entry, typoed in several places as STPE.

[wanton orchid]

there have been numerous patches about kvm and host memory management lately

maybe that is related ?

[twin dew]

Yeah, most likely actual kernel bug in the specific version in use as the main OS kernel.
But cannot know, and the specific line isn't referenced anywhere I can find.

But might not be relevant.

[twin dew]

But the VM insides did something stupid with memory and then the main kernel barfed up handling the violation.

[wanton orchid]

qemu also got updates

[twin dew]

Or the KVM did something stupid in assigning memory to the VM.

But can also be HW fault in the actual HW corrupting memory.

[maiden coyote]

I can try to get the full log and paste it'd help you help me xd

I appreciate it. my kernal version is root@pve-N100-16G500:~# uname -r 6.8.8-2-pve

[tough owl]

That's an old kernel

Not even lts

[twin dew]

6.8.12 was the last 6.8 and that whole line was EOLd in may this year.

What distro still uses such kernel version?

[maiden coyote]

I'm using proxmox pve 8.2

[twin dew]

So the idiots seem to be following Ubuntu non-LTS kernels on "LTS" OS...

[maiden coyote]

but the issues randomly started about a week ago when I was trying to setup docker in a container. proxmox seemed to die in slow motion. at first the network options vanished. then it It was laggy and I couldn't start/stop VM's.. then the web panels stopped loading and it because unresponsive.. the VM's were still running but I couldn't interact with the host

[twin dew]

No, 8.2 VE is based on Debian 12.5, but then upgrades to completely insane kernel...

[maiden coyote]

this time I was trying to update a windows 10 VM.. and it got laggy, cpu usage was went up to 80.23 and then the whole thing died.

[twin dew]

Released in April, when the 6.8 was already out of support, and using old version even then...

[maiden coyote]

maybe it's their own custom thing and they're behind normal releases? iono.

old kernal or not.. It shouldn't suddenly become unstable after running without issue for 3 months.

[twin dew]

Yes, it is customized version, as that -2-pve shows.
But that means they need to backport everything all the time.
Instead of following some LTS kernel with their patch-set.

But check for HW issues?

[maiden coyote]

is there a command I can run that'll list hardware errors?

[twin dew]

Probably no?
As even without monitoring corrected errors wouldn't matter and uncorrected but detected errors should panic the kernel.

The monitoring is for corrected errors to be able to do stuff before the uncorrected one hits.

Proxmox 8.2 with 6.8.4 kernel had lot of stability issues, and people have had to downgrade to 6.5.x kernels.

But this "we shall use this completely random kernel as the base for our patchset" makes the Proxmox devs seem like idiots to me...

That update from 6.8.4 to 6.8.8 was in attempt to fix those crashes it seems.

Fixed for some, but not for all.

[maiden coyote]

alright, I appreciate the help.

[twin dew]

But see if the final crash cause keeps the same or changes?

To differentiate between kernel issue and physical hw issue that would need stress testing or like to find.

[maiden coyote]

next time it crashes I'll downgrade my kernel to 6.5.x. I was doing totally different things when it crashed. I've only had two crashes so far. the one I posted the log from yesterday, and one last week. that'd I'd have to try to find. before that I had like 64 days of uptime.

[wanton orchid]

I don't know how they are doing it

as 6.6 is lts
not 6.5

heavily modifying a non lts kernel is absurd

unless you plan to switch to the next lts as soon as possible

Also I don't know why they would not use upstream
I don't know about any significant virtualization tech that have not been upstreamed after one year

imo
either you continuously support your build
or you give people means to easily upgrade to latest
there should not be any in between

that's why I find both "stable" and "rolling" release schemes to be stupid
I'm all for "rolling" like management with "stable" fallback
that way people can keep their system actually stable
without breaking everything for "cutting edge"

you basically get cutting edge except for the things you need some stability (which most of the time is only interface, not low level system)

versionning does become more complex
but it is not impossible
nor that difficult to use in the end (I'm working on it)

btw
rolling != fast channel
stable != slow channel
I also want people to be able to select their channel speed

breaking rolling is often linked with rushed channel

Rolling should not rhyme with testing

[twin dew]

The amount of fucking around with various programs SystemD unit files to make them behave correctly is way too high...

NUT was still using .services for UDev that were depricated in 2011...

[maiden coyote]

could secondary drive with VM's on it fucking up cause a kernal panic like I had?

[twin dew]

About same as just rebooting without warning, so yes.

[maiden coyote]

the secondary drive that hosts serveral VM's just died.

[twin dew]

Shouldn't die-die, just can lead to file system corruption.

[maiden coyote]

I mean it's gone.

I was trying out truenas before setting it up on bare metal and moving a file to it and poof, drive gone.

if that's the hardware issue that's making my proxmox server sick I'll be happy.. and I'll stop using drives I've found in the trash for setting up important stuff.

[twin dew]

So needed to create udev rules to add the UPS to SystemD .devices and add static alias for it.
And then override one of the NUT services to not want old depricated shit and instead bind to that ups.device (so started when the device appears, stopped when device disappears).

[maiden coyote]

are you using nut, or peanut?

well peanut is nut, it's just a frontend for it.

[twin dew]

Text mode only here.

[maiden coyote]

might be worth looking into peanut. it looks like it gives a nice webui for it.

[twin dew]

Why would I want webui?

[maiden coyote]

i think if the ups supports it, it'll pull power usage data and it'll graph it for you.

and line voltage and all that other good stuff.

[twin dew]

And the old depricated Wants just caused "you are stupid to use depricated stuff" entry in journal in bootup.

[maiden coyote]

?

[twin dew]

Just uses the upslog log in the NUT for that collection part.

As in the thing I fixed for myself by tweaking the udev and NUT SystemD service files with override in another file.

But that in 2024 the actual devs haven't fixed using stuff depricated in 2011, that doesn't actually do anything today except log a warning.

Also shit like following by design, but this cannot be fixed without causing problems updating:
nut-monitor[794]: fopen /run/nut/upsmon.pid: No such file or directory nut-monitor[794]: Could not find PID file to see if previous upsmon instance is already running!

That is expected, but they don't handle it and not throw useless warning.

Hmm... maybe I can add another override that creates that file as empty one if it doesn't exist yet...

Before the actual unit runs.

[twin dew]

But the official UDEV documentation is just painfully bad.

[wanton orchid]

did you set some redundancy at least ?

[wanton orchid]

as is systemd with almost everything under "dont worry we got you backed, undefined behavior be we swear it works great, oh also bunch of implicit behaviors you can't easily see and setup, and that could be changed un the futur without warning"

I swear there are some microsoft hired engineers in the back of the systemd team

the way of doing things is too close

[thick radish]

Considering how it seems designed to sabotage Linux...

[night girder]

Yeah this makes sense.

[maiden coyote]

Nope, I had the windows vm backed up.. I completely lost my docker vm

[wanton orchid]

I would be ok for you to use trash hdds if you had setup redundancies
hdd will die
trash hdds will die sooner faster

[maiden coyote]

It was a wd blue ssd. Probably the one that the ssd list says to avoid

[maiden coyote]

Well, drive seems okay, crystal disk says it's fine

[twin dew]

Haven't really had problems with it.
It is just that bunch of other projects set their unit files wrong.
I had for example have to add unit dependencies into console-setup to make it work right, as by default it is in race condition against other stuff etc.

console-setup explicitly sets to not use "default-dependencies", which breaks the load order for it to cause that race condition against the base units.

And at least the systemd documentation is pretty complete and has examples.
Just split into multiple documents that you have to shift through to get complete picture.

Just lot of devs in other project that think that udev and systemd work like the old init.
And so the unit files included in distros by default are just often bad and incomplete.

Old INIT was basically deterministic and single-threaded, only starting one thing at a time, always in same order.
SystemD is multithreaded and tries to start everything that it can at same time, and you need to use the unit file Before=, After=, Wants= etc. declarations to other units to set when your unit can be started by declaring everything it needs, and then everything that needs it, so SystemD starts it in right position.

With sshd.service, on computer where I have set sshd to only bind to one specific IP, I need to add requirements to the sshd.service to only start after that specific network interface has been configured.
As again the default sshd.service unit doesn't contain the necessary dependencies to only start after networking is actually up and running.

Works fine for the default "bind to everything" mode, but not with more limited sshd.config for that.

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

[languid gulch]

Crowdstrike is making stuff go splat around the planet, have fun tomorrow!

[soft bloom]

the heck happened to tech today?

[twin dew]

And AMD fucked up on the chipset naming...
X870E is same as X670E.
X870 is same as B650E.
B850 is same as B650.

X870E is dual-chipset, X870 is just single.

[soft bloom]

wtf happens to windows? i heard german airlines are suffering. here we had outages in banking...

[river mountain]

sounds like azure's had a shit

[languid gulch]

#off-topic-tech message

[twin dew]

An update to a product from infosec vendor CrowdStrike is bricking computers running Windows.

Causing BSODs

[soft bloom]

why tomorrow?...

[twin dew]

Probably time zone thing.
His very late night, and the next morning shift has to pick up the pieces.

While in europe, the picking up the pieces is ongoing right now.

[soft bloom]

maybe. just wasn't expecting 10 am to be evening. that seems waaaay to far

also i still can't get over the fact that indonesia is only 4h away from me

[languid gulch]

yea now it's hitting the US

gonna be a long weekend

[midnight osprey]

Microsoft what are you doing?

[twin dew]

Not MS thing?
But external antivirus/security program that has very deep impacting driver being fed garbage data by bad update, causing BSODs.

[midnight osprey]

Ah

Trigger clickbait shite then.

[languid gulch]

i hope it hasn't hit Starbucks, because techs are gonna need their coffee

[midnight osprey]

Haha

[twin dew]

So every customer of that specific security solution company is being affected.

[languid gulch]

which is a LOT

[midnight osprey]

Which security solution?

[twin dew]

CrowdStrike

[midnight osprey]

Ah, never heard of them

[languid gulch]

they're gigantic

[midnight osprey]

Which means my stuff is safe.

[twin dew]

Enterprise only.

[languid gulch]

every major airline is down right now

[midnight osprey]

Wow

[languid gulch]

banks, grocery stores

[twin dew]

And back to this.
The SystemD documentation is very good.
Everything documented, every option well explained, lot of explained examples etc.
Just spread out along one document per concept, but with very good linking between the documents.

And EVERYTHING is settable and tunable.

Just that the defaults work in most cases, so if you are just copying someone else, without reading the documentation, you will never know the tunable setting you would need exists.

But as the system is complete paradigm shift from the systems it replaced, and people don't read the documentation and learn, the other software hasn't been updated to work with it properly in many cases.

And that override system is just great, to be able to just change partly how some other packages things function, in a way that survives package updates as is without file conflicts etc.

Working like the various /conf.d/ directories for program specific settings in programs that support that.

Override system for the unit files I meant.

Unfortunately doesn't have way to remove dependencies the original unit file sets, only add to them.
For other things you can remove the base settings.

(Except completely replacing the unit file, but then any changes to the original unit file get "lost" without manual intervention)

For UDEV, you can do all the same things it does by default, customized, if you want to.
Just that the documentation is absolute shit so you need to look into the actual "hidden" default rules in many cases to see what is happening and now it is done.

And lot more than it does by default as there is lot to match to select what to trigger on, and then you can basically do anything when triggered.

[wanton orchid]

completely wrong to say that
it is easy to consider systemd stable and complete coming from older init which had very little control nor efficiency
but for instance the "stops in the reverse order it started" is complete bullshit, and lack of knowledge from systemd devs, and especially does not make any sense with the multiple targets involved

[twin dew]

Stops in reverse order, IF you have set the dependencies correctly.

The shutdown uses same "trigger everything you can at same time" approach.

[wanton orchid]

I'm talking about one unit

[twin dew]

Unit stops how it is set to stop.
Most units don't bother to set how they stop so they just get killed.

[wanton orchid]

also the multiple stages of booting under systemd made to work around some problem, are not as efficient at all as systemd claimed to become

[wanton orchid]

nope, any After turns as a Before in stopping sequence

[twin dew]

But that isn't for one unit, that is for relation between multiple units.
And how is that non-suitable?

You are supposed to set things in those stanzas that you cannot work without.
So if the reverse wasn't used for shutting down, stuff should completely break.

[wanton orchid]

I dont get your last sentence at all

systemd completely fail at understanding startup and cleanup of units

[twin dew]

My unit cannot work without X, so I set suitable dependency for my unit to be only started after that X is started.
On shutdown then SystemD shuts your unit down before the X too, as expectation is that your shit would be broken and couldn't shut down cleanly without the X.

No? It understands the set dependencies perfectly.
Just that other packages often have incomplete dependencies in their units for their software.

[wanton orchid]

for very simple things like "check your dns every x min" then yes it would not make sense to not use reverse
but for anything more complexe like lvm volume scanning and setup deamon
that's no near how it works

[twin dew]

LVM units are complete mess currently, but that is LVM side problem that they haven't made proper unit files.

They could be fixed, but I haven't done it as I don't use LVM.

Basically missing dependencies in the units by LVM devs.

[wanton orchid]

I would call these stuff "opportunistic reordering"
which is actually pretty much expected from a naive user pov
(but is not honored by systemd)
for instance the sshd shut down early

[twin dew]

Or in cases, too many dependencies to wrong things.

Yes? SystemD does that "as soon as possible, at same time" if you don't have your dependencies set right.

[wanton orchid]

I'm not saying that systemd is worse than old init systemds though

not near that

[twin dew]

And the various other program devs haven't caught up and read the documentation, and set their unit files up correctly.

[wanton orchid]

but it have a lot of design flaws
no wonder so many units are so badly set up, when the environnement is pushing for misconceptions and implicit default

[twin dew]

The defaults are for standard daemon type things that only get run after the system is up and ready.
Anything more complex needs to set their own and set the DefaultDependencies=no if don't want to just add to those limits.

Single line in the unit file to get rid of the defaults for load order things.

[wanton orchid]

but what scope of default ahah

oh yea sorry forgot they assumed you are in the dev team mailing list

back to my first point

[twin dew]

For .service files:

Service units will have dependencies of type Requires= and After= on sysinit.target, a dependency of type After= on basic.target as well as dependencies of type Conflicts= and Before= on shutdown.target. These ensure that normal service units pull in basic system initialization, and are terminated cleanly prior to system shutdown. Only services involved with early boot or late system shutdown should disable this option.

Instanced service units (i.e. service units with an "@" in their name) are assigned by default a per-template slice unit (see systemd.slice(5)), named after the template unit, containing all instances of the specific template. This slice is normally stopped at shutdown, together with all template instances. If that is not desired, set DefaultDependencies=no in the template unit, and either define your own per-template slice unit file that also sets DefaultDependencies=no, or set Slice=system.slice (or another suitable slice) in the template unit. Also see systemd.resource-control(5).

[wanton orchid]

you want to get a "simple standard typical unit" ? get a template for that

no implicit default bullshit

dont get me started on the default mount lock and remount at shutdown late procedures

[twin dew]

Then set that you cannot be active after that in shutdown?
Most stuff CANNOT be active outside that default dependency window.

[wanton orchid]

and so they will be set for

[twin dew]

Stuff like LVM etc. that need to be set up before that default (for just that reason), and quit after that window (again for same reason) need to set their dependencies very carefully.

And LVM isn't.

[wanton orchid]

everyone is focused on the starting behavior

but oh man is shutdown procedure a freaking mess of shit getting killed and data staled naturally, but never is reported because storage is under a lock

[twin dew]

Yeah, because you are supposed to quit before the storage gets locked...
There is even separate targets for that on shutdown, instead of just the generic shutdown.target.

www.freedesktop.org/software/systemd/man/latest/systemd.special.html#

[wanton orchid]

you said "you want it to shutdown later, then setup different targets" oh btw "it will shutdown in reverse order no matter what"

you cant set it right

[twin dew]

You can with middleware units for example.
And Conflicts entries.

[wanton orchid]

that will not do it

conflict before is also reversed as a after

ngl just seems like you started digging into it

[twin dew]

Conflicts=shutdown.target is the standard addon pulled by the defaults for .service for example.

Causing the thing to get stopped before the SystemD can enter shutdown.target, after which the various umounts are done etc.

Same can be done for any other unit or target only run on shutdown.

To specify when it needs to be down.

[wanton orchid]

that specificially does not work at all

your workaround does not even exist

and I demonstrated it on my side

[wanton orchid]

^

basically for this kind of retarded reason

[twin dew]

Example of what you mean with that?

For shutdown behavior.

[wanton orchid]

I already gave an example

you start a monitoring process early after hard dependencies are setup

it works great

but now you want to add a before conflict to something so it shutdown in order

but now it added a ghost dependency After= on start, and it does not work anymore because it starts too late

or is performance clutter because it does not get up on the storage setup stage but only after

and that is HARD behavior enforced by systemd

[twin dew]

Because you don't use Before for shutdown? But that Conflicts.

Or StopPropagatedFrom

There is load of various forward and reverse dependencies that almost no-one uses.

Or BindsTo=

Ok, that BindsTo doesn't work in this case

Where you need to start before it, but stop before it too.
Mostly need to use that Conflicts to a target that happens on shutdown before the slot the other unit sits in.

Like that umount.target which happens before file systems are unmounted.

So if you conflict with that, you get shut down before the file systems are removed.

final.target to be shut down after unmounts and most services being down.

Remember also that you can have one unit that handles the starting of another unit on bootup, but doesn't kill it on shutdown.
So that first unit gets started in the slot you want for the startup, and then it just shuts down as one-time thing.
And then the actual unit doesn't have dependencies and isn't on the autostart at all, but has suitable dependencies for the shutdown.

Type=oneshot services etc.

[wanton orchid]

that's how I work around it for me

but really there is no real reasons for hard reverse enforcement
(appart from the compatibility with the way systemd is implemented)

anyway I'm working on an alternative solution that does not assume every unit are user setup

[twin dew]

Because it makes sense for most units?
And that two-unit solution is easy for the ones that it doesn't work for?

As the default makes it much easier for most.

And the two-unit one isn't that much more complex for those that the default doesn't work for?

[wanton orchid]

^

[twin dew]

And when the template would need to change?
And people will then still use the old template for 15 years onwards for new development because the first search result was using the old one?

[wanton orchid]

you mean people will setup a unit without care, like what implicit default is encouraging

it will not break more than having to change the implicit defaults (I mean systemd devs changing the implicit defaults here)

[twin dew]

Which works for most standalone daemons and programs, and then it will keep working no matter the changes to the defaults.

[wanton orchid]

why

[twin dew]

If the base premise is kept to by SystemD people.

[wanton orchid]

you are basically implying a change to the default would not change the default behavior

when it is exactly what changing the implicit default means

[twin dew]

It would change the default behavior, but the point of those defaults is to give suitable state of system for normal stuff.
If the requirements to provide that state changes, then the implicit changing with it is good thing.

[wanton orchid]

systemd devs can't know better than you how to setup your unit
so stop saying so
and require people to understand they are the one responsible for setting up their unit

[twin dew]

When disabling the implicit is made easy too.

[twin dew]

Which is why that implicit dependencies is documented so well and can be easily disabled, IF your unit needs that.

Because it works for most shit.

[wanton orchid]

there is no such thing as "changing the default suitable for unit already running good on older defaults" because these units have been made/tested on the older default
changing that is breaking
there is no such problems if you are changing a template instead

making a template for default is pushing people to know what the shit they are actually doing

[twin dew]

And how would a specific template differ from just providing the various examples in the documentation that the SystemD already does?

If there wasn't implicit defaults for "start only after base system is set up" (After=sysinit.target basic.target) and "quit on shutdown before base system is taken down" (Conflicts=shutdown.target, Before=shutdown.target), then all those examples just would need those lines added to them instead.

And the documentation says in very early before actual contents:

Only services involved with early boot or late system shutdown should disable this option.

Specifies that default dependencies are a thing, what they are, and how to disable them, in very long form.

And protects from lot of cases of user stupidity.
Like the Console-Setup removing those and not replacing with suitable other units, thinking "We are early boot stuff!".
But then failing on some boots as file system and/or console devices aren't correctly up yet in race condition situation.

IIRC it was that /dev/tty:s weren't there yet when console-setup tried to set settings for them.

On some boots.
Causing the on-physical-computer text consoles to not always be set up correctly.

When it doesn't matter when those are set up, as long as it is before you can actually get login prompt from getty.

And IIRC gettys only get started after the system reaches multi-user.target, unless there is a problem.

So I just made override to add that "default-dependencies=yes" back in, so the Console-Setup happens between that basic.target and multi-user.target.

[wanton orchid]

I already explained why it is fake protection of user stupidity

and in the end just lead in more stupidity

[soft bloom]

is crowdstrike even good?

[mental oriole]

Idk, never heard of.

[edgy hazel]

It's big because investing into proper security is expensive. The product itself is mid

[tough owl]

https://fxtwitter.com/9to5Google/status/1814016979809771982

another one for the grave

[twin dew]

And seems I really need to do more research on Xen vs KVM front...

[tough owl]

In at work and everyone in the office that doss finance stuff says they can't work right now. I presume because of the CS issue

[twin dew]

Ok, the Debian support for Xen is not usable, as the Debian Xen packages don't really support SystemD.
Upstream Xen provides SystemD support but the Debian implementation leaves those out, trying to currently read why.

Because way back when the Debian Xen team made stuff to support multiple versions of Xen installed at same time and are still running that giant patch set, and haven't updated it to work with SystemD...

The idea why they do support multiple ones is very good, but that they haven't updated to SystemD support is just...

Completely up to you?

That isn't hot for current CPUs.

Seems the SysV Init script rewrite was completely normal thing, as the devs involved say that the upstream ones are complete mess and aren't responsibility of anyone.
And just hope that someone from Xen supporting company gets assigned to actually work on the SystemD units to make something similar so that failed upgrades etc. don't kill everything.

And the current Debian team just doesn't have the resources to do the same to the upstream Xen SystemD units.

So KVM it is then.

[jagged snow]

Kinda wild to have the MS and CrowdStrike outages in the same day

Sounds like it sent all of you europeans for a spin

[twin dew]

What MS outtage?

AFAIK just all Windows machines with CrowdStrike installed getting BSODs.

[jagged snow]

MS 365 was down overnight

[twin dew]

Ah.

Xen would probably be more secure, but as I'm not selling VMs, that extra-extra-level of secureness isn't worth the price.

[tough owl]

Seems like a struggle in Windows land. Glad that isn't me

[jagged snow]

The perpetual struggle to stay off of linux so that I'm actually productive

[twin dew]

But I went and bought two 64GB Samsung Pro Endurance MicroSD cards for the Pi:s.
As those seemed to also have great random read and write performance in tests people had run, better than most A1 (Application) rated cards.

Below the A2 rating requirements, but cards with that are rare.

Actually a 32GB Samsung Pro Endurance gave better IOPS for reads than A2 rated SanDisk Extreme 64GB.
And the SanDisk didn't actually fullfill the A2 requirements in the test.
But might be down to the high-end USB card reader used, from the USB part.

Also won in both against Lexar Professional U3 A2 V30 128GB 🤣

That Samsung EVO Plus is actually marked with A2 now.

On Samsungs pages for it.

These Samsung Pro Endurance cards are meant for dashcams and surveillance cameras etc. that do constant writing, with exceptional endurance ratings.

[languid gulch]

so as it turns out, starbucks is down too along with everything else 🤣

[twin dew]

First with Windows defaults with default exFAT formatting on the stick, with my USB card reader.

[tough owl]

What will all the white girls do with their daily venti mocha iced latte 💅💅💅

[pure karma]

be obnoxious probably 🤷‍♂️

[languid gulch]

i was thinking more about all the techs that run on caffeine

[tough owl]

monsters

https://tenor.com/view/monster-energy-drink-smoke-can-gif-15042566

[languid gulch]

lol

[tough owl]

or go to McDonalds for coffee

[twin dew]

And if McDonalds was also affected?

[tough owl]

local coffee shop

Idk ones locally aren't

developer.nvidia.com/blog/nvidia-transitions-fully-towards-open-source-gpu-kernel-modules/ cmon guys. Do the rest of it

[twin dew]

Doesn't seem to actually mean much, except that Noveau will get better:
https://old.reddit.com/r/hardware/comments/1e5xumz/nvidia_transitions_fully_towards_opensource_gpu/ldrnhez/

[tough owl]

Oh yea. It's not a lot but ifs nice the kernel modules are open

if only the vulkan driver portion was open

[twin dew]

As in that Noveau can now use the full Nvidia blobs and not cripled ones that Nvidia deemed to give them with suitable license for distribution.
And use the Nvidia driver as technical documentation replacement instead of pure guesswork.

[languid gulch]

yea i first saw it as all the mcdonald's in japan having issues

and was like "wtf"

[tough owl]

I'm so glad I turned my work computer off last night

[jagged snow]

Running CrowdStrike?

[tough owl]

IT just showed up to fix the issue here 🤣

[jagged snow]

Ah nice

[tough owl]

Whole school is 💀

[jagged snow]

Man 💀

[tough owl]

everyone who just let it sleep got the issue

[jagged snow]

Ah

My pc just sleeps overnight on weekdays

Usually gets shut off on weekends

Tops out at like 20 days of uptime

[tough owl]

I bet they'll issue a group policy to have all computers turn off overnight unless the person who uses it submits a ticket to let it sleep overnight

[jagged snow]

It's honestly probably good practice

[tough owl]

Save them on power slightly so they don't have awful windows sleep sucking down power

Or if the computer detects sleep for over 3 hours it'll shutoff if that's even possible

[jagged snow]

It is

[tough owl]

then that's what I'd do

But hey I'm not the head IT guy

though I've had him message me

i can't imagine how long it'll take them to hit computer lab on campus. I don't think those shutoff. Just go to sleep

[jagged snow]

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

[twin dew]

You didn't notice I linked it little under 12h ago?

[twin dew]

AMD doing fuckery with the chipset naming...

Dual-chipset non-E doesn't exist, and the single-chipset E has been upped to be named X870.

So now X870 and X870E have major differences on how much IO they can support, as first is single-chip, second is two-chip.

Now the defining difference is that the X870:s must support USB4 and PCIe 5.0 for both M.2 and 16x slot.

When X670:s were dual-chip and E:s were PCIe 5 for both.

[wanton orchid]

that part makes it more clear though

it ties hardware with name

[jagged snow]

I was asleep, sorry

[wanton orchid]

btw how far can a bunch of added io be useful with the static x4 pcie 5 cpu lanes ?

[twin dew]

How so?
X870E: two chipsets, USB4, "full" PCIe5.
X870: one chipset, USB4, "full" PCIe5.
B850E: doesn't exist.
B850: one chipset, USB3, partial or "full" PCIe5 depending on MB maker choice.

[wanton orchid]

E vs non E still is name noticeable difference

the big misleading part is B840

[twin dew]

Point was that using B850E would have been more logical for the one-chipset "more MB maker included features" variant.

To go with the 600-series naming convention.

And leaving X870 not used for now.

[wanton orchid]

they can't make every generation E be having pcie 5

pcie 5 is becoming standard now

[twin dew]

Point was that 850 would have been single-chip, 870 be dual-chip.

Like on 650 and 670.

[wanton orchid]

I never understood it that way

it just happened to be dual chip

[twin dew]

And PCIe 5.0 M.2 slot was standard on 600-series too for all four at start.
It was later dropped from the base B650-non-E

That was the difference between those two on 600-series.
X670 had more IO as it had two chipset chips.
B650 had less as it had only one.
And then E denoted raised PCIe 5 requirements from just single M.2 slot.

E just meant that the MB HAD to have 1x PCIe 5.0 M.2 and 1x PCIe 5.0 16x slot.
Non-E had to have one of those at first, and later B650 had the PCIe 5.0 requirement completely dropped, as that raised minimum MB price significantly from need to use better PCB material.

[languid gulch]

i wish for once that anyone would stick to a naming scheme

[twin dew]

So that X870 vs B850 being again two chipsets vs. one chipset would have made much more sense than this marketing clusterfuck.
And then suffixes for the other differences.

[languid gulch]

there's also no way in hell they're coming out with literally 100 different model names of motherboard

[twin dew]

No, but the rest they just marked was the USB4, and PCIe 5 partial or full.

[languid gulch]

there's not gonna be a B810

[twin dew]

And now basically X870 means "Must have USB4".

With the more or less IO being delegated to the E suffix for that.

[languid gulch]

which they could have called something like X945 for "overclockable, ryzen 9000 series, USB4, PCIe 5.0" or something

[twin dew]

And they skipped 700-series just to get up to same as Intel is for chipsets...

[languid gulch]

yup

[twin dew]

Or will be with the next gen.

[wanton orchid]

isn't just Intel using B700 already ?

[twin dew]

14th gen uses 700-series chipsets.

[wanton orchid]

what would be the point of naming your product the same as competitor

[twin dew]

B760 and Z790.

[wanton orchid]

you see

[twin dew]

But they also used B660 and Z790 before that

[languid gulch]

did Intel retire the H boards?

[twin dew]

And B560 and Z590 before that.

[wanton orchid]

So naming B750 would have been clash