off-topic-tech

[twin dew]

Example of actual Intel 13th and 14th Gen desktop CPU RAM speed support on 4 slot MB:

[dire igloo]

It was mostly true on DDR4, but I've seen DDR5 be unstable at JEDEC speeds

[twin dew]

It was there even with DDR4.

[winged valley]

Is the 2 R referring to 2 ranks of DIMMs or 2 dual rank DIMMs?

[twin dew]

First is single 1R or 2R DIMM per channel, and then 4x 1R DIMM, then 4x 2R DIMM.

AMD 5000-series:

[winged valley]

Gotcha

[twin dew]

AMD 7000-series:

[winged valley]

So then running 4 sticks just has both channels operate 2 sticks instead of 1? I think

If I'm following this right

[twin dew]

Yes.

[winged valley]

Ah okay

Yeah I can see how that can impact speed

[twin dew]

And the second stick on channel causes more load, which limits the stable frequency.

[winged valley]

Data load or power load

[twin dew]

The load on the bus drivers.
Data load.

[winged valley]

Because couldn't power just be solved with more rails?

Okay that's what I thought

[twin dew]

Need to pump more amperage into the bus to get the capacitance fed to switch the bus state fast enough.

So with DDR4 2DPC was a problem.
And with DDR5 it is basically dead if you want usable RAM frequency.

Also goes in part if the MB uses Daisy Chain Topology (one DIMM slot after another) or T-Topology (traces go to between the slots and split to both).
At start of DDR4 time some MBs were still using T, but near the end of DDR4 and now with DDR5 all are Daisy Chain.

Daisy Chain is better with just one stick.
T-topology was pretty good with two.

[winged valley]

Baldur I gotta ask how do you know so much about low level hardware properties and theory

Are you an engineer?

[twin dew]

So to get max one stick per channel frequency, daisy is better.
But to get max two stick per channel frequency, T would be better, but as it loses on one-stick, and is harder to design, it went away.

[twin dew]

Now to see how many decades that 25g syringe will last for me...

But finally time to run the comparison between the current setup with ceramique 2 and non-offset cooler first.
And then try the Noctua pastes with offset cooler, then to ceramique 2 with offset mounting.
And then do test with that with lessening amounts of paste until results start to get worse.

[narrow folio]

how about a test with ptm7950?

[twin dew]

Don't have any right now.
Will try to get some later.

[gilded helm]

Is it widely available yet?

[twin dew]

No, as it is not officially retail product.

[narrow folio]

Somehow 'everone' is hyped about that stuff

[twin dew]

Been waiting for the Thermalright Heilos pads to actually get availability.

[twin dew]

It is great, but getting hold of it is hard.

[gilded helm]

I don't necessarily care if PTM is a top performer, I just think it will be a good recommendation for first-time builders who might be on the fence about doing it at all.

[narrow folio]

Kryosheet?

[twin dew]

As all that is available is someone buying in bulk and selling in snippets.

Thermalright Heilos pads should be the stuff in retail, but like normal, even the official Thermalright retailers cannot get the stuff from them.

Jimm's PC Store is the official importer and retailer for Finland, listed on Thermalright pages, and they have tried to get the stuff for 5 months now, since Thermalright announced the stuff.

[twin dew]

Not that easy to use.

[narrow folio]

PTM should last as long as you don't take off the cooler, right?

[twin dew]

Yes.
And even if you do take it off, it should be reusable by just scraping it and putting it back on the CPU.

[gilded helm]

Is it harder to get good mounting pressure with pads?

[twin dew]

Even when it isn't a pad anymore.

[twin dew]

It isn't really a pad.
It is just preformed "paste".

At around 40C it turns liquidish.

That is the phase change part.

And when the temperature drops again, it turns solidish again, and party pulls back in.

Which is what helps against the pump-out.

The PTM7950 strip is 0.25mm thick when provided, but expectation is that once it goes the first change to liquid, it gets pressed down by the cooler to less than 0.05mm thickness and the extra is pushed out.

The stuff is also available in cans for "printing":

The PTM7950 datasheet:
www.caplinq.com/index.php?option=com_filecabinet&cid[0]=404&lang=en&task=download&tmpl=component

[twin dew]

Anyways, start 10 minute specific Y-cruncher stress test, wait 1 minute, reset HWiNFO, start 8 minute timer, take results as screenshot. Let the test finish.
Take temperature measurements from inside the case and room temperature at intake level before and after.

[twin dew]

Now thermal cycle on the H1 with offset mount, before real test.

[twin dew]

And same for NT-H2...

[twin dew]

And the Ceramique with same type of paste application as with the other two...

[wide void]

anyone here a whizz with Apple?

[night girder]

better to just shoot away your question 😉

[wide void]

I've given up, its the spawn of Lucifer

[night girder]

what's up?

[wide void]

they can pay to get it fixed as penance for buying the fucking thing

[wide void]

trying to update from El Capitan to Catalina

[night girder]

Ok, and what is going wrong?

[wide void]

its not having any of it

[night girder]

Oh. You get an error?

[wide void]

its not showing under updates

[night girder]

Huh 🤔

What model? And are you trying to update from appstore or from the system settings / preferences ?

[wide void]

a quick look online suggests you have to do an upgrade to a mid point (Sierra?) as the file system changed at that point, so the newer updates aren't compatible with anything older than that one

[wide void]

macbook pro, from 2013ish

but even Sierra isn't showing in app store or from preferences

[night girder]

MacBook Pro (Mid 2010 to 2015) - Siera compatible. So that's not it.

[twin dew]

Because High Sierra is first one to be on App Store.

[night girder]

"And are you trying to update from appstore or from the system settings / preferences" ❓

[twin dew]

Update download link in there for Sierra:
support.apple.com/en-us/102662

[night girder]

and I wouldn't go High Sierra. I mean do what you want. But if you are on El Capitan, might be safer to go El Capitan -> Sierra -> High Sierra.

updates-http.cdn-apple.com/2019/cert/061-39476-20191023-48f365f4-0015-4c41-9f44-39d3d2aca067/InstallOS.dmg - link to Sierra 10.12 dmg

[twin dew]

The instructions how to use were on the site next to links, so linked that.

And it has "All" the versions.

[wide void]

righto, ty. I'll give it one go then let someone else worry

[night girder]

I remember when Apple OS upgrade just bricked my whole macbook.

[wide void]

I've already backed it up just in case, lol

[night girder]

like a true Apple user would, haha 😄 (or any smart person)

[wide void]

in truth, I do the same for windows..

[night girder]

This was the error I got back then, pretty funny. (Image from google, too lazy to look on my NAS).

[wide void]

on my old rig there was one update that just took it into a restart spiral that I could only stop with a reinstall...

so its not just apple

[night girder]

Yeah, I I think I was able to fix a PC crash with GW2 with a clean OS install of W11 😒

[twin dew]

No real difference so far, and no real benefit in my case from the offset mount.
Need to take the cooler off again and add little more paste, think I went too far in minimizing the amount on the latest spread.

[night girder]

Testing different pastes?

[twin dew]

Quick test.
Not proper enough for real comparison.

First was ceramique 2 on normal noctua mount that had been there for long time already.
Rest are with the noctua offset mounting.

[edgy hazel]

support.apple.com/downloads/macos
tried on of these yet?

[twin dew]

The three with paste names were with Noctua suggested dot pattern application.

[night girder]

Was about to say. Isnt the digference in paste the quality, aka degredation over time

Comparing pastes performance over 6 months seems to be more usefull

[twin dew]

There are differences in performance, but they are pretty minor.
And that was the point, I know the Ceramique 2 is easy to apply and lasts for 10+ years.

Which is why I bought that 25g tube of it.

[wide void]

I've given up, they have forgotten their apple ID and it just says 'unable to proceed' or something when I try and change it

[twin dew]

Wanted to see if the Noctua pastes I had left over would make difference or not.

[night girder]

…

[twin dew]

At the new stage.
Of course if I added more temperature cycles, there might start to be more difference.

[edgy hazel]

I think my local supermarket has started to use Ai image creation

[wide void]

you never know, that bit of rosemary might actually be cast into the glass.. lol

[night girder]

One way to find out. But I mean it would surprise me. How much difference can there be in paste? How much headroom is there to improve it.

[twin dew]

Just if it started to pump out to some extent and the bond line got thinner with time.

[night girder]

I would expect differences of 2-3° celcius but not 10° celcius for example.

I am speaking of new paste btw. Not over time. Just to be clear 😄

[twin dew]

And yes, the last test didn't have paste contact on far corners on the IOD side.

So final pre-stress heat cycles with added paste and then final test.

But also shows that I don't specifically benefit really from the offset mount so far.
And so I can remove it and switch back to standard mounting, if the fan mounting wire clip is too close to GPU PCB back in final installation.

[twin dew]

Final for now:

So switched paste, made two load runs, let the system cool.
Started 10 minute Y-cruncher BKT test, waited one minute.
Cleared the HWiNFO, took screenshot after 8 more minutes.
And took measurements from physical room & outdoor temperature meter with the outdoor sensor inside the case in the suction airflow for the CPU cooler.

Before the start of the load, and after taking the screenshot.

So using NT-H1 or -H2 might give tiny bit better temperature, IF I bothered to make perfect paste application.
Which isn't really possible with -H1 because of the consistency.
And never tried to do with -H2.

And as there isn't any realistic difference that couldn't just be normal variance, not going to even try, as I know the Ceramique 2 doesn't care about time at all.

0.5s polling interval on HWiNFO.

[night girder]

I am still curious of quality of pastes.

They all seem to cool. But what is the drop off over time.

[twin dew]

Runnier the paste, easier it is to get good result with it when you just add some and plop cooler down.
But the same runniness makes it get out from between the cooler and IHS with time.

Or some dry out, and with dry paste any movement of the cooler breaks the contact badly.

So often the "best in quick testing" pastes will be crap over time, and need constant reapplies.

[night girder]

Wonder if anyone ever did that. Apply different paste on multiple mobo’s. Measure them over months.

Will check later on google to find some data. Not home atm.

[twin dew]

Not that I know off.
And I have specifically tried to look for that kind of testing.
As I thought I couldn't get more Ceramique 2 anymore.

And went looking for something to replace it that could do 5+ years without problems.

That top result was with cooler installation I did when I got the replacement CPU.
And the MB was later handled by the cooler when it was installed into case.
And has gone through weeks of constant high load with the stress tests.

But hasn't been on nearly the full 6+ months.

By top I mean that Base result without offset mount.

And because the heat output of 7800X3D is so low, the differenced get even smaller.

[edgy hazel]

https://www.youtube.com/watch?v=FdtJResj1A8
goes hard

[soft bloom]

is it any good of an idea to store lots of metadata in filenames, or just go with JSON db file? (sql not an option)

[jagged snow]

Depends on a lot of factors

Whats the use case?

I typically wouldn't use filename for more than a timestamp

I hate this kind of transition with a passion and I hate that Adobe popularized it for big software https://arstechnica.com/information-technology/2023/12/broadcom-ends-vmware-perpetual-license-sales-testing-customers-and-partners/

[edgy hazel]

ouch

[soft bloom]

to know what's inside, haha.
primarily - timeframe of the data, and timestamp of when file was created

essentialy, to be able to continue work, sourcing progress from these filenames

trick is that you can't just store somewhere "we did from date A to date C", cause in reality it might be a lot of a-b, c-d etc timeframes within the large target timeframe

[edgy hazel]

as long as every file has a unique identifier, I don't see a reason why a database isn't an option

[soft bloom]

because it's not a web-development environment, and some other constraints

*i'm from web development, but this is quite a different scenario

big HW question:
if i were to put my pc behind powerstation (you know, that kind with a battery, ac plug and a bunch of USB ports)
how quickly would it's battery degrade if at all?

I don't like it beiung like this because it often wants to cool down, making noise, unlike some cheap UPS i was using before.
but that ups seems to be dying, few times it just turned off without allowing to bring power back unless i disconnect it from the grid...
i heard that batteries eventualy degrade, but not sure how true this is, and is it applied to pwoerstation as much

[wide void]

UPS will have a rating somewhere in the tech specs about charge speed

[soft bloom]

true, but i don't see yet how its related...

[wide void]

as long as your draw is lower than that, the battery won't discharge in normal use

or am I misunderstanding your question

[soft bloom]

i don't think my UPS could support discharge when connected to grid - i believe it can only work with 1 source (grid or battery) at a time. it's line-interactive.
or are you calling powerstation a UPS too? i might be confusing.
quesiton was about if frequent use of powerstation will degrade its battery (capacity)

[soft bloom]

wait, why would battery discharge at normal use at all?

[wide void]

if you are pulling 1kw from your UPS that only charges at 500w, its going to go down

[soft bloom]

i see where confusion comes from.
powerstation, when connected to grid, passes it to its clients. so battery charges separately

[tough owl]

FOSS FTW

[wide void]

the 'UPS' output is generally not 'straight through', you are always pulling off the inverter behind the batteries

[soft bloom]

at least, that's how it was configured. maybe there's a setting to break the link, but for now its like that, i guess its good enough for pc

[wide void]

ahh OK, not familiar with that distinction!

is it still a seamless switch to the inverter?

[soft bloom]

uh, i can't link the manual

[soft bloom]

i an say that i didn't notice problems with powerstation (idk how else todistinguish betwee the two) nor UPS

UPS claims to have up to 7ms delay

[wide void]

having a 'mains' passthrough is a much better way to do it as long as the switch is quick

[soft bloom]

powerstation doesn't define that time, but i found confirmation in its manual that it will pass-through AC power

[wide void]

mmm, 7ms is pretty quick, most PSU's probably would not notice the break

the smoothing capacitors would have enough juice hopefully, lol

[soft bloom]

especially since i have PSU that is almost twice the requirements under normal working condition

too bad its known for its explosions 🙂

[wide void]

lol

[soft bloom]

but, despite powerstation claiming to passthrough ac, it kicks up fan periodically

and this is what got me worrying about using it on daily basis.

[wide void]

how frequently, some sort of discharge/charge cycle on the backup battery? (to maintain capacity/life)

[soft bloom]

main problem, including even JSON-file-db idea, is that i'm not sure if it is going to be good idea overall, haha
i fear things like concurrent program execution, abrupt stops etc.
file is a good thing because if it exists, it has the name and so all the infromation you stored inside of it.
i was thinking about JSON dumps (or sql for that matter) for some time, but it seems both more complicated and risky.
i think i'm missing something

[jagged snow]

@soft bloom what lang are you using for the project?
If you use Java, there are some very good sql libs

[soft bloom]

https:// www.bluettipower.com/pages/bluetti-eb3a?view=eb3a-spec
3-6 months
2500 cycles total (which shouldn't matter when power is on, which is sort-of always)

[soft bloom]

python.
yes, using sql is easy. but that would be a dependency (system one) for a minor part of the library (to be fair, that minor part already introduces some optional packages, not system onces though)

[sharp oasis]

I was reading how Arm makes the clusters work. And it's actually incredibly ingenious. And ik this is tech chat. But I can apply what I read to satisfactory in a sence

[stray badger]

Why does that need to be applied to satis?

[soft bloom]

that would be more of a math-and-meta talk

btw i believe that's how super compact bps in DSP work

[sharp oasis]

Item sorting and rudimentary logic

As the way I've been sorting items is garbage compared to that

[soft bloom]

without explanation its just teasing and tricketry

[sharp oasis]

Basically. I and a few buddies made our own method of mass sorting and routing items at high speed. And with attempting to sort even larger numbers it's becoming quite a task to design systems to sort it. I was inspired by how Arm's big LITTLE system works and supposed I tryed to recreate it with satisfactory and sorting items. Having big high bandwidth sort cores and much smaller lower bandwidth ones. Instaid of building all large ones. In the world we have made we have used zero load balancing so this could resolve some problems with under utilizing certain areas by shifting the load thru different speeds depending on the bandwidth received.

[pure karma]

challenge accepted

or should i say practically completed...

[sharp oasis]

So basically I'll be using LITTLE cores to deal with say a quarter train load. Instaid of using the big ones. As it's seeming to be an issue with the way I've come up with to sort items if there is a large number of one item it can reduce the speed of a system from peak speed. So Makeing wider bus systems for the large sort cores. Currently I've devised a 6 lane sort core that has a 2:1 sort ratio. (every two lanes in is one lane out (basically it will have a reduction in unsorted item bandwidth)) and it could have problems with being saturated if we only use it all the time. Now if we use a bunch of smaller single to dual 1:1 sorters beside the big one. Lower bandwidth but each lane is 1:1 (no reduction in unsorted item bandwith) making it more effective under full saturation. Now the bigger cores are really fast under 100% load. Being able to sort effectively. But as soon as there are many items that cannot be sorted that does pose a potential bottleneck.

That and using some basic item logic to determine the amount of bandwith we need rather then building a core per station. Saving me space, time, and unessasary work.

[soft bloom]

is this about sushi trains?

[sharp oasis]

No idea what that even is

[jagged snow]

Sushi trains carry more than one type of item per freight car

[sharp oasis]

Oh

Yeah

We always run sushi trains then

[night girder]

https://tenor.com/view/food-hot-dog-sushi-train-gif-11928848

Sushi Train? 👀

[sharp oasis]

Apologies I have like never learnt anything about what the community has called these systems. And basically this is all innovation based on the three of us dudes all trying to solve a problem.

[pure karma]

whats the point of a front usb-c conector if it dosent even work

[jagged snow]

Is it plugged in?

[pure karma]

yea

[rotund hamlet]

at the other end

to the mobo

[pure karma]

unless they just iddent include the wiring for it all the front usb wiring is conected to the mobo

all the other ports work including the power button so i guess its just broke

the motherboard one works so i guess atleast i have one

[jagged snow]

Many motherboards don't have a front-panel usb-c connector... and they can be a little annoying to plug in properly

[pure karma]

yea double checked all the front ports work other than the usb-c one

[pure karma]

is it the anoying vertical cable that like pokes outward really far and that is really anoying?

or another one

[jagged snow]

That describes a lot of them

[night girder]

What case?

[pure karma]

dont have another shot of it so thats that

[pure karma]

H9 flow

il check the wiring but everything else is working so i doubt its the port seeing how i have nerver used it

[pure karma]

yea everything is plugged in

[night girder]

seperate cabble it seems, looked it up in the manual.

[pure karma]

so its dead

[night girder]

could try replacing the USB 3.2 GEN 2 TYPE-C cable.

[pure karma]

or rgb is drawing to much from the second usb-c thingy

[night girder]

second usb-c thingy?

[pure karma]

well theres two on the motherboard one is hooked up to a rgb hub the other one is hooked up to the front pannel

[night girder]

should be fine.

[pure karma]

yea just double checked both cables are in fact plugged in

do they perhapse disconect on the io side or something?

bruh

the rad mounts cover the io screws so you have to take out the entire rad

[edgy hazel]

wish I had your issues

imagine having to disassemble your entire cooling loop and rebuild it differently because you dared to get a new GPU that is bigger

[night girder]

isn't air cooling handy

[edgy hazel]

yes but it isn't as cool

[night girder]

True, but cool enough for some

[edgy hazel]

tbf a dark rock pro is fucking sexy

[night girder]

I find that one pretty sexy too 😄

[edgy hazel]

nah that's too tiny

tiny man cooler

[night girder]

that's what I wanted.

[edgy hazel]

a tower made for ants

[night girder]

So I can access my RAM atleast.

I am very happy with it. I also got good temperatures.

[edgy hazel]

My raspi has one like that

[night girder]

sure

[night girder]

yup, that's a noctua

[edgy hazel]

yup

[night girder]

I though you throw the raspi in the bin by now

[edgy hazel]

same size as it too

[night girder]

probably same prize too

[edgy hazel]

nah it's currently running my bing bot

[night girder]

ah, good usage

[edgy hazel]

need a proper server tho, something where I can add my old 5700xt into, so I can run my own LLM

[night girder]

you said the mac mini, or try to find old hardware cheap.

would be pretty dope project

[edgy hazel]

I mean I just need a mobo with a cpu and a case

[night girder]

yup

[edgy hazel]

I might have some old xeons in the attic

[night girder]

you don't even need a case technically

[sharp oasis]

+205mhz core and +1000mhz vram on a 1660 Super gets me almost 60fps in satisfactory max settings 1080p

Wile still staying very cool at 58c

And insanely efficeint at 135w

[jagged snow]

Not that effecient

Less than 50% of the performance you'd see with the same wattage from a modern card

[sharp oasis]

fair

im still amazed by thw temps

[languid gulch]

oh, one thing that was kinda weird with all my drive swaparound stuff: for some reason it's never liked to keep reBAR on, but now it's fine

[sharp oasis]

Yeah

Does the 1660 super not support reBAR

As it's enabled on bios

But says it's not enabled on gpu-z

[languid gulch]

it's possible but i think you need to do a gpu flash

[twin dew]

Nvidia only supports it on 3000 series and up.
And for early 3000 series units there is firmware update tool.

[edgy hazel]

you do know that hwmonitor can pick up icue sensors withoput you having to install icue

[night girder]

that's why I avoid corsair.

[tame swan]

is that rpi 5?

[night girder]

2 afaik.

[edgy hazel]

2b please

[night girder]

B2?

[tame swan]

hmm why did they change hdmi to micro hdmi in newer versions

[night girder]

space

[tame swan]

i bought rpi 4 recently, ngl i kinda feel like i might have bought on spur of moment

[night girder]

Rumor: GTX 1650 & 1630 production will stop in 2024. Ending the GTX line.

[tame swan]

welp, looks like it is possible to run pretrained LLM on rpi 4 so i will give it a chance and try to make smh like a home assistant

[night girder]

can do all sorts of things with it 🤷‍♂️

[tame swan]

I bought it for hosting my bots / learning linux but it is kinda pointless since i had to buy 24 + 6GB dedi anyways

but since i am interested in electronics might be worth, idk

[edgy hazel]

possible doesn't mean good

[night girder]

bro, you got 2B.

[edgy hazel]

been looking into it the past couple of days. You basically can forget running any form of LLM on any rpi model

[sand saddle]

Yes

[night girder]

Why wouldn't LLM be able to run on a PI?

[sand saddle]

Avoid corsair, not like they make anything meaningful anyways. Their best products are simply rebrands.

[night girder]

Sure, you might need to attach a few drives to stash away all the training data

[tame swan]

yeah but the project i am planning to do includes use of gpio pins so can't really run it on my computer

[edgy hazel]

mostly RAM

[night girder]

Does it crash? Or just slow AF.

[sand saddle]

Friend got a 4k oled touchscreen laptop. Very fancy stuff.

[tame swan]

Mini LLM is able to run, i also seen ones that was able to generate image using ai tho it took him 45 mins for 450 x 450

[night girder]

I want a PI, so I can turn it into a Humidity meter for my plants.

[tame swan]

why not use arduino for that?

[night girder]

because a PI can do it

[tame swan]

but an arduino is way cheaper if you don't plan on using anything else

[night girder]

Stuff like that 😆

[tame swan]

yea ik, i have a sensor as well for that

[edgy hazel]

arduino will be a bit harder to integrate into any IOT stuff.

[tame swan]

idk meaning of IOT but if it is a moisture level checker, can be def done in arduino

[edgy hazel]

Internet of things. Smart home stuff

[night girder]

yes, but some people just prefer other hardware. I am more of a PI guy.

The real reason is: PI is easier to say than Arduino. 😆

[edgy hazel]

fuck the AI is gonna turn on my lights and turn off AC 😦

I'm gonna fucken die when I can't read the moisture levels of my window plant on my phone

[night girder]

Found it

[edgy hazel]

🗞️

[night girder]

WTF - 150 GB available space!!!

[edgy hazel]

Why you think we're buying 4TB SSDs now?

[tame swan]

valid reason ig

[maiden coyote]

There's the pi pico. Pretty much an arduino, and about as cheap

[night girder]

The crash is back 😦

[edgy hazel]

change your name then

https://tenor.com/view/betty-white-golden-girls-hands-gif-24420157

[night girder]

🖕

[edgy hazel]

ima say it's storage

gw tryna do something it can't so the system just goes haywire and dies

[night girder]

fuck me, it ran 2 days without an issue.

All I did today was install steam 😒

[narrow folio]

try to cool the steam a bit and install water instead

[edgy hazel]

that's why I'm iced out

[night girder]

Yeah, or just sell this build and make another one. Sick on tired of this AMD piece of shit.

[edgy hazel]

shit of piece

[night girder]

4-6 days of debugging. Then it rans fine for two days while running the game.

I install steam to download BG3 and let GW2 run. Come back upstairs and PC crashed.

[edgy hazel]

have you ever checked the GW2 logs?

[night girder]

Fuck. Pc wont voot anymore

Switched around the RAM sticks and it wont boot

[edgy hazel]

reseat ram, clear cmos

[narrow folio]

2 or 4 sticks?

[night girder]

Dont have jumper for cmos

2 sticks

[edgy hazel]

screwdriver

or you mean you don't have any pins?

[night girder]

Nah just the jumper to put on

[edgy hazel]

yeah coin, screwdriver anything metal will do

there are jumpers just for that?

[twin dew]

Which error light lights up?

There are four error lights above the ATX connector.

[twin dew]

There are specific parts that are used to permanently bridge such pin groups, and the ASRock manual shows one, even when that isn't really what is needed for this case.
Just leftover from time when there were actual settings jumpers that is still in the MB.
When the Clear CMOS one is only one left on the MB.

But this part is the important one if the system doesn't boot:

But pain on ASRock as they are all the same color, and so bright that it is very hard to see which one it is.

[night girder]

Dunno hard to see

I see two lights. Cmos didnt do much it seems.

[twin dew]

And the word "jumper" actually means the bridging extra part.
Not the place where such jumper can be inserted.

[night girder]

It was what the manual said

[twin dew]

So CPU and DRAM on.

[night girder]

Yup

[twin dew]

So did the CPU die then for some reason?

[night girder]

Didnt do anything.

Wait if the light is on it means it died?

VGA is not posting.

[twin dew]

That the specific thing isn't working according to MBs logic.

The light turns on when that part should be starting, and turned off once it is functional.

[night girder]

Huh, all i did is swap my ram around 😒

[twin dew]

So CPU light on: CPU is dead.
If the CPU was ok, but RAM was dead, then the CPU would be off and DRAM only would be on.

At least dead on the current settings, and if you reset the CMOS, then it is dead at stock settings.

[night girder]

So I killed my CPU? By dwitching my RAM units.

How do I know if CMOS worked?

[twin dew]

Might have already been dying and causing the issues.

[twin dew]

You cannot if the system doesn't start.

Remove power cable, remove battery, short the two pins for 10 seconds with screwdriver.
Reinstall battery, reinstall power cable, try again.

And you can remove the RAM and GPU etc. and try without.
So that they don't cause issues on the CPU detection.

In that case the CPU light would go out and DRAM or GPU etc. turn on and stay on.

Once the CPU booted and tried to boot DRAM.

And GPU should only try to start after DRAM is initialized and working, turning on, and then off.
When both CPU and DRAM are working.

Ok... Seems that ASRock keeps the CPU light on even when the problem is just RAM...

[night girder]

Oh…

[twin dew]

Pretty bad idea.
But at least one case of ASRock AM5 board where bad RAM kept both on.

Asus only ever has single light on at a time.

[night girder]

So RAM is fine?

Its VGA?

Oh wait.

It posted!

[twin dew]

No.
I said to remove RAM and GPU and see what the lights do.

[night girder]

I did

[twin dew]

The first long post after RAM replacement that can take several minutes?

[twin dew]

"Nice" for ASRock to have link to Tutorial Video about those lights in the manual...
ONLY as QR code...

[night girder]

Ok, we're back Wtf. is going.

So two days straight I have stable system of clean OS. (can be coincidene) ???

Today I let GW2 run as a test while I was working outside, I also downloaded steam to play BG3. So now I am thinking that Herg might be on to something, and it's the m2.

Also I noticed that when I plugged in my screen, my monitor (who is powere by same brick) goes on and off 🤷‍♂️

[night girder]

also without a bluescreen, I cannot minidump can I?

[twin dew]

Not when it isn't the OS doing the reboot.

[night girder]

yeah, keeps crashing now again 😒

And all that is happening in the background is steam downloading a game, so maybe it's my ssd? 😒

[night girder]

And there was a windows update yesterday:

[edgy hazel]

https://tenor.com/view/microsoft-edge-edgymicrosoft-microsoftedgy-edgyyymicrosoft-gif-23136106

[night girder]

I can try to plug my PC directly into the socket. So make sure it's not the powerbrick. I can also try disabling my iGPU. Or I can buy extra SSD, put GW2 on their to see if that helps.

[dire igloo]

What cooler is this?

[night girder]

#off-topic-tech message

[gleaming shadow]

Unless is a fancy powerbrick and not a simple hema one with a red switch, I cant imagine it being the power brick...

Btw, I am so happy I'm on Android...

The power to install my own APK files... Discord v204 welcome back!

Horrible new ui, why do they do that?

[night girder]

I mean at this point...

[gleaming shadow]

Yeah I came in at an end point, I have been noticing power drops,

If I may, what province? I work in north Holland and live in south, but in my home I see the lights drop for a milli second

And at work I see "dips" from 245vac to 233vac

And they had a power surge in north Holland last week

[night girder]

Nothing to do with power surges. This has been happening since I got the build, two months ago.

[gleaming shadow]

Headache machine it is

Nothing in the minidump?

[twin dew]

Something with damage, but no way to say what without part swapping...

[night girder]

Hard crash. Nothing being written.

[night girder]

Not an option afaik. Most I am willing to do is buy a new CPU next year The X3D or something.

[night girder]

Also, that's why I tried swapping my RAM around to check if it was faulty RAM or something.

[gleaming shadow]

No old PSU laying around to make sure that ain't it?

[night girder]

No

[gleaming shadow]

Is it after a while? Like power settings? C states, for power saving etc?

[night girder]

I don't know anymore.

Sometimes it happens a lot. Sunday I did clean OS. Then I had 3 good days without a crash and today it started again.

The only thing that changed between today and yesterday is: yesterday there was windows update I installed. Today I installed Steam and let it run in background. While it was downloading a game.

I am now messing around with my powerplan in Microsoft. Disabled putting the disk to sleep, monitor to sleep and PC to sleep.

Because it happened when I was AFK today, so might be a power plan thing. Long shot (99% it's not that).

[gleaming shadow]

There's also options in bios for that,

Long long time ago I had a pc who couldn't come out of sleep mode, had to hard power off to get it to turn on, hence I never use sleep anymore

[night girder]

Yeah, I read similar stories on the internet. By now I've read about anything 😄

[gleaming shadow]

And power settings always to max performance, I hate power saving in pc

[night girder]

It was put to performance. But then it crashed. So I put it to balanced. But then it crashed too 😆

[gleaming shadow]

The advanced settings have more options that dont change with those plans

[sand saddle]

That's stupid

It just increases idle power consumption

Doesn't increase load performance

Balanced it where it's at

[night girder]

Ok, I made a list with other solutions. I am going to download BG3, let it run for a day. If that game is stable, it is what it is I suppose.

Because then it means it's only GW2 <> my setup acting funky.

[jagged snow]

Chrome updated ui

It's terrible

[gleaming shadow]

I had bad experience with power settings turning off my stuff, and I have 3 PCs running, I don't care about idle power

[night girder]

Ok, so neither disabling PBO, turning of the powerplan or disabling iGPU helped. Since I just crashed when alt + tabbing. mhm. Plugged PC straight into the socket.

Anyone know a good way to view HWinfo logs? csv?

[dusk rivet]

i don’t mind it, will take a minute to get used to tho

i usually try to give UI’s a second to get used to before i say they’re shit

[jagged snow]

I just don't like the rounded aesthetic

I don't mind it in windows, but the way they've done it wastes a lot of space

Wasted space is bad

[dusk rivet]

yeah that’s fair

if they had just rounded stuff off without wasting space it would’ve been better

[night girder]

Baldur, what is GPU ASIC power?

[twin dew]

ASIC: Application Specific Integrated Circuit.
And in this case usually the total GPU power draw.

[night girder]

So, seems to be pretty high before the crash.

had HWinfo log to csv file. Disabled buffer.

[twin dew]

Try what happens if you lower the GPU power budget with the Adrenalin OC section?

[night girder]

Yeah, wanted to ask. We did that CPU offset remember? You said it can also be done to GPU?

The last crash was when I alt + tabbed.

I noticed that when I sometimes alt + tab, the game freezes for 3-5 seconds, but survives. So I assume this creates power spike to GPU, gpu has to work again boot up render and then the freeze stops ingame.

[twin dew]

That was just extra voltage to CPU.
There is the PBO Eco mode to lower the power budget for the CPU.
And then there is the TDP slider for the GPU in OC tools too.

And there usually is also voltage offset slider for the GPU too.

So you:
Can limit the power the CPU can use.
Can limit the power the GPU can use.
Can give the GPU more voltage.
As tests.

[night girder]

I was hoping there was a way to do the same we did to CPU, like the offset.

[twin dew]

Yes?

[night girder]

Because if I remember correctly, the offset had to do with load spikes and switching from load?

[twin dew]

That was what I was mentioning when I talked giving the GPU more voltage.

[night girder]

needs 1.10V, Vsoc gives 1.15V and 0.05V is lost.

[twin dew]

And this was the GPU power tuning:

Making little upward adjustments to those voltage steps would be to check if the GPU isn't stable.
And lowering the power limit would be to check if the PSU cannot provide enough.

[night girder]

So weird it ran for 3 days without an issue, 20 hours + game time. But I wrote your advice down as one of the next steps.

[twin dew]

1.150V on the screenshot from me.
So would example try adding 50 to each level, which would give 1.200V in my case max..

Or try first with for example -25% power limit, to close the PSU out, with CPU PBO at defaults, or even set to some ECO mode.

[night girder]

I did disable PBO to try that out, but it crashed.

[twin dew]

The CPU PBO difference is pretty small, and it doesn't really do spike loads.

Stock vs. PBO enabled I mean, on the CPU you have.

But just things you CAN test, not that you HAVE to test them.

But I really need to head to bed now...

[languid gulch]

was wondering why my system would ramp up randomly, & i figured out it's my antivirus. 0 reason for it to be pushing my cpu to 70C 🤣

[edgy hazel]

question is how long

it trying to scan zip files can very much take up some power

[languid gulch]

it's only for about 20s before the fans slow down, but man it reminds me of my old cooler that couldn't keep up with my old cpu

it was the "scan during idle" setting, so i just set it to a regular schedule instead

plus i finally got around to turning on PBO 😛

[edgy hazel]

yeah that'll do it lmao

anyway I wouldn't worry about it

[languid gulch]

beefy cooler but that fast demand will still make it kick the fans on

[edgy hazel]

antivirus is just doing it's job 😦

[edgy hazel]

guess it's time for WAAATEEEERRRR

[languid gulch]

i know, but it's still doing it

🤣 that might be up next depending on any changes next year

[edgy hazel]

https://tenor.com/view/kermit-the-frog-kermit-kermit-meme-kermit-water-kermit-waterfall-gif-24226895

[languid gulch]

might be going to a smaller & redesigned desk, & that might mean a smaller case

[edgy hazel]

oh you 100% definitely with absolute guarantee need to go water then

[languid gulch]

stealing that shit because i've never seen it before & it's hilarious

[edgy hazel]

no other way I promise

[languid gulch]

🤣

[edgy hazel]

go on. spend 400€ on water cooling stuff

and then another 300 in 4 months

[languid gulch]

nah, i'd probably just do a 280

[edgy hazel]

and another 200 in 8

[languid gulch]

& not custom

[edgy hazel]

BOOOOOOOORIIIIIIINNGGGG

[night girder]

Herg, what was your problem again with your SSD? Was just one drive?

[edgy hazel]

it was onedrive accessing multiple drives while also gaming

[languid gulch]

god i hate onedrive

[edgy hazel]

I got 365. No reason to not use it

[languid gulch]

i'm still trying to unfuck some of my stuff that still insists on saving to there

[edgy hazel]

just uninstalling should be enough

[languid gulch]

oh i tried that

[night girder]

don't have one drive installed. But was thinking what can go wrong if something funky happens to SSD. Shouldn't crash a PC should it? 😒

[languid gulch]

now some stuff is saving to some kind of ethereal plane

[edgy hazel]

it very much can

false data being used can def crash your shit

[night girder]

The thing is, I don't have steam running atm and now the game has been running for 2 hours. If I enable steam, it crashes. But the only thing steam is doing is downloading and writing to disk.

[edgy hazel]

tried another slot for storage?

[night girder]

I hope it's SSD or PSU. Farely "cheap" to fix.

[edgy hazel]

also had an issue once where using the OS on the lower m.2 slot caused hefty crashes

[night girder]

Also underclocked my GPU power usage by 10%.

[dire igloo]

PSU is an utter annoyance to fix

Unless you used extensions

Btw is it the pic you sent or is the bottom end of the CPU cooler fan further back than the top end?

[night girder]

How do you mean to fix? I am talking: buying bigger PSU 😄

[dire igloo]

Troubleshooting is basically "replace and hope it works" and replacing a PSU is tedious

[edgy hazel]

that's no tedious wtf

[languid gulch]

is there a site that lets you find a case based on specific parameters?

[night girder]

Yeah, my mood is pretty low atm

[edgy hazel]

you just unplug some shit and replace the fucker

it's like 4-5 cables

[dire igloo]

Unplug and undo all cable management you did

Also good luck doing EPS

Or anything you usually preroute

And then IF it works, redo all cable management

[edgy hazel]

do you cable manage with glue???

the point of cable management is to specifically make maintenance easier

[languid gulch]

hot glue is the bane of my existence

[edgy hazel]

hot glues you to the wall

[dire igloo]

Ever undid a zip tie (even if it was Velcro) just to take out one cable?
All the others go all over the place

PSU is number three on the most annoying to replace list after case and mainboard. Tied with CPU

[languid gulch]

hot glue is the tar & feather of the pc world

[night girder]

give me a few, will answer. Raiding atm 😉

[dire igloo]

A few what?

Something's just uncanny about it

[edgy hazel]

you are using the wrong velcro technique. every cable on it's own, velcro is done in a different direction each time so you can just "snap" em together

the "all in one" tech is garbarge and fake

(I do not do cable management)

[edgy hazel]

lmao

my currently favourite gif is

https://tenor.com/view/b99-holt-balloon-pop-gif-10094260

[dire igloo]

Wha

Share your wisdom with me

[edgy hazel]

blew your mind didn't I?

[dire igloo]

This sounds like some wizardry fuck

[edgy hazel]

might be worth ~~connecting ~~ velcroing together cables with the same purpose/location like fan cables or RGB cables

[night girder]

I don't see what you mean with the CPU fan. Seems fine.

[edgy hazel]

But most importantly I got invited to my first pen&paper d&d campaign

BG3 fucked me up and then I watched Critical Role and now I can't wait

[dire igloo]

Probably a weird perspective on the pic then

[dire igloo]

Anyone in there that you know?

[night girder]

yeah, was a quick pick for the boot

[edgy hazel]

all of em

[dire igloo]

Lifelong friend of mine started a campaign with uni friends, I eventually joined

[night girder]

have you seen the cartoon that CR made on Netflix?

[dire igloo]

Thing is: I helped with world and lore building, so out of character I know a fuckton of stuff that's about to happen

[edgy hazel]

I wanna go in blind and have fun and roleplay 😄

[night girder]

The Legend of Vox Machina, can recommend if you into Critical Role and DnD.

[dire igloo]

Nope.
Did you see Tales from the Stinky Dragon on YouTube?

[edgy hazel]

I just watched that yesterday

[dire igloo]

Me: "I know you guys need a cleric/healer. I'll be one and that's about all for minmaxing I'll do"

Also me: becomes Aaracokra Twilight Cleric

idek how often I got to cheat an encounter through flight

[edgy hazel]

I wanna be a druid because I'm german and it's expected for me to go furry mode

I'll build my lore about a s** dungeon

and whole grain bread

Nah I don't know shit but my first BG3 char was a druid and it was really fun

[gilded helm]

Be an arcane hierophant wu jen+druid.

[edgy hazel]

I do not know these words

[gilded helm]

I was just hoping that would make sense. I have no idea what game you're talking about.

(third edition d&d)

[edgy hazel]

it'll be my firast real D&D and I've watched a bit of yt d&d

[night girder]

I am not sure if I am good enough actor to actually roleplay DnD

just not my strong suite, to act like a character.

[edgy hazel]

now the question will be: 🍁 or 🍺

[languid gulch]

idk how a maple syrup flavored beer hasn't taken over the beer world out of canada. that shit would be amazing

[edgy hazel]

wellllll we have something called the Reinheitsgebot in germany

"According to the 1516 Bavarian law, the only ingredients that could be used in the production of beer were water, barley and hops."

[dire igloo]

Love this actually

Name the dungeon "pain in the ass"

[edgy hazel]

nah that's french

[dire igloo]

But a solid pun

[edgy hazel]

it'll be the "Vollkorn Dungeon"

[dire igloo]

Whole grain, whole pain

*whole grain, (w)hole pain

Even better

[edgy hazel]

lmaoooo

[charred pewter]

the g.skill ram i got are type-m, despite people on amazon in recent month claiming they got type-a... guess the bins are back to m now

also half the heatspreaders on them dont even touch the ram chips. typical g.skil crap, i shoulda went with corsair like i knew i should have

[night girder]

😒

[edgy hazel]

never had issues with g.skill. tho I have only had two sets of 2 sticks each which are both running in a 4 stick config rn

[charred pewter]

well so far no errors in memtest.... so thats good. vs my last set which spewed so many errrors I couldnt count them

[daring cedar]

This game have a high CPU usage, my graphic is bored while the CPU is short before a burnout.
Most stressfull is, when the hoverpac is equiped, the change between powered area an non powerd area - then the usage will rise by 15% and the frame rates drops to the ground - may I have to switch from a 5800X to an 7800X3D... Or just wait till the UCIe will be released

[dire igloo]

A CPU reliant game being CPU reliant isn't really surprising

Especially if you're in the lategame and/or have a high performance GPU

[daring cedar]

Sometimes surprising that the load isn`t redirected/shared to the compute cores of the graphics. Especially in this cases the parallel computing capabilities of graphics will be an game changer.

[edgy hazel]

CPU more like CPooU

[daring cedar]

With extended object limit your CPU beg for mercy 😨

[edgy hazel]

I got a 5800x3d CPooU. I'm fine. My poo is in another dimension

only limit is my coffee intake

[jagged snow]

That's not really feasible

[edgy hazel]

hey hey hey if he wants to code in assembly let him

[jagged snow]

And to an extent its already done in many games... systems like physx

[edgy hazel]

no offence. Can you name one that isn't dead/discontinued for years as an example next time?

[jagged snow]

No because they dont exist

[edgy hazel]

https://tenor.com/view/giggle-snoopy-laughing-haha-so-funny-amused-gif-17213549

[wanton orchid]

steam download use 3 things :
cpu
network
disk

does not sound psu
does not sound gpu here
though you should note that cpu usage can trigger a brittle gpu driver fall off (delay at wrong moment)

yes a ssd can cause crash if it is your system ssd
even simply too high delay for system process can crash your OS
if your ssd is faulty then the increase in pressure could also make it send wrong data, or having wrong behavior
depending if it is NVMe or SATA different behaviors

[pure karma]

could the cause of my not functional front usb-c conector be caused by the merged front io plugs

i have confirmed everything is plugged in where its supost to but still somethings not right

[wanton orchid]

what are you calling high cpu usage
if you talk about satisfactory the it uses at most 3 of the full 16 threads

[wanton orchid]

you seen to assume a graphic unit works like a cpu unit
if that was the case we wouldn't bother having both good cpu and good gpu

[jagged snow]

What're you on about? Satis has never been limited to three threads, and even less so in u8

It scales easily to 6 or 8 threads, and much more than that on large saves

[wanton orchid]

"scales"
❌
I'm not talking about just using some thread
I'm talking about fully loading it

[jagged snow]

As am I

And why did you ❌ scales?

That is the proper term for what the game is doing

[wanton orchid]

it's not scaling as it does not spread the whole load
only some of the work is done on other threads
so it's not scaling, more spreading than scaling
you have like : main loop, render loop, render off load, background graphic driver processing, background satisfactory sim processing, off load of satisfactory sim on 1 or more thread
I confused with dedicated server out of habit sorry
but it still doesn't "scales"

counting that makes like 7 threads

[pure karma]

yea it dosent scale everything for example trains are limited to two threads each

or one full p-core

[jagged snow]

How are you defining "scale" here? But either way, you're incorrect about the manner in which the game functions

[pure karma]

found that out but yea

[jagged snow]

Don't confuse software and hardware threads

From Ben

you have the:
Major threads:

- Gamethread
- Draw thread
- renderhardware inteface thread

then each thread has access to a pool of background and foreground threads
the number of those threads are determined based on your hardware
On my machine I believe I have 12 foreground and 12 background thread.

Gamethread with controls the:
Physics thread and animation thread do "task" calls to those foreground / background threads.
The Draw & rhi-t make use of those threads too.```

[wanton orchid]

did you heard of "the economy of scale" ?
it has the same root idea
scaling is when it is following the curve when you throw [threads here] at it
doing more or better of the same thing
talking about the game : it is NOT doing more of the game as you add threads to it
even though it uses some threads
and use more when bigger saves
you could say big save scales from 1 to 4 threads
but otherwise it's not scaling

[jagged snow]

That's not really the way I've heard scaling defined, even in a software development context

[wanton orchid]

that's exactly how the word can be used in software development context

[jagged snow]

But either way, that is exactly what's happening, more threads gives a performance boost just like faster individual threads would

[wanton orchid]

usually talking about number of running applications or server machines

[edgy hazel]

that is extremely developer and gamestyle dependant

it's a factor that's basically worth ignoring if you do general benchmarks

[wanton orchid]

they confused threads with work queues/pool here unfortunately
the foreground and background pool part is mostly deciding how the tasks are scheduled
but actual threads are showing off of the 'tasks' dependencies
also means that here it can fuck up in scheduling at 2 or 3 points :
at game level (not choosing foreground/background pool correctly making foreground dependent on background for instance)
at engine level (miss spreading the "tasks" into the pools and slowing the whole process out of cpu memory binding, dependency starving and low level cache flushing/fetching [hello x3d big boost])
at OS level still fucking that as the threads may be physically scheduled in the reverse order of the software dependencies and further impact on memory binding and cache contention

[wanton orchid]

that is extremely not dependent for gaming on a specific engine that hold all the task scheduling and general game processing architecture

[edgy hazel]

yes if you focus on a single game

and even then, devs make mistakes, make threading not efficient or assume players use a mechanic a way that isn't actually used that way

[wanton orchid]

if you push 100 dependent tasks in a pool in one go without any more informations provided to the pool manager
then it's basically a lottery scheduling wise

and it haves many way to fuck up

[jagged snow]

I think you're mixing up when people are talking about hardware vs software threads

[edgy hazel]

Yes. CS2 won't have as much use of multiple threads as CS2 would have for example

[wanton orchid]

and the scheduling logic get much slower than the actual task hoping does
and is always left behind for catching up the dependency logic

[edgy hazel]

[wanton orchid]

(once it figured out "oh that task is better spread like this" the task behavior already changed)

[edgy hazel]

(although CS2 isn't as optimised as CS2)

[maiden coyote]

Just throw more cores at it

Install more cores

[twin dew]

Doesn't help if you cannot find independent bits to throw to those cores.
Too much interdependancies and splitting the work causes it to finish slower.

[twin dew]

Which is why most games get single-core limited still.

[gilded helm]

SOmeone in the Cyberpunk Discord was saying they think UE5 is a step backwards in some regards because it's not good at multicore?

[twin dew]

Not better or worse than UE4?

Multithreading games is HARD.
As you cannot multithread steps that depend or can depend on other steps.

For Satisfactory example, you cannot multithread belt handling fully.
You need to first split all the belts into systems, that might go over the machines too, depending on exact code.
And only then you can send each such system to be handled in separate thread.
And need to start from sinks or like end-points, then work backwards up the chain to miners.

[gilded helm]

I can't find the post I'm looking for, but they're also saying it's a shame RE4 is dead, because it's the most multithreaded engine that exists.

[twin dew]

Often it isn't about the engine, but how it is used.

And the engine used for Resident Evil 4 remake in 2023 has been in use since Resident Evil 7 released in 2017, and is being used for more games still with updates.

Or meaning the engine for original Resident Evil 4 in 2005, for GameCube only?
Completely different use-case and at that point the thing was hand-optimized for the specific game on the specific console to just get it to run well enough.

And cannot be the latter, as GameCube only had single core.

[gilded helm]

Red Engine 4

[twin dew]

So engine that was never used for anything?

[gilded helm]

Cyberpunk isn't nothing 🤷

I don't mourn RE4 though, I never felt like it handled crucial player status effects and timers well.

[twin dew]

Ah, that explains why it wasn't mentioned in the article I found.

Only up to Red Engine 3

[dire igloo]

DDR4 or DDR5?
With DDR5, both are desirable, iirc M more than A.
No idea for DDR4 tho

[dire igloo]

Corsair binning on DDR4 is crap btw, from what I've heard they're the least desirable brand

[twin dew]

No, DDR5 Hynix 16Gb A is better than M in everything.

[dire igloo]

Oh alright

[twin dew]

M is the first revision for Hynix, A is second.

For 24Gb dies, Hynix M is the best there currently is.

And Hynix M used to be the best for 16Gb, but the A is just better.
Higher frequency and goes lower in tRP than M.

[dire igloo]

Yeah, had to recheck.
M was best until A was released

And 24gb doesn't have A anyway according to this spreadsheet

[twin dew]

Yup.

But there is Micron 24Gb die(s) too.

[dire igloo]

https://docs.google.com/document/u/0/d/1efiUn8LWI9sOFz8kaXHlZcZHc0Q4caFy834OQc6iFdk/mobilebasic?pli=1

[twin dew]

But was just point that when talking about die revisions, need to include the size too.

And manufacturer.

[dire igloo]

Yeah

Tho at least when referring to M die, you know it's Hynix

[twin dew]

In theory Samsung might hit M at some point.
Micron usually uses multi-character die revision names that include the die size etc.

Not sure if Samsung hit M on some DDR4 die already or not.

[twin dew]

Which was what I said, that Corsair likes to intentionally mislead.
But for some reason only G.Skill was option then?
When there are LOT of RAM stick manufacturers around...

[night girder]

Thanks for your two cents. If only I had usefull data to work with and not need to guess the whole time.

[dire igloo]

In the end, DDR4 binning is all over the place anyway
I was just shedding some light onto the nuances of RAM and that Frigidman was a bit too quick to the trigger on his judgement

[twin dew]

As in stopped providing anything but CAS on their site for example.

And do their best with DDR5 to confuse between sets using Micron chips and Hynix chips.

[dire igloo]

I've recently seen Corsair DDR5-6000 CL36 kits with 1.25V

Was super confused that they used Hynix for that

But makes sense if they want to obscure their Samsung CL36 kits

[twin dew]

Not Samsung.
Corsair is owned by Micron.

Ah, sorry, again mixing with Crucial...

But Corsair does have lot of Micron spec kits, that they try to hide with Hynix kits by not releasing full specs.

[night girder]

The thing is, it could be CPU, but I had HWinfo log before the crash. Disable buffer (so straight to disk flush) in HWinfo. Max CPU %:

Around the 1H mark you can see a big peak. But the system keeps going, then the usage even goes a bit lower and then it crashes.

[dire igloo]

Is there CL36 Micron? I thought it was almost certain to be S16B unless <1.3V then it's M16A/M16M

[night girder]

I just found something out by reading the manual. I put my SSD in the m2_3 slot.

While technically both should be fine since it's Lexar® NM790 M.2 2280.

[twin dew]

If the SSD is in the far slot, you could try what happens if you force it to PCIe 3.0 max mode, instead of PCIe 4.0, in BIOS.

No, it is pretty close to the chipset.

So not being able to actually do 4.0 is somewhat slim chance for that slot.

[wanton orchid]

is the game crashing an online game ?

could it be network driver ?

[twin dew]

Problem is that it isn't crashing to OS, or OS crashing.
But computer doing automatic reboots.

With no notification to OS.

[wanton orchid]

ok

[twin dew]

Ah, one to try would be to lower FCLK to 1700 or lower.

Because that is one thing that is ECC checked, and doesn't give errors currently, just reboots the computer.

[wanton orchid]

put the cpu in the safest place and watch I guess

I thought that was what already done

[twin dew]

RAM is on auto, and FCLK is on auto, no idea what the specific MB sets it on both auto.

Can be 1733, 1800 or 2000 for example.

HWiNFO will show what it is currently.

@night girder So what does HWiNFO say about Infinity Fabric Clock (FCLK)?
Below the CPU power numbers, above the thermal throttle stuff.

Because it was FCLK max frequency dropping that happens with the VSOC damaged IODs, until the CPU dies.

Because with the settings changes that were done, that should be the last ECC checked thing in the CPU than doesn't do error reporting with current AGESA.
Cores do, PCIe should do, you don't have ECC RAM.
IF/FCLK doesn't currently do, and uncorrected errors cause reboots.

[night girder]

Hey, sorry. Let me check 😄

1,996.4MHz

[twin dew]

So at the highest that "should work on all".
Set it to 1733 (default AFAIK), or 1800 in BIOS.

If ASRock really sets it to that on Auto with XMP disabled, they are crazy...

The CPU I had to send for warranty replacement could only do 1900 at that point anymore.

[night girder]

Symptoms at random:

• PC shuts down when alt + tabbing to Guild Wars 2 (This causes the crash most often).
• PC shuts down when I am AFK in a city in Guild Wars 2
• PC crashes during events in Guild Wars 2 with a lot of people and monsters.
• PC crashes when I run Guild Wars 2 and have steam downloading full throttle.
• PC crashes when I changed the settings in Guild Wars 2 to low. Or to high. Or to anything.
• PC doesn't crash with Read Dead Redemption 2 and Hogwarts Legacy.
• PC was able to run the game fine for 3 days stable. Until yesterday. Probable just lucky to have it 3 days stable.

[twin dew]

Yup, 1733MHz is the default FCLK for Raphaels.

So it is currently significantly overclocked.

[night girder]

You think? Not that I don't* trust you. But where can I double check?

The same with the boot code yesterday 😉 Maybe it's just Asrock being special. Or AMD.

[twin dew]

Try with that forced 1733?

You mean where to check what the default for Raphael FCLK is?

[night girder]

I don't know Raphael.

But what the stock FCLK is for 7600X 🙂

[twin dew]

Raphael is the code name for the CPU series.

The 7000 series Zen4 non-APUs for desktop.

[night girder]

So that's bullshit then? 😒

[twin dew]

hothardware.com/news/amd-addresses-zen-4-memory-oc-details
skatterbencher.com/2022/09/26/raphael-overclocking-whats-new/#AMD_Raphael_Clocking_Topology

[twin dew]

If the CPU can do it.
Those are common numbers people can push it before getting constant unstability.

2000 is what almost all CPUs can do without problems, when new.
2200 is about max anyone can do.

[night girder]

If it's my CPU, is there a stress test I can do?

[twin dew]

Not really.
But point was to set it to default and see if the same problems continue or not.

[night girder]

But you just saaid this. CPU is two months old 😒

I am just confused what the default is 😂

[twin dew]

As like I said, the AGESA doesn't currently report the ECC errors happening on that bus, and just swallows the errors.
And reboots on uncorrected one.

[night girder]

but I will try it out. Just seeing if that -10% power consumption does anything.

[twin dew]

And that was what my Asus MB set it on all default.
And upped to 1800 in specific case, and then to 2000 with another case, depending on RAM settings.

So 1733 is default, 2000 is "what (almost) every cpu should be able to do), and up to 2200 in cases.
But as there is no way to check if it functions correctly, except to get reboots on stress testing...

[night girder]

So you don't think it's the M2_3 slot?

[twin dew]

Or in some other cases.

[night girder]

32 Gb/s vs 128 GB/s. I was theorizing, that if the GPU uses same lane as that 32Gb/s that it could cause issues when there is high load on SSD + GPU all using same lanes on mobo.

[twin dew]

No, GPU is direct to CPU.
Those lanes go through the chipset.

The CPU has 16 lanes for GPU, 4 for the first M.2 slot, and 4 for the chipset.

[wanton orchid]

so it crashs on cpu load/power spikes
could it be vrm ?

[night girder]

Sorry that I bring up the m2_3, steam really threw me off. 3 days fine. Install steam, download game and crashes. So I trying to put 1 + 1 together...

[twin dew]

We did try modest upping of VCore (+10 curve optimizer), to give the VCore more headroom.

[twin dew]

And it CAN be that the specific SSD doesn't handle being on just 2x slot properly, or that the slot cannot do PCIe 4.0 with that SSD.

[night girder]

But the symptoms are so random and hard to replicate... head scratch

[night girder]

I think I can check that in device manager maybe... let's see

[wanton orchid]

I'm still heading toward power delivery chain problem

[twin dew]

But the FCLK instability fits the symptoms better, as that always causes just random reboots.

[night girder]

pigtails?

[twin dew]

When the others would cause calculation errors etc. too.
Or PCIe should give errors into Event Viewer now.

[wanton orchid]

steam dowload does not spike gpu

[wanton orchid]

👍

flck should NOT be at almost 2000 on an unstable pc

[twin dew]

And is pretty easy to test for, by lowering it to default for now.

[night girder]

Yes I know Baldur, but if I do the flck thing. We wont know if it was -10% GPU power tuning or FLCK.

[wanton orchid]

baldur : should the motherboard match the ram automatically as it is on auto, if flck is manual ?

[night girder]

I try to do fix -> crash -> next fix. Because if I do fix -> fix -> fix -> crash. I don't know what fixed it 😂

[twin dew]

In theory MB should just always use 1733MHz if they followed AMD spec.
But they do OC the IF based on whatever the specific manufacturer decided.

[night girder]

Let me flck it.

brb. Or not if my PC explodes 😂

[twin dew]

For Asus it follows somewhat the RAM frequency up to 6000MT/s AFAIK.
But ASRock seems to just set it to 2000?

[wanton orchid]

I mean, if we set the flck manually, does the ram [auto] set ram to matching frequency ?

[twin dew]

No.

[wanton orchid]

ok so probably should check the ram manually too ?

[twin dew]

It is on auto right now and 4800MT/s.

[night girder]

This is stock settings.

I mean nothing is OC in bios afaik.

[twin dew]

So it was forced to that...
Why, what is going on?
And why is VSOC 1.2V?
And is VDDP the one below that and set to 1.100V?

[night girder]

How do I know?

[twin dew]

The whole pic is nothing but OC settings below the RAM speed, set to OC values...

[night girder]

I cant reset it. Thats the factory settings.

[twin dew]

Default for the three topmost voltages would be 1.100V, next is probably 1.8V default, the VDD_SOC is 1.00V default, VDDP would be 0.80V default.

[night girder]

I did load UEFI BIOS defaults multiple times.

[wanton orchid]

soc 1.2 is high

too high

should work fine at 1V iirc

[night girder]

Is there a hard reset button for BIOS or something?

[twin dew]

For that RAM speed, yes.
Up to 5200MT/s is officially supported at 1.0V SOC with single DIMM per channel.

So ASRock is doing idiotic things this generation...

[night girder]

Or did I get shipped an Overclocked motherboard?

[twin dew]

The Clear CMOS is the most extreme version of the setting to defaults.

[wanton orchid]

tbh everytime I receive a new motherboard I always fully reset the bios, but updating + cmos + no-batterying it

for both consistency and security purpose

you never know who touched the motherboard before arrival

[night girder]

I did cmos yesterday

[twin dew]

But it would seem that ASRock sets OC stuff on defaults...
Which has unfortunately been somewhat common over the years.

But take another pic where the left edge is shown too.

[twin dew]

But set what you can of those to auto.
Usually select field, press a, exit, or select field, press a, press -, exit.

Ok, the 1.1V was VMisc, that is 1.10V on default.

And the 1.8 was the VPP, which is that as stock.
But the rest are not stock...

[night girder]

So UCLK also to auto?

[twin dew]

And that default 2000 FCLK is just weird...

[twin dew]

Auto should mean that setting up to 6000MT/s and 1:2 after.

Trying to be helpful to people who just set XMP/Expo On with over 6000MT/s RAM set.

[twin dew]

And try to set the various voltages to auto too.

[night girder]

VVD msc and VDD soc dont have auto

[twin dew]

What happens if you enter a and then press -?

[night girder]

Resets to numeric value

Oh wait.

Fixed. i think…

[twin dew]

Usually the Auto is before the first supported numeric value.

On that type of field, just tried one of the ways you can "shortcut" to it on some MBs.

[twin dew]

Better.
The RAM is 0.15V above "stock" but below what XMP/EXPO would probably set, so we can let it go for now.

[night girder]

Ram I didnt do anything it on stock

JEDECc

[twin dew]

Point is that the defaults on ASRock overvolt the RAM compared to JEDEC spec, which would be 1.10V for the top three. and 1.8V for the bottom fourth.

Ah, now you are setting the RAM to 5600 EXPO profile.

[night girder]

Nono!

Wanted to show the table.

oh wait!

[twin dew]

Yeah, but the top selection on that screen was at the EXPO 5600 set.
Which explains the RAM etc. stuff.

[night girder]

Wtf, why was it on 5

[twin dew]

You need to set that box to something else.

[night girder]

I set it to auto :p

Ok. Seems good?

[twin dew]

So save and go check what happened, and remove the GPU downpower for now?

[night girder]

Ok thanks 🙏

[twin dew]

Mainly what the VSOC, FCLK etc. are now.

But if you didn't update your BIOS before installing the CPU, it might have been hit by the 1.35V VSOC feeding.
If the MB back then autoapplied some RAM stick settings too without manually setting XMP/EXPO on.
Like it was right now doing.

And that causes IOD to begin degrading over time, with stable FCLK capacity getting lower with time as the symptom for me.

Could do 2100MHz or more as new for the first 7800X3D, could only do 1900MHz when I sent it for replacement.

And that thing ran at 1.35VSoc for only about 15 minutes.

And at 1.05V after that.

[night girder]

Let's see how it goes.

[edgy hazel]

mathematical notation explained
symbol meaning
= equals
≠ not equals
< left
> right
! LOUD NUMBER
~ worm
π stonehenge
√ right answer
x wrong answer
⋯ soon…
∮ what Exacrly the fuck
∝ fish
∞ fish with 2 heads
↯ lightning
:⇔ he Scream

[twin dew]

"!=", not "=!=" or "=/=" ! XD

Rest are more near to correct.

[edgy hazel]

sry let me fix that

[dire igloo]

≠ > !=

[twin dew]

Separate character, vs. weird uncommon bastardization when != has been the "standard" replacement for very long time on computers.

[dire igloo]

≠ is nicer to read on paper tho

[twin dew]

Point was that the message originally read =/= as three characters.

[night girder]

I got used to the != because programming.

[dire igloo]

Which is eeeeeh

[twin dew]

And that was what I was talking about.

[edgy hazel]

mathematicians don't know how to use a computer tho

[night girder]

they do

[dire igloo]

Figured it out from context

[edgy hazel]

isn't ur shit crashing the whole time?

[night girder]

https://tenor.com/view/cry-me-a-river-justin-timberlake-singing-music-video-gif-16648841

[night girder]

Yeah it is. Since I listened to you fools to switch to AMD 😉 - ||This is a joke. ||

[edgy hazel]

nah it's cuz u do math

[night girder]

you mean meth*

[edgy hazel]

don't know to to use a PC

[twin dew]

If this ends up being unstable FCLK, I did tell you to reflash the latest BIOS onto the MB before installing the CPU, which you didn't do IIRC.

[edgy hazel]

https://tenor.com/view/hue-huehue-huehuehue-duck-laugh-gif-4916860

[night girder]

Sorry, if you did say it and I ignored it. Not by intention. So many people were helping me out, must slipped through the messages.

And no, I didn't do that. Plugged in everything. Had BIOS update ready on stick. So I updated BIOS before I installed the OS back then.

[twin dew]

Shouldn't have mattered, if ASRock wasn't doing insane things.

But seem they might be.

[night girder]

Can still be something else.

[twin dew]

Yup.

[night girder]

I think if it runs stable for a month, we can be sure 😉 (But I had the false hope that 2-3 days of stable = we fixed it. But nop).

[pure karma]

seeing how i need to order some DP cables should i also throw in a rgb hub while im at it i dont need one but im up to like 6 or 7 rgb conectors and there just really inconvinent to try and just shove somewhere

the only downside is il have 3 difrent hubs

and i can allready barely fit 2

[night girder]

What do the other two hubs do?

[charred pewter]

i kept reading that Type-A were the bom, what you want... and type-m are just fine, but not great if you try to do any overclocking (ddr5)

[night girder]

Frigid, what's up.

[twin dew]

Difference is in the max frequency at same voltage.
Hynix M gets you to about 7000 or so, A gets you to 8000 or so with extreme voltages, if the CPU IMC can do it.
And A can do lower tRP than M.

M can do about 4800-5000 on 1.1V, A can do 5600 or so.

And those extreme frequencies are at 1.45+V RAM voltage.

[charred pewter]

tRP ?

[twin dew]

One of the primary timings.

[charred pewter]

is that the 4th one....

[twin dew]

third.

[charred pewter]

ah... ok, the 4th one on these is horrible ... its 102: 32-39-39-102

[charred pewter]

fix your crash?

[night girder]

Oh yeah Baldur, when we were resetting the mobo, I saw tghat tPRwas from 36-> auto. That's because we reseted the OC profile of the ram (exo -> auto)?

[twin dew]

Yes.

Which I had thought had been done long time ago.

[night girder]

We are onto something atm. Mostly Baldur. A.k.a. get my motherboard to auto as much as possible

[twin dew]

I had understood that both RAM, FCLK etc. were on defaults (Auto), but it was still on XMP except for the RAM frequency.

Including that FCLK, which would match the symptoms if unstable.

[night girder]

I had no clue. I am not a bios man 😦

[charred pewter]

oh so you've had bios tweaks all this time?

[night girder]

We think so yes.

Although I reseted it multiple times to factory presets. And did a CMOS clear multiple times yesterday.

[charred pewter]

that kicked me a few years ago when i was trying to run ram at 'xmp' and kept getting random windows crashes ... lol, a total bios wipe/reset got it all happy

[night girder]

and I know the CMOS clear worked, I had to reenter my passwords

[twin dew]

ASRock might have been being "helpful" and enabling some of XMP/EXPO profile by default.
Infinity Fabric was as 2000MHz, when stock is 1733MHz.

2000MHz is "what every CPU should unofficially run at" number.

[night girder]

oh let me check what it is on now.

[twin dew]

Yes, I did ask you to check the FCLK and VSOC etc. in HWiNFO now.

[night girder]

Shit, so the auto still puts it at that 😒

[twin dew]

"Nice", so go back and set it manually to 1733.

Check the VSOC too before

CPU VDDCR_SOC Voltage, below the core temps.

And the VDD (SWA), VDDQ (SWB) in RAM stick section.

[twin dew]

Might have that the Auto on next reboot enabled that EXPO 2 again.

[charred pewter]

bios trying to be helpful ... 🤔

i like the bios on asus and msi ... they are pretty much 99% unhelpful 😆

[twin dew]

So those are at stock.

[night girder]

FCLK is good now 👍

[night girder]

I like Asrock bios too. I mean, I am not going to point any fingers at this momemt. No clue wtf is going on.

At this point, I would just be happy if I could pinpoint the faulty hardware. And just replace it. And then I can say "what a shit product"

[twin dew]

And it might not even be faulty at this time, just doesn't go to "normal" OC.
As the MB setting IF to 2000MHz on FCLK: Auto, and RAM: Auto isn't "normal".
Trying to be helpful and give people better performance, but it is still OC.

[night girder]

Stock MB = 2000MHz FCLK. MB -> FLCK to Auto = 2000MHz

[twin dew]

Asus does the same once you increase the RAM clock.
But not before.

[night girder]

I didn't even know that a PSU can struggle with certain types of GPU.

[twin dew]

Because the Nvidia cards do huge spikes way above their TDP.

Nvidia only manages average draw, not instantaneous one.
And the PSUs had quick overcurrent protections, that shut the PSU down when the current hit 110% nominal momentarely.

[wide void]

the 'average' is over a few seconds, the spikes can be double the average

[twin dew]

Yup.
300W GPU taking 600W part of the time.

[night girder]

I see, well it's fixed in the new models like GM/GX/PX. I have a GX so I should be good.

[wide void]

shit design really

[twin dew]

Which is why the ATX 3.0 spec now has spec for that, and allows up to double for 100ms or similar time.

Which is just crazy IMHO.

[wide void]

a little capacitor space should/would have smoothed out the demand I'm thinking

[twin dew]

As in 1000W PSU must be able and allow for 2000W load for short enough time.

[twin dew]

Or just using the VRM features, instead of rolling their own...

[twin dew]

For ATX 3.0 PSUs.

[night girder]

But like 750W is enough right? For 7600X and 6950 XT. I always wonder how to know if you have plenty or not.

[wide void]

750 isn't a lot these days

[twin dew]

Should be.
And PSU protections should shut the PSU down, not reboot.

With audible click when relay opens.

[night girder]

Exactly. And if it was PSU, my RGB should cut out in the case. Which never happens. All lights stay on, just monitor black and reboot.

[twin dew]

But IF uncorrected errors cause reboot like if you pressed the reset button on the case.

With no warning to anything.

IF = Infinity Fabric, the bus between the IO-die and the Core-die, and inside the IO-die between different sections to some extent.
What the FCLK (Fabric Clock) sets speed for.

[night girder]

Toggled off automatic restart today. 0.001% chance I know, but low hanging fruit so I did it.

[twin dew]

That only is for BSOD.
You would still get the BSOD screen (or blank screen for time), and then reboot.
And there would be Event Log entry about BSOD happening.

Just means if the Windows sits at the BSOD screen until end of time, or automatically reboots once the data about the BSOD is written to disk.

[night girder]

Ah, read it somewhere that it could happen a PC reboots so fast, it doesn't show blue screen.

[twin dew]

If the first tick and the memory dump write were off.

[night girder]

Didn't know that BSOD are also logged to eventviewer.

[twin dew]

That is what that uppermost tick sets on or off.

[night girder]

I also checked, no mini dumps to be found (ofc).

I wonder if I still get those weird errors about perfdiag running.

Do you also get these annoying: Metadata staging failed, result=0x80070490 for container '{00000000-0000-0000-FFFF-FFFFFFFFFFFF}' ?

On my previous install (before I did clean install) I had device hardware metadata turned off. So I never got those. Now I left it turned on, so avoid issues, and I get those errors. I think it's the server being down or something.

So the perfdiag still does weird stuff too haha 😄

[twin dew]

Tries to find a WU supplied driver for device with GUID of 00000000-0000-0000-FFFF-FFFFFFFFFFFF, which seems to be something that isn't actually valid.

Ok, seems I do have device container with that GUID, the "Microsoft ACPI-Compliant System" one.

[night girder]

How did you find that, I am searching with powershell and now in the registry.

[twin dew]

Put search on the registry for that "00000000-0000-0000-FFFF-FFFFFFFFFFFF"

Found this as the first one, and then looked what was under it.
Computer\HKEY_LOCAL_MACHINE\SYSTEM\ControlSet001\Control\DeviceContainers{00000000-0000-0000-FFFF-FFFFFFFFFFFF}

[night girder]

"USB\VID_05E3&PID_0610\8&f2cb6fa&0&13" this one times out too for me. 🤷‍♂️

But I did google a bit around it, and people said it's mostly server issues 😛 Somebody used wireshark to monitor the metadata exchange 😂

[twin dew]

But there has been long time bug in Win10 times where you need to change the server address in registry, do a run to get the metadata, and set it back.

As MS is by default using redirect step in the process.

[night girder]

Interesting.

[twin dew]

Help-threads from 2015 about how to fix the issue with that.

[night girder]

found same thread 😉

[twin dew]

Multiple ones from that timeframe, and some later ones.

[night girder]

www.tenforums.com/drivers-hardware/13474-metadata-staging-failed.html

[twin dew]

Different from the ones I actually opened.

[night girder]

https://www.youtube.com/watch?v=LCaBm2uuCjw

Even a YT video, hehe

[twin dew]

Computer\HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows\CurrentVersion\Device Metadata
DeviceMetadataServiceURL
Need to temporarely be changed to http:// dmd.metaservices.microsoft.com/dms/metadata.svc
Without the space

Or that was the one, seems that is wrong now.

Moment.

[night girder]

Wtf, what a crappy video.

[twin dew]

Ok, it really is down right now.
The redirect URL works, but the redirect leads to that which gives almost instant timeout.

And now 502, Bad Gateway Azure error.

[night girder]

not sure if the url's accessible without sending the proper headers etc.

[twin dew]

Usually gives something, not the actual data, but something.

[night girder]

Oh wait. I might have something, not 100% sure yet.

[twin dew]

But as the behavior is same as when the QA site goes down for azure maintenance, might just be that right now.

[night girder]

Not sure if you remember that bug with the HP printers? But I assume Microsoft just disabled the server untill they have a fix

[twin dew]

Possibly.

[night girder]

"disabled" = just no access untill they know what the hell is going wrong.

[twin dew]

Well, handles big part of the WU driver support too if I understood correctly.

[night girder]

the WU drivers come with the updates. A few drivers are installed with WU, on my machine atleast.

[twin dew]

Matching the HW to WU packages.

And initiating the download for the immediate ones when there is device missing drivers completely.

[night girder]

👆 This one get's installed by a windows update. Based on my headset. When I disable "drivers with WU" in GP, I didn't get it last time.

So I am pretty sure, that according to Microsoft standards, this is a driver they ship with their updates, since I can disable it with GP.