overclocking

[kind walrus]

Use bios

[misty socket]

Whats the safest value to start off on?

[kind walrus]

That depends on your cpu and your testing

There's no one size fits all answer unfortunately

I will say this, use a separate windows install to play around with stuff like that

Since undervolting, cpu oc, ram oc, all carry the possibility of ruining windows

[misty socket]

good to know

ty!

do you know which ones you alter for that on MSI bios sorry first time ever doing undervolting/overclocking at all

I'm just worried about how hott my cpu runs

[kind walrus]

It's lga1700

And a K chip at that

It's gonna try and run at 100C almost no matter what

It's just how it is

Undervolting will do a little bit but ultimately unless you do stuff like ICCmax or power limiting you won't really get anywhere with temps

(and yes, this means you have to accept a performance loss in some capacity)

And to make clear as well btw @misty socket , these cpus could happily plug away at 100C 24/7 for months, years and likely still be exactly the same as they started

100C is a safe limit, it's going beyond 100C that you start to creep into concerning territory

CPU's are designed to self regulate at their throttle so they stay well away from damaging temps

If you want proper advice, avoid doing lots of excessive multicore tests beyond stock limits on 14th gen

Stuff like cinebench over and over and over have been known to degrade some of the late intel chips (Source: oc.net)

So try to avoid shoving like, over 250W into your i7

[bleak sable]

I totally forgot i used this photo as a HWBOT submission and got a bronze meal lmao

[sudden torrent]

I've totally done that with that exact fan

I still need to sub the scores but it'll be all golds once I'm sure I won't have the subs removed for using the wrong driver version (the latest version is from 2011 and doesn't work right when overclocking)

[alpine mantle]

[deft quiver]

L volt mod

[wheat lintel]

Average Marsy moment

[proven canopy]

What is that

[tall pelican]

soldering the evc wires directly to the through hole header points

instead of using a pin header that the board was designed for

or just super jank soldering?

[steady lance]

Does anyone know what these voltages affect? (X670-P mobo)

• DRAM VDDQ
• DRAM VDDP
• CPU VDDIO

[steady lance]

I can't for the life of me find any good information regarding what CPU VDDIO voltage affects, it's limits, and etc.

Can anyone provide some insight on the info I gathered?
From what I've read, VDDQ voltage does not need to match VDD and can be lower usually by 0.5v reducing heat and also that multiple A die users seem to be dailying 1.6-1.9v for about a year now without degradation?

Oh wait, is CPU VDDIO the memory controller voltage?

[steady lance]

Got lucky and remembered I had posted my stable OC timings here. Will tune the voltage needed later.

[steady lance]

Hmmm weird, my tReFI doesn't work anymore it seems.

Dropped it to 32k will mess with that later

[steady lance]

I am running TM5 and I notice this show up even when I run it with Administrator. Any idea why this is.

[kind walrus]

If it's still running the test I believe it's fine

[steady lance]

it was I think considering my pc slowed down a lot.

[kind walrus]

It'll say whether it loaded the profile though is what I mean

[steady lance]

profile is loaded

PCBdestroyer

also the tune was more effective then I thought. getting about 20-35+ fps on average

[steady lance]

Somehow my former ram OC isn't stable on this new windows install.

[wheat lintel]

Not surprised

[zenith palm]

Probably wasn't stable in the first place

Just appeared stable

Ptsd to my samsung c die oc

[steady lance]

Never crashed until this new win install though.

[zenith palm]

My oc i thought was stable for 6 months then one day the windows boot manager boom gone

[steady lance]

ouch

[wheat lintel]

It killed your windows OS, it never was stable

Doesn’t have to crash to not be stable

You barely tested it

[steady lance]

that was a different issue.

Found out it was something to do with the Typical Idle Current.

[steady lance]

Not worth the extra hassle so I am probably just gonna stick to XMP for ram till I upgrade.

[dim plank]

My iGPU is currently 100% but the VRAM usage is not fully using the 2GB shared memory. Does this mean the GPU speed is so slow and the VRAM is not the bottleneck? Im not very familiar with these stuff. https://ibb.co/Dg5qP3b

[barren ridge]

You're gpu limited

[dim plank]

So it means the VRAM doesnt have any problems yet. Only the GPU compute speed is overloaded?

[sudden torrent]

Your VRAM is full, the first number is the total usage, the second is the usage of the program using the most memory

So it's both the GPU core and the VRAM that's fully loaded

[barren ridge]

^

[eternal gust]

I'm very new to PC technicals and have heard the term overclocking thrown around quite a lot. I understand that you're trying to get the most out of your CPU or RAM (and I imagine its possible with your GPU too), but what are the risks that I keep hearing? Can you fry your whole PC if you get it wrong, or would it be limited to the part you were overclocking? Is it something that a beginner should approach at all, even cautiously, or is it strictly something that's recommended for people with much higher skill? Do you guys have any resources if I do want to learn more about overclocking, even in theory?

[dull ginkgo]

Overclocking nowadays is not necessary and not significant enough on your CPU and on your GPU for dailying. The main gains from ocing today comes from ram. Main concerns with overclocking are instability and deterioration, sometimes windows becomes borked, sometimes pushing too much voltage makes the silicon deteriorate. Damage to the functionality of the components is possible but more likely due to heat than what you set. There are some resources pinned if you want to look around.

[sudden torrent]

RAM overclocking can get you more stable frame rates, that's the most effective overclock for gamers

[fervent talon]

I never overclocked. Was too scared

[zenith palm]

most effective highest pain in the ass

[steady lance]

agreed

[alpine mantle]

You said that but many people lose their minds, me included

[faint tangle]

Ddr5 it's much easier to see some pretty major gains

[sudden torrent]

Mostly because there's only 4 memory dies to worry about and most sticks overclock the same

So you can just copy/paste other people's settings

[faint tangle]

Yeah its so much simpler and a lot more beneficial

[late kraken]

I'm very new to PC technicals and have heard the term overclocking thrown around quite a lot. I understand that you're trying to get the most out of your CPU or RAM (and I imagine its possible with your GPU too), but what are the risks that I keep hearing? Can you fry your whole PC if you get it wrong, or would it be limited to the part you were overclocking? Is it something that a beginner should approach at all, even cautiously, or is it strictly something that's recommended for people with much higher skill? Do you guys have any resources if I do want to learn more about overclocking, even in theory?

[kind walrus]

Overclocking can be anywhere on the gamut from very easy, to as little as one button, to all the way on the other end of the spectrum where it can be completely risky, outright will destroy hardware and/or carry actual risks depending how unprepared a person is

For people who first dive into this however, provided you avoid the physical side of overclocking

And avoid some obvious things like say, running 2 volt into your cpu core

It's fairly safe

The most you risk with some proper care is a lost windows install and time

[late kraken]

Thank you for so much information 🙂

[kind walrus]

It's very user-specific as everyone's pc is different

But the gist of it is there's always a degree of extra performance baked into your pc parts (there for the sake of margin of error, safety, and stability) that you can leverage to make the pc faster

[late kraken]

Oooo

[kind walrus]

To say free performance*, always has the * as a disclaimer

As you push the limits of your stability to eek more out

[sudden torrent]

On that note, I noticed after a bios update that for some reason my board wants to put 1.5v+ through the CPU. It was more like 1.325v before under load.

[kind walrus]

1.5 under load??

On 5800x??

[sudden torrent]

Yes

[kind walrus]

Sus af

[sudden torrent]

LLC 3 so very little vdroop

[kind walrus]

You don't have two pieces of monitor software running right

[late kraken]

It shouldn't be right

[kind walrus]

If you run hwinfo64 with say aida, it can cause misreading

[sudden torrent]

Nope just hwinfo

[kind walrus]

Roll back bios

[sudden torrent]

Bios said it too

[kind walrus]

Roll it back

That's sus imo

[late kraken]

Yes

You're right @kind walrus

Roll back bios @sudden torrent

[kind walrus]

Ryzen boost and pbo are safe, but...only as safe as the agesa codes it to be

[sudden torrent]

I just set a manual voltage to compensate for now because that bios update addresses the critical security issues they recently revealed

[kind walrus]

You can do offset no

Offset it back to old voltages

That lets you keep pbo

[sudden torrent]

Offset only goes down to -100mV

[kind walrus]

Oof

[sudden torrent]

But I can CO down more and see what it does

[kind walrus]

I'd just rollback for now

[late kraken]

Yes @sudden torrent
You should rollback bios

[sudden torrent]

[kind walrus]

That's better

If it's within a few mV imo that's fine

[sudden torrent]

Peaked at 1.55v tho

This is idle

[kind walrus]

1.5+ idle is fine

It's the combination of voltage and current

That's what kills it

[sudden torrent]

No, the 1.35v is idle, the 1.55v was loaded

[kind walrus]

EH

[late kraken]

@kind walrus
Can heating environment increase Pc Performance and give better overclocking performance?

[sudden torrent]

That's what I'm saying lol it makes no sense

[kind walrus]

Cooling it does

[kind walrus]

That sounds like a glitch or horrible mistake

[late kraken]

Oh thank you

[kind walrus]

Just roll back and update later

[late kraken]

Else I have would burned up my PC 😂

[kind walrus]

Nah

Unless your pc is major old then it will manage it's temperature

It won't overheat unless it's truly struggling

[late kraken]

It would emergency shutdown itself

[kind walrus]

Yes

Which is still safe

[late kraken]

Maybe Cooling as your opinion

[sudden torrent]

It might be a LLC glitch now that I think about it, I'll play around with things. I know LLC 1 is negative droop (increased V under load) on this board.

[kind walrus]

?

[late kraken]

At what temp of cooling CPU should have a gaming PC?

[kind walrus]

Under throttling

Anything not throttling is fine

[late kraken]

I mean that at what temperature is fine for a cooling CPU

[sudden torrent]

Ideally the CPU would be under 80C but it's designed to run at the limit 24/7/365

[kind walrus]

The answer varies depending on the cpu

So the perfect answer is under tjmax

i.e not throttling

[late kraken]

Okay thank you so much for so much information

You helped me a lot

[kind walrus]

Np

[late kraken]

What's the difference between benchmarking and overclocking?

[zenith palm]

Benchmarking is basically like taking an exam, it's testing what score will be gotten in whatever exam.
Overclocking is making your parts run faster by increasing the frequency/decreasing latencies (delays)

Your clocking it over what the manufacturer set it at

[steady lance]

and just to be clear, TJMAX refers to at what temperture the CPU will downclock itself to reduce temps.

[tall pelican]

might want to look at svi2 tfn in hwinfo, its the on die voltage reading

[sudden torrent]

Playing Starfield, total CPU usage is at 75% in New Atlantis and SVI2 is 1.41-1.45v

And that's after a -100mV offset

[tall pelican]

iirc voltage offset doesnt do anything, because boost algorithm will just downlock it to match vid table

so I guess lower voltage+clocks with the offset

[sudden torrent]

Update: My CPU died

[steady lance]

f.

[deft quiver]

It's ok

I killed my 7900xtx

• apex encore

[late kraken]

Everything's Fine?
Hehehehehe

[kind walrus]

💀💀💀

[mental turret]

https://tenor.com/view/greetings-hello-johan-liebert-monster-gif-26082272

[steady lance]

AMD seems to be slipping a lot lately.

Would be kinda concerning if this was somehow intentional to encourage people to upgrade.

[sudden torrent]

AMD just denied my warranty too because it expired 2 weeks ago

[steady lance]

Well that's bs.

[kind walrus]

I'd make the argument they've been a lot better lately tbh lol

My AM5 experience has been 10x better than AM4

[sudden torrent]

MSI did too but that's more fair. They already replaced the board once and I'm almost a year past warranty on the board.

[kind walrus]

Emphasis on my exaggerated 10x

[kind walrus]

That sucks

[steady lance]

I gotta go work now so I will be back later.

Honestly if AMD gets me a new CPU that actual works I might stay on the AM5 platform otherwise I am likely going back to Intel since I never had issues with them.

[sudden torrent]

I wanted to hold off upgrading until 9000 came out but I might not have a choice now

[kind walrus]

7800x3d kinda nice

[sudden torrent]

Yeah but it's also $400

[kind walrus]

Tbf I warned you an ungodly amount of times that a 7950x3d was not an easy chip to work with

[steady lance]

The bloody thing can't even stay on 24/7 on stock settings.

[kind walrus]

Though not this level of hard, I think you drew the bad lottery there for luck

[steady lance]

Anyways have a good one folks and Fal, hopefully things work themselves out for you.

[steady lance]

My AM4 experience was even worse with USB issues.

[kind walrus]

Yes

That's amd

When it works it's great

But when you have problems it can be really painful

It took me years to get to the point where I am 1000% stable on AM4

[sudden torrent]

The USB thing was a chipset flaw and mostly the motherboard's fault

[kind walrus]

Yes it was

Very annoying though

Totally empathise with the feelings about it

[sudden torrent]

I had the drops too for a while, luckily agesa managed to fix it

[kind walrus]

It was fixed on both of my boards

But I put up with it for like a year or two before going to LGA1700

And now I want off LGA1700

Hot af

AM5 is converting me

[sudden torrent]

Maybe I'll just get a used 5800X3D or something, I need at least 8 cores for things. Problem is, besides gaming I do other things and those things prefer raw clocks.

[kind walrus]

7700?

If you're mainly doing clock bound tasks then yeah x3d won't be that significant

[sudden torrent]

Maybe? Or 7700X for the hell of it since I've got the cooling

[kind walrus]

Ye

Or maybe even 7900/x

Kekboom

[sudden torrent]

No way, not ever. Not doing dual CCD, you couldn't pay me enough.

That extra latency would be more damaging than lower clocks for my purposes

Maybe I'll get a 13600k out of spite for AMD

At least then I can keep my memory

[kind walrus]

Dual ccd isn't that bad

Lolol

[steady lance]

https://tenor.com/view/destroy-computer-gif-19603724

[steady lance]

this is how I felt about this 7950x3D at the worst of it (every hr or so crashing)

at least after multiple windows installs, bios updates, driver updates, etc it can stay stable for about two days at a time as it stands.

[steady lance]

yes, yes. Still you'd think a big company like AMD would make sure their chips are better binned or whatever the name of the process is for testing stabillity.

[sudden torrent]

They don't even power on the chip to test it iirc, they only check if it's electrically sound. At least that was what Intel did during a GN tour.

[steady lance]

You must be joking right?

Seems like they should be testing the chips more or does someone else test the chips before they get into AMD/Intel's hands?

[dull ginkgo]

That's pretty standard afaik, it's just comprehensive testing for the electrical properties of it

They won't be able to slot everything into a mobo before shipping

[deft quiver]

Bruh there isn't even that big of a difference

Latencuck

[sudden torrent]

I'm not talking about gaming, I'm talking about actual work where a half second stutter can ruin the entire batch

[steady lance]

If you don't mind me asking, what kind of work are you referring to?

[sudden torrent]

I don't want to get too specific since it'll greatly narrow down what company I work for, but I can say it's a variation of protein folding

[steady lance]

Ah alright.

Just looked it up and Protein Folding seems like a very interested subject.

[deft quiver]

Oh that yea monolithic die all the way

[alpine mantle]

Spent the whole day trying to make my 13600k reach 5.5ghz on all pcores and it was a success, had to tweak the cpu voltage, lite load, and loadline calibration on my msi z690 pro-a ddr4 motherboard. Cinebench R24 was a pain as it was prone to crashing far easier, finally managed to make it run for a while. Here's my score for R23.

OCCT Linpack - 1 Hour (Success)
Cinebench R24 - 30 Minutes (Success)

CPU
P-Core: 5.5GHZ
E-Core: 4.3GHZ
Ring: 4.8GHZ
Voltage: 1.170v @ -0.150v offset (Override + Adaptive)
Lite Load: Mode 11
Loadline: Mode 2

Max Temp on linpack: 96c
CPU Cooler: Thermalright phantom spirit 120se

sipped quite a bit of power today i cant lie

but GG for me

[short blade]

meanwhile my 13700k needs 1.28v load vcore to do 5.4/4.8/4.3

[alpine mantle]

silicon lottery is tough...

i just had to put more effort into this because i knew that this cpu still has some performance left on the table somehow

i was thinking about giving up midway too

[deft quiver]

Kinda low score

I did 5.7 5.6 5 easy

[alpine mantle]

1.53v needed...

[deft quiver]

Set voltage

[alpine mantle]

You running it daily or just for points?

[deft quiver]

It was more like 1.37

[deft quiver]

Daily

[alpine mantle]

Huh wow nice

[deft quiver]

No degradation after a year

[alpine mantle]

Honestly I just find mine to just be satisfactory, nice low voltage and not too much power for 5.5ghz oc

Yours is an insane bin

[deft quiver]

Nope

Average tbh

Just good cooling

[alpine mantle]

I'm running mine on an aircooler

You can only do so much

[deft quiver]

Fair

[alpine mantle]

Ye

[sudden torrent]

Yoooo Furmark 2 just dropped https://www.geeks3d.com/dl/show/705/UEEYzxyKzVp6ayq

[alpine mantle]

if furmark 2 dropped that means pc 2 can

[fervent talon]

No way! PC 2? The era of pc building is over!!!111!1!1one
We can get into pc 2 building

[alpine mantle]

✅

[dim plank]

In amd adrenaline you can lower down the stock minimum frequency of the GPU to -28%. Is there a downside to this? Because currently I dont see one.

[sudden torrent]

Yes? Reduced performance overall

[dim plank]

What if the max frequency is overclocked but the minimum frequency is underclocked? Will the performance still be slower because the minimum frequency will make it longer for the clock speed to reach max clock speed?

[sudden torrent]

You can get very strange behavior in low load situations, like games that don't fully load the GPU or playing youtube videos. The frequency would jump drastically in those lower load situations. What's happening is the application doesn't load the GPU fully, so the GPU goes into the lowest power state, but then it's too low, so it goes to the max power state for a second, but then it's not loaded so it goes to minimum... and so on in a loop.

You could set it down a little bit without issue probably, but test it while watching the frequency

[dim plank]

Thanks for the info.

[wind prairie]

Bruh

[alpine mantle]

😂

[kind walrus]

nah

@sudden torrent

@barren ridge

[barren ridge]

a sec

[steel zephyr]

A little liquid helium helps with clock speeds on the 14900KS. https://www.youtube.com/watch?v=qr26jxPIDm0

[steady lance]

If 9.117.75GHz were stable, what do you think the increase in gaming performance would be?

[sudden torrent]

You would need someone to manage the temps for you while you played, and it's unlikely you'd get much if any increase when gaming due to the GPU being the limiting factor in most games

And if not the GPU, then the game engine, and if not the game engine then the cache (you saw that a lot with FX-8350 processors that were totally stable at 8GHz but the scaling significantly decreased after about 5.5 - 6)

[tall pelican]

its also an incredibly light load for cpu validation, you would not be able to game at 9ghz (even with lHe)

[steel zephyr]

Someone else had it benchmark stable at 8.4-8.5 GHz for some benchmarks.

[steel zephyr]

You would probably still be GPU limited at 1080p and only have a small increase. The performance increase would be small even with a 4090 or 7900 XTX.

[tall pelican]

my 13900ks was regularly cpu bound with my 4090 at 1440p, even with 8400mts ram

would be a safe assumption that at 1080p, you'd see a very large increase with a 30-40% clock increase

[kind walrus]

Fitz with the real takes

💯

[sudden torrent]

30-40% yeah as long as the cache can keep up

[steady lance]

20,000 fps in Minecraft xD

[steel zephyr]

If you look at processor benchmarks and game FPS with high end GPUs, in most games its not a linear relationship. As you improve CPU processing speed/ability the improvements don't increase rapidly.

[alpine mantle]

Rip 5.5ghz oc I thought it was stable enough

I knew something was off with the freezing n stuff. Thought it was my GPU hoping that it wasn't my CPU dang it

[mental turret]

moar voltz

[alpine mantle]

one of the cores about to thermal throttle

i can only go up to 5.4ghz

ill just keep it that way

[mental turret]

moar cooling

[alpine mantle]

lol

[steady lance]

Just daily LN2 24/7/365 😏

Damn, I can't find a single LN2 meme gif.

[alpine mantle]

[mental turret]

https://tenor.com/view/raiden-metal-gear-rising-jack-the-ripper-balls-chemical-gif-25229269

[steady lance]

Ehh

[deft quiver]

Bruh

Is the 4090 that fast

[tall pelican]

yes

especially when shunt modded and left to do its own thing

in ow2 for instance, high (not ultra) was ~550-580 fps with the ks, and now just sits at the 600 fps cap with 7800x3d

gpu usually is at like 80% utilization

if I go ultra, it drops to the 500s and 100%

[deft quiver]

Bruh

So 4k

Is kinda needed

[tall pelican]

imagine buying a 4090 to not let it do its thing

[deft quiver]

Why shunt mod when you can flash 1000w bios

[tall pelican]

or 1440p buttery smooth oled

[kind walrus]

I said this for like a whole year and people would try start arguments over it

I was literally witnessing it first hand

[deft quiver]

Bruh

[tall pelican]

1: 1000w wasnt available to public at the time
2: I also cap modded, so shunts were like an extra 2 mins of work

[deft quiver]

Ah I see

What did you replace the caps with

[kind walrus]

Yeah the best you could do was 600w vbios iirc?

[deft quiver]

Just higher rated ones

[tall pelican]

nothing, just populated empty pads on suprimx

[kind walrus]

Fitz yours is suprim right

[deft quiver]

Ah

Kk

[tall pelican]

yeah

[kind walrus]

So it would've been liquid vbios

[deft quiver]

I killed my 7900xtx volt modding

[tall pelican]

it was 600w, but would regularly hit PL flags in I think bf2042 at the time?

[kind walrus]

Damn

2042 gpu stress test 2023

[tall pelican]

2022 I think?

I got it like launch week

[kind walrus]

Is it really that new? I swear I've known about 2042 forever

[deft quiver]

I'm probably just get a 4070 or something

[kind walrus]

I'd have no idea tbh

[deft quiver]

5090 should be out later this year

[kind walrus]

Painge

[deft quiver]

Yea

I didn't even do that bad

[kind walrus]

Did it manage to bridge or

[deft quiver]

Idk

[kind walrus]

Hm, maybe late to say this but a multimeter might discover something if you haven't looked

[deft quiver]

Na

I just gave it to a friend

Haven't talked to him in a bit

Let me go ask if he was able to fix it

[kind walrus]

Alr

[deft quiver]

Nope he said it's dead

[kind walrus]

Sadge

[deft quiver]

Probably a internal short on the die

Cuz the PCB looks fine

[kind walrus]

Very painful fix then

Reballing a die is not easy afaik

[deft quiver]

Yea

That's the only thing I can think of

[abstract merlin]

You wanna send it to a gpu repair specialist?

[fervent talon]

Can send it to me

[deft quiver]

Who

[abstract merlin]

Northwest repair

[mental turret]

sned it to me

(ill steal your gpu core)

[deft quiver]

It ain't even that good

[deft quiver]

https://www.3dmark.com/spy/42683748

[sterile flame]

By overclocking you implicitly give up all manufacturer warranty and warranty from place of purchase
This is a sample policy

[sudden torrent]

In theory yes

In practice there's generally no way for them to know if you overclocked a component

[mental turret]

overclocking profiles go crazy

[sterile flame]

It can be determined by measuring the voltages

Warranty providers aren't stupid

[sudden torrent]

What voltage

It'll run at defaults when you plug it into the test system

[sterile flame]

Oh I am too new to this stuff

[dull ginkgo]

All the profiles are stored only on the motherboard, unless you edit the spd on the ram sticks to change the stored XMP profile for some reason

Deterioration from overclocking is also very hard to distinguish from normal wear from normal use

And some companies will explicitly permit overclocking

[tall pelican]

iirc there was something that amd had mentioned they can tell when a fuse is blown, but they dont care about it for warranty purposes

[sterile flame]

Not NVIDIA.

GPU overclock is not supported

[dull ginkgo]

Warranty would go through the AIB partners, and for example, EVGA and MSI have said before that overclocking is fine

(rip EVGA GPUs)

[sterile flame]

Yeah. MSI Yeah
GIGABYTE NO

[dull ginkgo]

But in general, if you don't tell them, they usually can't prove it

Unless you did hardware mods

[sterile flame]

d applies to oc probably

They can interpret it at their discretion

[dull ginkgo]

Yea, most of the time they'll interpret it in good faith, but it's still there to prevent abuse

[sterile flame]

I believe OC != (abuse or misuse)

Unless obviously you knowingly raise the frequency too high

[sudden torrent]

Frequency too high won't cause damage at all

Only voltage will do that, and GPUs are locked so you can't increase the voltage much anyway

CPUs you can definitely kill with excessive voltage

[mental turret]

naahhhhh

[kind walrus]

The cards they ship out the factory these days are too hard to kill with general OC. There's absolutely no chance of them figuring it out

The fingerprints from someone handling the card are easier to detect than playing with the sliders in afterburner

They'll say they don't endorse it but they very much provide the playground to support people doing it

If they didn't allow it, apps like afterburner wouldn't work...

[dull ginkgo]

I guess the one thing they can tell is if you flashed on a different vbios for a higher power limit

[sudden torrent]

But if you flash it back to stock bios before sending it in they wouldn't know

[kind walrus]

[mental turret]

Explode it

[frigid locust]

The exception is CPUs having fuses in them that blow if you OC it

[sudden torrent]

It's unclear exactly what conditions trip that fuse, and it's entirely possible that the auto overclock that a motherboard does at default settings could trigger it.
As a result, both AMD and Intel still honor the warranty with that fuse blown.

[frigid locust]

it's still a little "Trust me bro" on that

[steel zephyr]

I've had Asus replace under warranty an Apex series overclocking motherboard that still has some vaseline and liquid electrical tape on it, obvious signs of sub zero cooling overclocking. They had no issue with it. EVGA replaced two of three 3090's for me that they knew I overclocked with sub zero cooling(coolant chilled in a dry ice bath).

I haven't found any cases of people having their PC hardware warranty denied due to overclocking unless there was major physical damage to the component or they replaced soldered components on the PCB. Maybe its happened, but its so rare that its a very minor exception and not normal.

[frigid locust]

my point was really not that it happened, but that PC part manufacturers are ultimately multimillion dollar companies, not our friends

[steel zephyr]

Unless the company is financially struggling or they have very high warranty return rates, they won't look into warranty returns that closely. Not only is it expensive to do a detailed analysis, but the negative press they would receive from someone falsely accused would hurt their sales. In addition to that, most people working at PC hardwarewarranty replacement centers are lower paid employees who jsut don't care.

[steel zephyr]

Now that i think of it, the only company that I dealt with that regularly denied warranty claims for the stupidest and most minor BS reasons was Gigabyte USA.

[steady lance]

Really? thought gigabyte was supposed to be a good brand.

[frigid locust]

I mean

There ain't no "good brand"

None of them are our friend

all we are to brands is wallets

[steady lance]

I wish there were more purpose driven companies.

Especially wish there was one in the PC area.

[frigid locust]

EVGA was pretty good, but they are currently going under so

[steady lance]

Yeah that's unfortunate.

[steel zephyr]

Their US RMA/Warranty service center has had major issues for 15+ years now. THey often send warranty replacement parts that are defective or DOA. They damage customers' computer components and claim it was sent that way. They damage customer's PC components and deny warranty coverage because they damaged the components.

[steady lance]

jesus christ.

Seriously?

[steel zephyr]

For example, I take photos of every piece fo PC hardware I ship for shipping insurance. I sent a motherboard into Gigabyte for RMA repair. 5 weeks after they received in and 3 weeks after they told me it was a BIOS issue that they needed to repair, they said the board was damaged by me and theya re denyign the warranty claim. It was a scratch ont he board that wasn't in any of the photos form when I shipepd it. The manage at their RMA center called me a liar and accused me of photoshopping the pre shipping photos.

One of my friend sent his GPU to Gigabyte USA in for warranty repair and received it back, 5 or 6 weeks later, with a long and deep gouge in the plastic shroud. The gouge was not in any of the photos of the GPU from before he sent it in. Gigabyte claimed they received it that way and refused to replace the shroud.

[steady lance]

how tf arre they still in business with that kind of support.

[steel zephyr]

Gigabyte has sent me DOA replacement motherboards more than once.

[frigid locust]

TBF the shroud being damaged doesn't really affect things

But that they're trying to deny warranty on grounds of damage they caused is a travesty

[steady lance]

I am very glad I did not go Gigabyte when I upgraded then. I would'

[steel zephyr]

When Gigabyte was ransomwared during the pandemic, they lost multiple customers GPUs and motherboards and some were never replaced.

[frigid locust]

Yikes

that's literal theft

[steady lance]

deleting the message cause mods may not like that statement of mine.

how is MSI's support?

that's my current motherboard's brand.

[steel zephyr]

I haven't had issues with MSI, but I've rarely had issue with MSI hardware to require support.

[frigid locust]

It's funny how hard I got whiplashed

I'm sure I already told the story but here I am again

[steel zephyr]

I've been building custom PCs for other people as a side business along with taking care of friends and family members PCs for 25-20 years, so I guess I get to experience more RMAs than the average user. Gigabyte is the only company I had issues with consistently.

[frigid locust]

I used to have a really old Asus laptop (X751LA if anyone is interested, i3-5005u and GTX 970M)
I had it for 9 years straight with no core component failure whatsoever (though the hinges literally broke out of the plastic bottom shell, the keyboard started falling apart after 6 years and the CPU fan finally died at year 9 so I had to stop using it because it thermal throttled at idle)

and now that I got an Asus mobo for my desktop thinking "It's Asus, it'll be awesome and work for a long time"

I've had NOTHING BUT PROBLEMS

[steady lance]

ouch.

[steady lance]

Lmao, I have a Dell laptop with almost the exact specs.

[frigid locust]

first it was Bluetooth refusing to work at all
Then it was CTDs under any kind of load
Now it's the BIOS simply refusing to update whatsoever

[steady lance]

same gpu, just a newer cpu from 6th gen it's a 2 core I5

thing is now 8-9 years old, still working..

[frigid locust]

yeah mine is a 2c/4t i3 from 5th gen lmao

it'd still work if the cooler would work

but it was NEVER repasted and the fan is dead on it

[steady lance]

I ripped the storage out of it and plugged it into my desktop.

[frigid locust]

and the chassis is so messed up that if we open it up we are NOT putting it back together

[steady lance]

ahh in my case it's minor damage to the USB covers but otherwise works. had to replace the battery once and regular dedusting.

[frigid locust]

I actually replaced the 1tb HDD in it with a 480gb SSD
And use the HDD as my bulk storage in my desktop

now I'm half tempted to rip the SSD out and add it to my desktop

but again if we open it up, we ain't putting it back together ever again

[steady lance]

I replaced mine with a 512GB hybrid ssd.

thing is very weird.

it shows up as two drives in the system.

split in half

[frigid locust]

that's cool

but yeah I do need more storage

[steady lance]

same.

[frigid locust]

I'm eyeing a 2tb SATA drive for my second upgrade

first is a new case that actually has airflow

[steady lance]

Ideally I'd like a setup with 1TB boot, 8TB game, 8TB work, 8-16TB cold/extra storage.

Not so bad nowadays but when I was doing my work and video recording/editing I was deleting storage in the hundreds of GBs a day.

[frigid locust]

my "ideal" is a 1tb PCIE Gen5 boot drive, 2tb PCIE Gen 4 game storage, 4TB SATA for game captures/stream VODs and 8TB cold storage

well

technically that's not THE ideal

[steady lance]

pretty similar setup just different capacities.

[frigid locust]

but what's actually affordable here

Double every number and you get the ideal

[steady lance]

bloody hell.

I just realized I have been trying to get this CPU replaced with AMD for 6 weeks now.

and an additional 5 weeks of testing prior to giving into the RMA process.

[frigid locust]

Yikers

@steady lance BTW it appears that the issues relating to game latency penalties with the 2 CCD X3D chips got fixed

so as soon as you get your replacement, you'll actually get the sort of benefits you'd have wanted

[steady lance]

Oh neat.

I had set the games to use specific cores so I never really had any issues tbh

Bu that's good to know.

[steady lance]

Okay I couldn't help myself. I just tried to apply 6600mhz on my ram with 1.4v DRAM/VDDP/VDDPIO didn't boot so I turned it down to 6400mhz at 1.38v and all booted

running a quick and dirty test of Anta777 on TM5.

How do I configure the config to use all 64GB of my ram? it's only using up about 45GB.

success. no errors detected.

6400cl32-38-38-80-118

[sudden torrent]

You can edit the config files manually, they're just text files

6200 is the best I got stable but I think that's a memory stick limitation

I got that jedec speed kit that's known to be hynix dies

[kind walrus]

You need to change the page size

In tm5

For 64gb I think it's like 1096mb or something

Been a while since I opened it but I change that, then you can also edit cycle count if you want to be more thorough (3 is fine, other stability tests on top would be more thorough than more tm5 alone*)

[steady lance]

Gotcha.

[steady lance]

Oh a 7950x3D or? Also what do you mean by memory stack limit?

[kind walrus]

He's on 7900x3d

Different die types, different rank sticks, different heat output, trfc settings

All weigh in a lot on what you can do, also just having a mid or meh imc can cause that too

Like a cpu just particularly doesn't like a certain type of die

[steady lance]

Ahh gotcha. I forgot he got the 7900x3D from a friend.

For a 7950x3D is 6400mhz the norm for ram limit at 64GB?

[kind walrus]

6000-6400 is very common ye

6600 is extremely lucky

Given we're talking ambient

[steady lance]

Part of me feels tempted to see if I can get 6600mhz to boot stable. The nightmare cou I had before would boot but was in no way stable with 1.6-1.65v dram, vddp and vddio.

[kind walrus]

You can also ask @modern walrus if you want a bit further debrief on things you can try

He's OC'ing his 7900x3d rn

[steady lance]

Oh okay.

[kind walrus]

He's new to AM5 but he's very good at ram lol

He has the patience I don't

@sudden torrent u too, he might know something to get you to 6400

[steady lance]

@modern walrus, when you have time I would be very interested in learning from you about RAM oc. I am also on the AM5 platform.

[kind walrus]

[steady lance]

I wonder if I have the patience, xD.

[sudden torrent]

Nah it straight up would not post no matter what I did, even SoC to 1.35v just to see

[kind walrus]

The difference between him and me is he's already spent longer than a day on it lol

[kind walrus]

Omg

Don't set 1.35v again btw

Don't risk cpu popcorn

Lolol

[sudden torrent]

It was only for like 5 minutes lol

[steady lance]

Thought popcorn was 1.4+?

[sudden torrent]

Yeah

But still

[kind walrus]

Even still, remember the msi ryzen master incident

[steady lance]

Can't say I do.

[kind walrus]

There's a ryzen oc tool from msi, it basically lets you set whatever voltage

[sudden torrent]

Also remember when AMD GPU drivers automatically overclocked your CPU?

[steady lance]

Huh?

[kind walrus]

You can immediately kill an x3d if you use that software

[steady lance]

Yikes.

Glad I did not touch it assuming this was very recent.

[kind walrus]

PBO is

VERY

Smart

It can accomplish what static and manual numbers can't really do

[sudden torrent]

CO alone gets me a good 200mhz when boosting

[kind walrus]

So if you think an X3D idling at 1.3V means you can run 1.3V through it, think again

Instant death

[steady lance]

From what I've come to understand about AMD CPU oc is this. Just use PBO because the gains otherwise are little to non existent and often lead to death of processor.

[kind walrus]

I think in that article the x3d died at 1.2

Just instant death

Immediate 00 post code

[kind walrus]

Full marks

You got it in one lol

[steady lance]

Thank you.

[sudden torrent]

1.1V is the absolute max for a reason on X3D

[steady lance]

On 7000 series?

My main board defaults my 7950x3D to 1.25-1.3v SOC

[kind walrus]

I think x3d in general

Soc is different

This is vcore

[steady lance]

Oh I never touched vcore but my board gives me the option to albeit limited iirc.

[kind walrus]

You can look in hwinfo 64 to see what the cpu vid is reporting (rare time it's useful to look at)

Under load you should be somewhere around 1v

0.9v

[steady lance]

Gotcha.

[kind walrus]

Depends on the chip

[steady lance]

I will take a peak at it in the morning.

For now I left HWinfo on my PC to run overnight with logging enabled.

[kind walrus]

Fair enough

[steady lance]

My plan is this,

Give it till Monday next week to make sure all is working (leaving 6400mhz mild oc) then if all is good then I will try tuning my ram then pbo. Or do you gets recommend trying pbo first?

[kind walrus]

If you're worried about the cpu run stock for a bit longer

Do ram with stock cpu, then pbo

Pbo can take a long time or not long at all

[steady lance]

Aight.

[kind walrus]

But if the cpu is at all unstable the ram is much harder to test

[steady lance]

So tune ram with stock CPU settings then tune CPU with pbo?

[kind walrus]

Mhm

[steady lance]

👍 will do.

[kind walrus]

If you have an eclk board, that's also an option

Slightly better/same

[steady lance]

It's a MSI X670-P

Not sure if it hZ it or not.

[kind walrus]

Hm idk that one

You'd have to look into it

[steady lance]

Aight.

[kind walrus]

Anyway I'll be back on later

[steady lance]

Eclk is for base clock oc right?

I am gonna hop off and sleep. It's been fun chatting with you guys.

[sudden torrent]

Ace and Godlike are the only MSI boards with ECLK iirc

[sudden torrent]

Yes but the advantage is you can apply the OC to only the core frequency and memory, so you don't get SATA and PCIe bugs

[steady lance]

Was not aware of those issues.

[sudden torrent]

Yeah there's a chance that your SATA devices simply decide to die after BCLK 103 or so

[steady lance]

What does 103 stand for?

[sudden torrent]

And if the OC gets applied to the PCIe slot then it's overclocking the GPU too

103MHz

Normal base is 100

[steady lance]

3mhz can make that big of a difference?

[sudden torrent]

So going 3% out of spec can kill a SATA drive

[steady lance]

Yikes.

[sudden torrent]

And GPUs automatically overclock themselves with the boost algorithm so it's entirely possible that it just crashes with an extra 3% it didn't expect

[steady lance]

I have a sata drive with a lot of important stuff on it so I probably won't attempt base overclock without a eclk board.

[sudden torrent]

You shouldn't attempt a base oc on X3D at all

[steady lance]

Just use PBO right?

[sudden torrent]

The results are worse than stock

Yeah PBO

[mental turret]

(or buy intel if you wanna OC)

[sudden torrent]

Let me know when Intel can get me the most fps under 200W

[steady lance]

Honestly I was two clicks away from having gone Intel at the start of this until I decided that I'd prefer to have the option to upgrade.

[mental turret]

you know whats funny

i daily am5

[steady lance]

7600X I am guessing?

[mental turret]

77

I only do because I got my board for like 50 bucks

[steady lance]

Ahh nice.

My 7950x3D Core VIDs in HWingi are reporting 0.87-1.39V with a average of 1.25v

[sudden torrent]

At idle though

[steady lance]

Yes

[sudden torrent]

Idle doesn't matter because there's very little current

[steady lance]

What's the max it should reach?

[sudden torrent]

Under load 1.1V if you have PBO on

Idle can tag 1.4V no issue

[steady lance]

What about PBO off?

I have not touched PBO.

[sudden torrent]

Around 1V

Depends on bin

[steady lance]

Gotcha. So just keep an eye on it when running under load here and there?

Is there any benefit to enabling PBO without touching overrides like +200mhz?

[sudden torrent]

PBO enabled (not advanced) will still allow it to boost higher under load

[steady lance]

Ahh okay.

Will do that tomorrow.

I gotta sleep.

Thank you, Fal.

[sudden torrent]

Sure, g'night

[modern walrus]

sounds good I'll try to help as much I can. I'm currently trying to OC memory myself lol

[modern walrus]

I tried up to 1.3 for 6600 & it wouldn't cooperate. For whatever reason, I seem more stable with VDDQ lower than VDDD. 1.32 VDDQ

I lasted 104 iterations of ycruncher VDT at 6600c34-38-36 at 1.4 VDD/1.32VDDQ with everything else on auto. SOC voltage 1.255 & FCLK set 2133

VT3 is rough... ycruncher website says there were systems they tested at stock settings that don't pass VT3 lol

maybe I'll just use Karhu & pretend everything is fine lol

VT3 makes JEDEC timings your friend

[sudden torrent]

I've tried every step of 25mV on SoC from 1V to 1.3v and it simply would not post 6400. Even with the extra training times enabled. That took several hours alone and I'm done, settled on 6200.
Haven't tried much past BZ easy timings yet. I can do that when I have time but today is not that time.

[steady lance]

Could you explain to me the difference between DRAM, VDDQ, and VDDIO voltage? My board likes to set them all equal to DRAM. On my previous 7950x3D I had tuned 6400cl30 with the rest of the timing pretty tight at 1.6v. I am using Hynix likely A die.

[steady lance]

#overclocking message

I believe those were the timings.

[wheat lintel]

Tell that to my 3700X

[kind walrus]

*Looks at it with a face only a man who uses PBO would make *

[wheat lintel]

Well what is left of my 3700X at this point

My damn dog ate my 3700X

Was able to bench 4.4GHz all core at 1.35V set voltage, thing was fun af

[modern walrus]

I gotta be totally honest, I don't know exactly what each does. I know that DRAM voltage (VDD) is the voltage fed to the ICs and is typically the highest. I think VDDIO & VDDQ is more the communication between the memory & the CPU. VDDQ is usually set the same as or lower than DRAM voltage. VDDIO seems to be the main voltage that affects frequency

This is my first time using AM5 & first time using an AMD CPU in about a year or so

I'm kind of trying to make sense of it still

which board sets them all the same? On auto they're all like 1.1v for me but to get higher frequency than auto I've been setting them manually

[modern walrus]

what did you set VDDIO to?

[modern walrus]

there are similarities but each CPU behaves different

[steady lance]

@modern walrus what CPU, board, and ram are you using?

[modern walrus]

7900x3d, X670E Taichi, & some cheap Klev A-die 2x32gb

[steady lance]

I am running a 7950x3D, MSI X670-P and 2x32 Kingston fury hynix ram (never confirmed if it is A or M)

[modern walrus]

like cheap enough that they don't even have XMP lol

[steady lance]

Yikes.

[modern walrus]

lol A-die is A-die

[steady lance]

I think it's A cause low temps at 1.6v and 65K tRFC.

[modern walrus]

on my 13900KS/Z790 Apex setup, G.Skill 7200 a-die & TeamGroup 5600 A-die run the same

[steady lance]

Impressive stuff you have.

[modern walrus]

I kinda like the cheap memory anyway because I get to avoid the mission of getting the heatspreaders off for watercooling

[steady lance]

Gotcha.

[modern walrus]

thanks. It's kinda dumb ig cuz I mostly use it for benchmarks & hwbot lol

like my Z790 Apex has never even connected to the internet lol

[steady lance]

I was lucky and the kit I got was on sale at I think 20% off making it the cheapest 6000cl32 or better kit in my area at the time.

This was after purchasing a Teamgroup Create ram kit that was damaged in transit delaying my PC for 2 weeks.

[modern walrus]

I used to think AMD memory OC was silly until I found out MCLK was running 1:1 to UCLK

Intel runs 2:1 on DDR5 always

[steady lance]

1:1 is preferred right?

[modern walrus]

generally ya

maybe if you're running like 8k+ it might be worth it but I kinda doubt it

I should try running a 2x24gb kit on this board at both 2:1 & 1:1 to see how much difference it makes

2x24 is still single rank so higher frequency is a lot easier

I couldn't get this 2x32gb kit to POST any higher than 7600 on Apex

2x24 kit will do like 8600

[steady lance]

Gotcha.

Do you happen to know the limits for VDDQ and VDDIO on AM5? I can't find it anywhere.

[mental turret]

all my homies love gdie

[sudden torrent]

100mV below mem VDD

Originally tried VDD=VDDIO

Found no difference

[sudden torrent]

The most I see people recommend doing is 1.35V because lower is often more stable, and they shouldn't be set above memory VDD which is usually good at 1.35-1.37V for any frequency you can reasonably expect on Ryzen 1:1.

[steady lance]

VDDIO?

Just enabled PBO and I am already seeing higher boost clocks past 5.7GHZ

[steady lance]

Cores 8 and 9 (non v-cache) seems to be my cpu's best cores. Hits 5.75Ghz+ regularly like 50% of the time and HWinfo records their average frequency to be 5.5ghz.

so far, core wise this CPU seems to boost higher then my old one did.

all cores seem to boost higher and more frequently and there are two cores instead of just one that can hit beyond 5.7.

[modern walrus]

heck I'm running VDD 1.43 & VDDIO 1.32 for 6400

anyone remember how to test only FCLK?

I swear it was like Prime95 Large FFT+OCCT something at the same time

nvm I found it. P95 Large FFT+OCCT VRAM

[steady lance]

I am looking forward to the weekend so I can spend a solid few hours or so to start tuning

[wheat lintel]

I should get back to working on stabilizing 8000 MT/s

[modern walrus]

good luck lol

I think I'm calling this good enough

[steady lance]

Interesting, with PBO enabled during a lite-workload I noticed all my frequency cores hitting a all-core workload of 5.55ghz

@modern walrus, what test suite do you use for ram stabillity? and settings for each one.

[steady lance]

try pushing 1.6v like me 😄

[modern walrus]

I don't have PBO enabled yet. I was trying to test memory OC first before I introduce another variable. I'm using yCruncher (http://numberworld.org/y-cruncher/#Download) VT3 test because I heard that was the hardest to pass--I could be wrong or have old info, of course. When I open ycruncher, I press 1, 8, 18, 0 lol

[modern walrus]

ya I was thinking I might try to drop tCL since that's likely one of the few places left for improvement since I can't do 6600

the other thing I might mess with is tRFC but that's kind of a rabbit hole I wanted to avoid

[steady lance]

Agreed, that one took me days to work out.

[modern walrus]

but rn I'm also trying to push FCLK higher... speaking of rabbit holes lol

I haven't really gotten a clear answer on what the proper ratio is with FCLK to UCLK & MCLK

[steady lance]

When I start RAM oc this weekend I will disable PBO assuming I do ram first.

Just interesting seeing this CPU boost so much so frequenly unlike my other 7950x3D.

[steady lance]

FCLK iirc is unsynced so higher is better

[modern walrus]

I've seen MCLK/1.5 shold be FCLK but I've also seen it should be like 3:2:3

which is that the same math? I suck at math lol

FCLK = (MCLK x 2)/3? lol

[steady lance]

UCLK = MCLK and FCLK is completely unsynced so just tune that sucker up as high as you can.

[sudden torrent]

It doesn't matter much but 3:2:2 is technically slightly better, but if you can get it 66MHz higher that takes priority

[modern walrus]

ah heck I guess I land at 2133 regardless of which of those two equations don't I? heckin math

[steady lance]

Isn't FCLK unsynced on AM5?

[modern walrus]

2167 seemed to be running well enough

but I'm not sure VT3 would necessarily show FCLK issues unless the settings are way off

[sudden torrent]

Yes but there's a very minor latency improvement when it's 3:2

[steady lance]

Gotcha.

Is that latency overcome with for example that 66mhz you mentioned?

[sudden torrent]

Yes if you can increase the fclk by 66MHz that more than overcomes the small bonus to latency

[steady lance]

Good to know.

[sudden torrent]

33MHz, it's basically equal

[modern walrus]

these results seem suspicious https://www.techpowerup.com/forums/threads/ryzen-7000-am5-memory-guide-ddr5-6000-stability-slow-boots-memory-training-etc.315236/

[steady lance]

So at the end of the day, just tune the sucker to the moon if possible.

[sudden torrent]

Yep

As long as it keeps scaling

[modern walrus]

& stay under 1.3v according to Igor's labs

[sudden torrent]

Once it stops scaling that means you're getting silent errors that get corrected

I was "stable" at 2200 fclk but it would be worse performance than 2133

[steady lance]

1.3 vscoc?

[modern walrus]

heck I should just run ycruncher 1b like 5x each at various FCLK frequencies & see if there's a noticeable pattern

[modern walrus]

yes

[steady lance]

My CPU runs at 1.3 by default it seems.

Board has a hard limit of 1.35v

[modern walrus]

well Igor's lab could be totally wrong like they were about those alphacool fans so grain of salt lol

[steady lance]

Does Vsoc affect ram or cpu OC? It made no noticable difference when I was doing ram oc on my old CPU at 1.35v so I left it at default.

[modern walrus]

I believe it should really only affect FCLK

[steady lance]

ahh I never bothered with FCLK. by the time I reached there I learned my old CPU was faulty and spent a month trying to fix that

[sudden torrent]

Now's the time to do it then

[steady lance]

How risky is it to run 1.35v vsoc?

[sudden torrent]

Well, 1.4V+ was what caused the CPU popcorn

[steady lance]

So I can decide if I even wanna attempt higher vsoc for potentially better fclk.

[sudden torrent]

So it'll probably have accelerated degradation

[steady lance]

any clue as to if that's like 10 years cut down to 5 or?

[sudden torrent]

The platform isn't even that old so there's no data on that

[steady lance]

solid point.

Well I will see what I can do with the default of 1.3 and decide from there if I wanna try higher.

What did you guys say I should limit VDDIO voltage to? I forgot.

[sudden torrent]

A little below 1.35V seems to be the sweet spot. Too low, it's unstable. Too high, it's unstable.

[steady lance]

VDDQ should equal or be set below VDD (DRAM voltage) correct?

[steady lance]

gotcha. What does VDDIO affect on the ram? I think you guys said frequency but idk if I remember correctly.

[sudden torrent]

It's the voltage of the IO die, which communicates with the memory

Faster memory, faster communication, higher voltage needed

Until it becomes unstable

[sudden torrent]

VDDQ = VDD, or VDD + 50-100mV

[modern walrus]

gosh darnit windows

three 1b runs in a row & variation is like .020

the fourth jumps 2s

the fifth jumps 5s from the first three

[sudden torrent]

You're not using a neutered bench OS are you

[modern walrus]

I should be

[steady lance]

neutered bench os?

[modern walrus]

instead I'm on 23H2 stock lol

[sudden torrent]

An install of windows with everything you can disable, disabled. Telemetry, updates, apps, everything.

[modern walrus]

I restarted & disconnected ethernet lol

[steady lance]

ahhh I wish I had the option to do that.

[modern walrus]

I mean you can for benching but you destroy Windows Defender & other things that might be useful so it's not a great idea for daily use

for daily use stripped Windows maybe try ghost-spectre or tiny

[sudden torrent]

You're just asking to corrupt windows at this point.
I thought I was ok using my daily to test CO. I was wrong. -50 on CCD0 and I killed the bootloader.

[modern walrus]

lol win11 breaks easy

[steady lance]

"1b runs"? what is that.

[modern walrus]

ycruncher

[sudden torrent]

ycruncher, calculating pi to the 1 billionth digit

[modern walrus]

it's a benchmark that goes pretty hard on CPU & memory

if you want a bunch of benchmarks to mess around with, google "benchmate"

it'll have Cinnebench, ycruncher, SuperPi, & some others

[steady lance]

is the variation of .020 good or bad?

not sure what a normal run is like

[modern walrus]

I mean really heckin close

that's like 2 hundredths of a second

[steady lance]

Oh is variation the calculation time?

[modern walrus]

which if you don't know scores work, might not be obvious

[modern walrus]

yes

[steady lance]

Now I understand what you were talking about then.

I was confused by what exactly "variation" meant there.

[modern walrus]

ya sorry I should be more clear lol

[steady lance]

CPU overheating or?

[modern walrus]

this is the spreadsheet so far

[steady lance]

Seems odd that it jumps so much.

[modern walrus]

I doubt it since I'm on stock on custom loop

[steady lance]

weird.

This is yCruncher right?

Does it say if you run into errors? I am considering trying it out.

[modern walrus]

yes. it'll say Coefficient is too large if it can't finish

but that doesn't really give you much more than that

[steady lance]

So it's good to know if the system is unstable but bad at giving the cause of instabillity. that sum it up?

[modern walrus]

yes

it's a lot better for showing instability to run it by itself

just running for 15 seconds will only show you if you're really not stable lol

[steady lance]

So what do you do when you run into issues? Run a different test to find out the cause or tune settings down one by one?

[modern walrus]

I try to change settings one at a time or small groups so I know which one is the likely culprit

[steady lance]

Hope I am not boring you with all my questions, haha.

[modern walrus]

haha no no worries

I owe it to the universe to answer all these questions as best I can

cuz like 4 years ago, someone else was answering them for me lol

[steady lance]

Gotcha. Also I wonder if running VDDIO 1.6v could've been what caused my CPU to go from unstable to really unstable.

Cause my board sets VDDIO and VDDQ 1:1 with VDD voltage unless otherwise specified and I could not for the life of me find answers on VDDIO limits.

[modern walrus]

VDDIO 1.6 seems really high. Over 1.42, my motherboard changes the number to red

[steady lance]

for me numbers go red at 1.48v

and my board's voltage and no I did not actually run this but apparently lets me set voltages as high as 3.

[modern walrus]

it does the same for VDD & VDDQ tho which is exaggerating imo lol

3? holy crap lol

[steady lance]

Yeah. I never applied it but I wanted to see what the limit was.

[modern walrus]

highest a board let me go for memory-related voltage was 2.2

[steady lance]

It hardcapped CPU to 1.35v but lets me crank anything ram related up to 3v

doesn't make sense lmao.

[modern walrus]

it let you set CPU voltage on X3D?

[steady lance]

Vsoc. but I believe it does.

[modern walrus]

I thought X3D only allowed PBO cuz they dont' want you to bork vcache or something

[steady lance]

I will need to double check.

[modern walrus]

ohh SOC okay

[steady lance]

Vcore is the one your thinking of right?

[modern walrus]

yes

[steady lance]

I saw the setting somewhere in my bios.

never touched it though.

[modern walrus]

that seems high for LN2 lol

[steady lance]

I will need to confirm but I am pretty sure my board lets me set vcore voltage.

X670-P from MSI if you are curious.

[modern walrus]

ah okay. I'll see if I can find the BIOS manual. Come to think of it, I haven't really looked to see if I can set vCore

so I might be saying nonsense rn lol

[steady lance]

Vcore is what allows for proper frequency oc correct?

[modern walrus]

yes. That's normally how you'd increase frequency but with x3d I think it's different

like I don't think you can set 5ghz for all core like that

[steady lance]

Okay now I gotta run a all core run lmfao.

Cause in HWINFO I keep seeing my cpu on PBO hitting 5.55ghz

all core

cinebench fine?

I have some benching software saved in a folder might be a lil old though

@modern walrus, running R24 in the background on all-core test.

Note, results aren't perfect since I have a few apps open but I just realized when you said All-core you meant all 16 cores didn't you?

I was thinking core sets. like all-core for the Frequency set.

[modern walrus]

Cinebench is good for testing CPU but the score barely changes if you run JEDEC or super tight memory OC

well CB is kind of good for testing CPU stability

[steady lance]

ngl I love watching my CPU demolish apple's M1 and M2 processors.

[modern walrus]

There are much better ways to test CPU stability (OCCT, Prime95)

it was kind of a joke a few years ago to call your CPU overclock Cinebench stable lol

[steady lance]

I have OCCT.

[modern walrus]

lol ya I'm not a big apple fan myself

[steady lance]

running R24 to see all-core boost not for stabillity just to be clear.

[modern walrus]

OCCT small fft one of the better tests for all-core overclocks

[steady lance]

hitting 5.05 all core on CCD1 and 4.95 on CCD0 (V-Cache)

[modern walrus]

but for like PBO if you're setting each core magnitude, there's a thing called core cycler that'll test each core individually & log it so you can see which one crashes

heck I didn't realize they ran at different speeds lol

[steady lance]

I have a old version of core-cycler.

[steady lance]

Without PBO, CCD1 caps at 5.7 and CCD0 caps at 5.5.

[modern walrus]

what do they cap at with it? +200mhz respectively?

[steady lance]

I haven't used advanced mode on the cpus yet but I believe so.

[modern walrus]

looks like June 2023 was the last update https://github.com/sp00n/corecycler/releases

[steady lance]

Just enabling it by default adds at least 50mhz.

[steady lance]

ahh not to out of date then. mine is April 2023.

[modern walrus]

heck I hope core cycler still works or I'm never gonna finish core offsets lol

with 16 cores, it'll take you like a month lol

[steady lance]

If you want I can run Core-cycler while I work today to test for you.

[modern walrus]

nah no worries. If you're gonna test, maybe run the VT3 thing if you're not entirely sure if your memory OC is stable

[steady lance]

Sorry I just realized I think I gave false info on CCD0.

Caps at 5.2Ghz.

[steady lance]

currently all I did was just a quick frequency bump. not really an oc.

[modern walrus]

so far this is not showing enough variation to really call it conclusive lol

[steady lance]

old cpu was the one where I spent a month tuning the ram.

is lower or higher better?

[modern walrus]

ah okay. Well, whenever you end up trying again, VT3 is supposedly great for stability testing

[steady lance]

Is VT3 software name?

[modern walrus]

I mean, ostensibly higher is better but by such a small margin

[steady lance]

scored 1930 in Multi-core on R24 with like 6 apps open.

[steady lance]

I don't mean the FCLK just to make sure.

[modern walrus]

I haven't messed with R24 much only because it's not worth much on hwbot lol

oh okay thought you meant FCLK lol

[steady lance]

Talking about y-cruncher run numbers.

like 15.808

[modern walrus]

oh sorry

lower is better

[steady lance]

all good, I should have been clearer.

[modern walrus]

ya it's just measuring how long it takes to calculate 1 billion digits of pi

[steady lance]

Is R1-5 different ycruncher tests or?

[modern walrus]

they're just separate runs of the same test

[steady lance]

gotcha.

[modern walrus]

I should run longer tests but I don't really wanna spend all day doing this

lol

also, since I'm on regular windows, all it would really take is windows randomly trying to update or some other nonsense to totally skew results

[steady lance]

Be like me, spend 5 hours a day on ram oc 🎃

[modern walrus]

I did that with 12900K haha

I missed DDR4 so much when DDR5 was new lol

[steady lance]

So far I've enjoyed D5 much more then D4.

OC wise.

[modern walrus]

If you were using Sasmsung bdie, DDR4 was pretty predictable

especially on AMD lol

[steady lance]

No. I believe I had some kind of D die.

[modern walrus]

Samsung?

[steady lance]

I forget. I think it was Micron?

but I think it was mostly due to how horrendous the bios on my 5700G was.

Not user friendly what so ever.

B550M DS3H mobo.

[modern walrus]

I have some Rev E Micron but I never tried anything with a D from Micron

I had fun with 5600g cuz it could sync FCLK up to like 4933 lol

[steady lance]

My timings scaled pretty well with voltage up until around 1.4v

I was likely board limited frequency wise. couldn't get 4,000 stable but 3966 worked perfectly.

[modern walrus]

well... I dunno I mean... I guess there's somewhat of a trend here

[steady lance]

run 20 times.

and average each fclk section

best average wins is what I'd do.

[steady lance]

@modern walrus, running core cycler in the background to see what happens.

Made some changes to make the run heavier.

such as increased duration.

[steady lance]

Passed Core cycler.

Well about to pass. One core left to test.

I think I will try enabling two threads for the next run to make it more strenuous.

[sudden torrent]

That makes it boost lower

It's a very different test

[steady lance]

oh really?

the description in the configs made it sound like it would work the cores more.

[modern walrus]

I'm doing that now

I'm only -5 on all cores with +200 boost set

kinda weird C4 only ran 4305 & C5 4767

0 hit 4940; 1 5051; 2 5053; 3 5019.

[steady lance]

Decided I am gonna do CPU oc first

Applied +200mhz override and did -5 CO on a per core level.

running core cycler over night

[kind walrus]

You should really do that second

Lol

[steady lance]

I know. It

it's just I am so tired of ram oc.

I will save it then do ram only or smth afterwards then cpu + ram

[kind walrus]

Damn man this is like the easiest platform to do ram on

Tbf, I did watch you for days changing tiny things

[steady lance]

not when you have trauma from the first round

[kind walrus]

You did a very good job of it

But also

[steady lance]

ty

[kind walrus]

You did a very good job of it

Too good

I don't have the patience to do one value at a time

I'll often shotgun several then test

And then check if I get issues

[steady lance]

CoreCycler run it in 6m tests or 12m tests?

[kind walrus]

Skips 1000 steps

[kind walrus]

Longer is better usually

[steady lance]

aight.

Hopping off for the night, leaving core cycler on and 🤞 power outage does not happen again ruining my test

5.5GHZ ON x3d core!!!

hopping off now but it's a start

[mental turret]

Now overclock ddr3 @steady lance

[sudden torrent]

no u

d3 boring now

d5 is more exciting, you never know when you'll get a single error after 60 hours of testing

[kind walrus]

or the board retraining completely differently after months of perfect stability

overclock ur ddr5 @mental turret

smh

[mental turret]

I did

[mental turret]

Ok name every ddr3 ic

[kind walrus]

Don't you daily jedec?

[mental turret]

no that was like a year ago when i was dailying x99

i forget when that was tbh

[steady lance]

No thank you.

Large capacity D5 is already a big enough nightmare with minute, minute and a half boot times.

@kind walrus, @modern walrus, +200mhz and -5 co is a success.

ran core-cycler all night

Just changed the CO to -10.

[steady lance]

Last night I had a false-reading. x3D core wasn't the one that hit 5.5. In anycase I am running a new test now to see which ones hit what.

CPU peaked at 66c all night during the test.

[steady lance]

Just to confirm. With PBO's behavior it will only boost past AMD's specs if able to correct? wondering because I applied a +200mhz override and so far I am seeing a +50mhz increase on all cores peak.

[steady lance]

running OCCT small with CoreCycler in P95 mode.

Interesting, when hit with a all-core workload it appears the 7950x3D's P-Cores will drop it's speed to match or be slightly higher than the V-core's clock speed.

Because when it's just the P-cores in use they boost a lot higher even if it's hitting all cores.

(calling the non v-cache chipset P-cores for simplicity sakes)

[mental turret]

D3 is actually fun to mess with

Unlike ddr5

[steady lance]

I finally solved why my PC's fans were always running at the same speed and why I couldn't change it with software.

For some reason my Bios set the fans to DC mode rather then PWM.

[sudden torrent]

Correct, it only boosts if there's the power, voltage, and thermal headroom to do so.
Did you adjust your power limits?

[steady lance]

Not yet.

Any harm besides power consumption in just maxing that out?

[sudden torrent]

Well that's probably why you only got 50 out of it

No harm technically but the algorithm might bug out. I just set it to "motherboard limits" and that's plenty.

[steady lance]

Make 500W or whatever my board limit is available to it.

Oh, motherboard sets it to the limits and not recommended?

I assumed it meant set to Motherboard recommended settings

[sudden torrent]

Motherboard limit for me is like 400A

[steady lance]

also I think P-Core 1 for me is my golden goose.

[sudden torrent]

You can check what it sets in Ryzen Master

[steady lance]

can maintain +5.6Ghz for about 50% of the time.

Will go up the power limit now.

btw I am on -20 CO now.

So far no issues.

[sudden torrent]

My chip does -40/-30 per CCD

[steady lance]

Oh wow. that seems really good?

[modern walrus]

the multi-core boost won't be nearly as high as the single core

[steady lance]

Unfortunately so.

Upping power limit to see if that helps bnrb

[modern walrus]

are you manually setting TDC/EDC/PPT?