overclocking

[modern walrus]

@sudden torrent do you set them or put board limits?

I'm at 150/180/180 respectively

[sudden torrent]

The -40 is on the 3D CCD and that's pretty normal for those. The limit used to be -30 but AMD "discovered" (after being told) that some chips could go lower.

[modern walrus]

lol I remember tryin to go below -30 by also setting vCore to have a negative offset on 5900x

[sudden torrent]

I just use the "motherboard limits", it's always been fine for me even in my AM4 board.

[modern walrus]

clock stretching ftl

[sudden torrent]

Judging from my 1B score almost matching Forks' 7900X on LN2 I'd say I'm probably not clock stretching

[modern walrus]

sorry meant AM4 lol

[sudden torrent]

Oh yeah, that I noticed too. I could actually do -25mV offset and still do -15 CO and get gains. Golden sample 5800X fr.

RIP

[modern walrus]

think this is gonna work?

[sudden torrent]

But any lower offset and yeah it would stretch

[modern walrus]

gonna smooth it more later lol

they're not compatible with x3d in their default state lmao

[steady lance]

I just applied board limits, might try manual later.

[sudden torrent]

I should lap my CPU...
Yeah the TG DD stuff is usually good. Roman knows his stuff.

[modern walrus]

before

I didn't mean to cut as much off as I did tho ngl

slipped with the tin snips

[sudden torrent]

Should still be plenty of material

[modern walrus]

also kinda not even but I got a little $35 dremel to do a bit more sanding jic

[sudden torrent]

Looks a bit bent but that can be fixed

[modern walrus]

it should still apply pressure all the way around I hope

tbh I was gonna get another one & start over

cuz that's a heck of a haircut

hopefully delidding isn't too bad

what do people do with Zen 4 CPUs when they delid? It's not like Intel where you can remove the solder & slap LM on instead then put the IHS back on

cuz of the silly IHS cutouts

[modern walrus]

I think where it looks bent is where I sanded from less of an angle so it shows the aluminum under the paint more there

it seems straight in person at least

like sanded at like, say 20 degrees instead of like 45 or 90

btw Fitz Polish

removes LM so well compared to anything I've tried

this had like one year old liquid metal on it that I couldn't even make it look different with acetone or isopropyl

gonna do it a few more times tho

ordered a whole bottle of fitz polish

That's how it looked

[tall pelican]

winks

[modern walrus]

lol well done

[steady lance]

Current peak is 5.825GHz on the P-cores.

V-cores remain the same at 5.25Ghz however the clocks are being sustained much longer

[steady lance]

@modern walrus, @tall pelican 5.925Ghz!

4/8 P-cores hit 5.8-5.825Ghz and one core golden core 10 hit 5.925Ghz.

the rest so far have peaked at 5.750Ghz but hit that very frequently.

V-Cores have not boosted paste stock of 5.25Ghz but are hitting their peak numbers regularly.

running Aida64 Extreme stabillity test

So far seeing a All-Core average of 4.7 +/- 100mhz.

with all core peaks up to 5.25ghz

[modern walrus]

heck that's good lol

I'm still trying to do per core magnitudes

cuz I love misery apparently

[steady lance]

thank you.

[modern walrus]

on hwinfo there's this option for snapshot polling for AMD CPUs

[steady lance]

also this is with -20 co which I am finalizing with aida test before trying -30

[modern walrus]

I think it gives more accurate clock numbers (not saying yours are wrong) so I always activate it on Ryzen

[steady lance]

ahh gotcha.

CPU core voltage is 0.93

[steady lance]

found the average.

aida is reporting 0.97v core average.

this is under 100% load

[modern walrus]

on all cores or on one?

I'm showing 1.032 rn running Prime95 on C0 with corecycler

[steady lance]

looks like all core

[modern walrus]

eh iirc there's a more accurate voltage reading

[steady lance]

is the aida64 test infinite duration wise?

running the Aida64 Extreme stabillity test

[modern walrus]

VRVOUT is most accurate iirc

dunno about Zen 4 but Zen 3 it was lol

[steady lance]

which program is that?

[modern walrus]

for some reason I have three different VRVOUT readings & theyh're all diferent so who the heck knows

it's in HWiNFO

[steady lance]

btw since fixing my fans not speeding up accordiingly my peak temps on this 7950x3D under load is 66c.

[modern walrus]

on Aida stress test? I never really used that one

Prime95 & ycruncher are really stressful so I figured those were good enough

[steady lance]

Aida and Cinebench R24.

this is also after running core cycler for two days almost without stop

[modern walrus]

sounds like you should save BIOS profile & keep dropping offset magnitude or manually increase PPT/TDC/EDC? that was my Zen 3 approach I dunno to what extent it still applies

[steady lance]

yeah I will definitely save bios profile before trying -30 co.

PPT is power limit, I forgot what the other two are.

currently set to board limits

[modern walrus]

increasing those allowed higher all-core frequency for me back then

TDC & EDC are the amperage limits

like Thermal Design Current & Electric (I forget?) Design Current

Electrical Design Current

[steady lance]

is it safe to increase the other two?

..... ngl when this chip works it's amazing.

[modern walrus]

I always treated those numbers as limited only by ability to cool them lol

[steady lance]

peak wattage of 100 and it performs +2x better then my 5700G did which consumed 89w

ran aida64 for 25m I think it's safe to say the oc is stable.

running CoD next with my friend to test game stabillity

what the?

It looks like C0D is using all 16 cores.

[sudden torrent]

And I seem to have zero, unless I hid it but I doubt that

[steady lance]

5/8 P-cores hit +5.8Ghz and one core hit 5.925Ghz with -20 co stable.

Now doing -25 co and running core cycler over night.

-30 did not boot guessing one of the v-core did not like it.

[sudden torrent]

The vcache cores tend to like lower voltage more than the regular ones because they boost lower

[steady lance]

@sudden torrent, @kind walrus , @tall pelican, @modern walrus, do you guys happen to know what the safe and max limit is for TDC Limit and EDC Limit using PBO? It's measuring in mA.

Wondering if it'd be a bad idea to just set it to 500, 500, 500.

[kind walrus]

yeet it

[steady lance]

PPT, TDC, EDC.

[tall pelican]

"safe" is max, but there will be performance regression after a certain point

[steady lance]

So just set it to the moon?

[tall pelican]

And max is like 1000w, 520A

[steady lance]

Mine lets me set it up to around 9999999

[tall pelican]

500mA will throttle in everything

[kind walrus]

it won't pull 500A if that's what you're thinking Apex

It'll hit some other proc event that will stop it going anywhere near that

[steady lance]

Oh okay. So it only pulls as needed?

[kind walrus]

Peak 7950X3D is about 165W

[tall pelican]

520,000mA is 520A if I can math at 9am

[kind walrus]

It won't try to go higher

Lol

[steady lance]

Oh okay. That's good to know.

I was worried that it was like setting a static voltage.

So I wanted to make sure.

[tall pelican]

Just leave scalar at 1x and you'll be safe

[steady lance]

Okay. Also what is GFX curve?

[tall pelican]

Igpu stuff

I disable igpu in bios so windows doesnt try to install igpu drivers

[steady lance]

Gotcha.

I noticed PPT limit is in mW.

Same as a watt or smaller measurement?

I don't know why but it bugs me that PBO boost override maxed out at 200.

[kind walrus]

wym

Mine doesn't even do 200

Lucky duck

Lol

[steady lance]

I don't think I have touched any there called a scaler or has any measurement that would imply things like 1x, 2x, etc.

[steady lance]

I don't know about lucky granted the suffering but ty.

Wow that image took like 8m to upload.

[tall pelican]

Pbo scalar ctrl

[steady lance]

Wth?

My CPU is running at 545mhz.

Feels like it too.

[tall pelican]

Because you have it at 350mw lmao

[steady lance]

Oh....

Well that explains it.

[tall pelican]

.

[steady lance]

500mA?

[tall pelican]

Tdc and edc

You need an extra 3x 000 at the end

So 350 would have to be 350,000

[steady lance]

So in order PPT, TDC, EDC, do this?

• 300,000
• 500,000
• 500,000

And other limited will activate but it actually draws those numbers correct?

[tall pelican]

If you want 500W/500A/500A, then yeah

[steady lance]

How much power can a 7950x3D draw if it really wanted to?

[tall pelican]

150-200

[steady lance]

Note I am using a 750W PSU rn.

[tall pelican]

My 7800x3d with igpu off maxes at like 70w

[steady lance]

Okay gave it 250,000 mW then so a little extra head room.

[sudden torrent]

I'm using the iGPU to drive my secondary monitor to avoid the Nvidia bug where the GPU doesn't sleep with multi monitor

[tall pelican]

Ooohhhhh rip

[steady lance]

Should I set TDC and EDC lower then PPT?

[tall pelican]

That's like an extra 20-30w even at idle

[steady lance]

Also, is there any benefit to using the 7950x3D's iGPU for production applications?

[sudden torrent]

Most apps don't know how to multitask like that

[steady lance]

Like with Intel's iGPU for example.

[sudden torrent]

Intel quicksync works only in a couple programs

[steady lance]

Ahh okay.

Does AMD's iGPU have anything similar that is used? If not I think I will disable my iGPU.

[sudden torrent]

Yes but it's not used often afaik

[steady lance]

Will look into that then.

Is the boost frequency on the V-cores locked to 5.25GHz?

I see the P-Cores go beyond stock but not the V-Cores.

(Calling P-Cored the normal ones and V-Cores the ones with V-Cache)

[sudden torrent]

It's probably hitting the voltage limit on 3D cores

[steady lance]

Gotcha.

Is running P95 L1/2/3 in conjunction with OCCT EXTREME Large a good stress test?

Does Prime95 call the cores "Worker"s?

[sudden torrent]

It depends on what you're testing.

[sudden torrent]

The threads are workers, there's 2 per core by default. The OS handles which worker is which logical CPU thread.

[steady lance]

So Worker 5 is likely v-core 3 (thread 1) then? Cores 0-7 are V-cache.

[steady lance]

cpu stabillity.

[steady lance]

okay I think I've figured it out.

found where I see all the cores/threads in P95. It appears core 5 is the only that fails, going to lower the CO on that one.

this is bugging the hell out of me.

Worker 5 keeps on failing.

Does P95 treat each core as a worker or each thread?

[dull ginkgo]

Depends on what you set it to do, default is every thread is a worker I think

[sudden torrent]

No they're usually not sequential, it's just a thread number not a core number

The OS decides what core works on what worker thread

[steady lance]

How can I find out which worker is which core/thread? @dull ginkgo, @sudden torrent

[sudden torrent]

You don't

[steady lance]

....

[sudden torrent]

It changes according to the whims of the OS

Dynamically

[steady lance]

well f me.

This is going to be pain to figure out with 16 cores.

[sudden torrent]

Core Cycler only knows what thread is which core because it assigns itself to specific cores

[steady lance]

gotcha.

Any clue why other stress tests haven't implemented that feature?

[sudden torrent]

Because they're multi core tests generally

It can only choose which cores it uses, not which worker thread is on which core

So core cycler does one core at a time

[dull ginkgo]

You could try like processor lasso or something similar and stress test 1 core at a time if you want

[steady lance]

Well here's to hoping I found the right core on this attempt.

[sudden torrent]

Yep that's basically what core cycler does

[steady lance]

Would that be different than using Core Cycler on P95 mode?

Wondering because this mystery core keeps passing hours and hours of Core Cycler without issue.

[dull ginkgo]

I think that should be what core cycler does

[dull ginkgo]

It could also be some sort of multi core problem

Like it only has problems when you're running into a current limit or maybe related to heat or power draw or something

[steady lance]

I am leaning towards the multi-core problem because I've ran CoreCycler in the background for about three days now excluding when I am running other tests.

Not heat because my peak temps are the same. So maybe it's current or power draw.

peak power draw is 147W.

🤞 I don't jinx myself just now but

Before Worker 5 stops normally 2-3m into the test. about 6-8m in so far without it stopping so far since dropping core 5's CO down to -5.

Running P95 (Small FFTs L1-3) and OCCT (Large, Extreme, Variable) at the same time.

[dull ginkgo]

An interesting combination

[steady lance]

Seems to stress my system more than P95 with Aida64.

I think I may've gotten lucky in that Worker 5 was also core 5.

15m in no issues.

Is a peak temperature of 64-66c under load good? Been running P95 and OCCT essentially without stop for a hour now. minor stops of about two minutes to go into bios and chance CO.

Does P95 ever end? If not, do you guys think 30m in one run of P95 + OCCT is enough to call the system stable?

[dull ginkgo]

P95 stress test is endless

P95 is more for stressing power/heat than general stability so it shouldn't take more than like 10 minutes on its own to test for that

[steady lance]

Oh okay.

For stabillity what do you recommend? also think my current combination is good enough or?

[dull ginkgo]

I feel like running occt variable load on its own is better for general

[steady lance]

I run CoreCycler overnight as well.

[steady lance]

what presets?

also how long.

[dull ginkgo]

I haven't messed with occt in a while, you'll have to find out what works best for you

Half an hour to an hour should generally be fine

[steady lance]

aight.

Well I just got back and no errors.

2hrs of P95 and 1hr of OCCT

temps peaked at 64.6c

All core freq:
V-core 4.6 average
P-Core 4.7 average

[modern walrus]

tbh I just set them as high as I can cool them

& realistically, TDP and EDC/TDC are the two ways to control the same thing

like you won't hit your TDP limit if your EDC/TDC limits are too low to let it reach TDP limit

you won't reach EDC/TDC limits if TDP is too low to reach EDC/TDC limits

just don't set scalar to anything but auto or 1x

I mean, I honestly have no idea why they leave scalar in BIOS or what's even so scary about it but supposedly it exceeds AMD's safe whatevers

I tried scalar back on 5900x & didn't see any meaningful differences but scalar bad

[sudden torrent]

It affects the boost algorithm. Basically it tells the CPU to disregard its own health in favor of boosting slightly longer.

[steady lance]

I am guessing it scales accordingly?

Does this by any chance also increase max clock?

[sudden torrent]

It's unnoticeable usually. We're talking microseconds more boosting in exchange for the CPU no longer caring if it lives or dies.

It does not affect clocks, only the boost timer

[steady lance]

Ahh gotcha

Guessing it's only ever used for bench scores?

[sudden torrent]

It's rarely used even for that

[modern walrus]

lol sounds so worth it

[steady lance]

What even is the life expectancy of a CPU from AMD?

[faint tangle]

my 3600 is fine from i think end of 2019

[sudden torrent]

The warranty is 3 years but if you treat it right it can last 10

[modern walrus]

heck I can't believe I'm stuck using 8 6+2s to power my GPU lol

I think I might cut out part of the venting in the PSU shroud instead of trying to run it thru the typical channels

[sudden torrent]

Am I reading that right? You need 8 connectors for your GPU?

[sudden torrent]

Is that the dual 12VHPWR 4090 that you have a double octopus on

jfc it is

[modern walrus]

yes

usually vertical GPU is to hide where I messed up bending PETG pipe

this time it's to hide my GPU's dreadlocks

[sudden torrent]

I think maybe a PSU with dual 12VHPWR native would be a better move

[modern walrus]

ya but also a minimum of $250 move lol

It's frustrating cuz I'm tired of throwing money at PSUs

the cheapest dual 12vhpwr is 1650w Thermaltake GF3

I'm gonna end up with an extra Thermaltake Toughpower RGB 1200w that's Platinum rated

if I buy another one I mean

was gonna make the 1200w PSU my benching PSU cuz I gotta use the 1600w EVGA G+

the 1200w TT PSU only has 6 CPU/PCIe connections

[sudden torrent]

Revo DF X has one native 12VHPWR and one dual 8 to 12VHPWR

https://www.newegg.com/enermax-revolution-d-f-x-ert1050ewt-1050-w/p/N82E16817194151

[modern walrus]

damn that's like half the price

I kinda didn't like my 850w Rev DF cuz it's multi-rail

it would shut off when TimeSpy Extreme started lol

there are so few 2x12vhpwr PSUs

the cheapest I found was a be quiet

the Seasonic is crazy expensive like the ASUS

[sudden torrent]

Straight Power 12? (that sounds really bad out of context)

[modern walrus]

lol ya that sounds right

[sudden torrent]

$200 for that rn

[modern walrus]

this I think https://www.amazon.com/quiet-Straight-12-1500w-connectors-BN518/dp/B0CC8JMDVH/ref=sr_1_10?

the 1300w has 2x for $200

[sudden torrent]

https://www.amazon.com/dp/B0CC88BCG9?tag=pcpapi-20&linkCode=ogi&th=1

[modern walrus]

yes this one 1200w not 1300w yes

[steady lance]

What model 4090 is this?!

[modern walrus]

Hall of Fame

[steady lance]

Also the cables, oh the cables. It looks like my fan cables.

Massive chain of dreadlocks.

[modern walrus]

it says recommended PSU at least 1000w. That would be fine with the 1200w TT PSU if I had enough connections lol

I thought about using some of the 6+2s that have 2 connections per cable but ehhh

it's not gonna run that high for daily use

I got like 784w out of it before tho

[steady lance]

Why not just buy a 1600w PSU with more io then you'll likely ever use.

[modern walrus]

I have one but it doesn't have 12vhpwr lol

[steady lance]

Oh rip.

Is that why you have the giant mess of cables?

[sudden torrent]

Yeah that's why he needs 8x PCIe connectors lol

[modern walrus]

the back of the thing is ridiculous

[sudden torrent]

Freaking nvidia ruining everyone's day/week/year by obsoleting perfectly good standards and melting GPUs

[modern walrus]

lol why are they like this?

I guess by 50 or 60 series 12vhwpr will probably be what 6+2 used to

[steady lance]

What was the point of the cable?

[modern walrus]

6080ti needs 3x12vhpwr lol

[sudden torrent]

I love how AMD's flex on Nvidia was "we have normal 8 pin power and don't melt"

[steady lance]

Was there any technical or theoretical benefit?

[sudden torrent]

Theoretically 600W over a single cable

[modern walrus]

ya but then nvidia is like ya uh come see me in benchmarks bro

[sudden torrent]

But way too small of a margin for error and poor connector design

[modern walrus]

https://www.youtube.com/watch?v=otCpCn0l4Wo

[steady lance]

Only benefit I've seen is that the cable assuming you aren't using chains looks far cleaner.

[sudden torrent]

XTX vs 4090
XTX gets 140 fps
4090 scores 0 because it started smoking

[modern walrus]

lol with gigabyte PSU or cablemod 90 adapters ya

[steady lance]

Well hopefully by next Gen issues will be resolved so we can have a functional cable that is not prone to issues.

[modern walrus]

lol or it'll just get worse & more expensive like usual

[steady lance]

The expensive part is likely gonna happen.

[sudden torrent]

We do have a new cable design that splits the 12VHPWR in half and it mostly resolves the issues with friction on too many pins that kept people from plugging it in

[steady lance]

Not sure what the odds are on the cable.

[modern walrus]

ya they extended the sense pin or something

[sudden torrent]

Shortened

[modern walrus]

lol that'd make more sense wouldn't it

[steady lance]

Btw so was the problem this whole time just people not fully plugging the cable in?

Or was there more to the story.

[sudden torrent]

It was poor design physically preventing people from plugging it in all the way in some cases

Which lead to poor contact on some of the pins

[steady lance]

Well that's definitely an issue.

[modern walrus]

it was arcing ig

[sudden torrent]

Sense pins said "you're good to push 600W" but only 2 of the 6 12V pins were actually connected

[steady lance]

Arcing is when electricity zaps to another source without direct contact right?

[modern walrus]

Core 4 does -60 for me

yes like it jumps thru the air

& making lots of heat

[steady lance]

Damn.

7950x3D?

[modern walrus]

7900x3d

[steady lance]

Ahh my bad.

[modern walrus]

suuuper arc https://www.youtube.com/watch?v=9GeXkussHfw

[steady lance]

Currently I save run into two problem children core wise.

One refuses to do -10 and one doesn't like -20. Fixed child one. Gotta turn down -co on corn nine.

[modern walrus]

Core 9 won't do better than -24

Core 5 will do -55 but borks out at -60

[steady lance]

My plan right now is to get a general idea of what each core likes by working in curves of 5 then once I have done that tune each core's co one by one in 1-2mv steps.

[modern walrus]

so trying -58 rn

[steady lance]

What frequency have you managed to get on your p-cores?

Cause I am guessing your v-cores are like mine and refuse to go beyond stock limit.

[modern walrus]

what is the stock limit? lol

not sure I'll check now

I wanna say like 57xx so far

[steady lance]

For 7950x3D it is 5.25Ghz (based on my two different 7950x3D)

[modern walrus]

gonna delid at some point

oh on vcache CCD?

[steady lance]

For 7800x3D it peaks at 5ghz.

[steady lance]

V-cache is always CCD0 afaik.

And yes.

CCD1 is the high frequency cores with the 5.7ghz marketed speed.

[sudden torrent]

I don't bother with anything less than a 5mV step personally

[modern walrus]

Core 3 is running 5156.7 with Prime95 HUGE

[steady lance]

All core?

Is that sustained or peak out of curiosity.

[modern walrus]

oh no this is just single core

trying to get COs

[steady lance]

Okay.

Gotcha, not bad then.

[modern walrus]

I haven't run all-core

probably should but it'll probably be spicy lol

[sudden torrent]

Right now I'm per CCD CO

I kinda want to see what it can do though

[modern walrus]

I should've just done that but I already committed lol

[sudden torrent]

12 cores CO is going to be straight up pain

But I do know that core 7 and 8 can't do -35 at least, they crashed corecycler

[modern walrus]

lol 16 must be nuts

[sudden torrent]

I think I'll leave it for now and do per core when I'm really, really bored

[steady lance]

be prepared, for me All-core was far more prone to throw errors then single core.

[modern walrus]

heck I guess core cycler is useless then

[steady lance]

What I found is do this.

Run it in 15m per core minimum.

[modern walrus]

were you running Prime95 in Core Cycler?

[steady lance]

Yes.

on the largest setting

[modern walrus]

I'm pretty much at stonk settings on core cycler. All I've changed is the order

[sudden torrent]

12V-2x6 is the new cable

[steady lance]

the extended duration makes up for the all-core test.

15m minimum per core, I recommend 20-30m if you can tolerate it.

[modern walrus]

heck I'm gonna have like 7 PSUs

[sudden torrent]

15m gets you bench stable at least

[steady lance]

then afterwards run 30m of P95 with OCCT just incase.

[sudden torrent]

Nah it's the same socket, just a new cable

[modern walrus]

ah okay that's better ig lol

well to heck with this CO thing

I'm gonna delid CPU

then test it before I put everything in the case

[sudden torrent]

Be sure to use sudden jerking motion when you delid it so you can crack the solder and the die

[modern walrus]

lol I'm using a vise & small chisel

jk

[steady lance]

now testing -25 co on all except core 5 and core 9.

[modern walrus]

good luck lol

[steady lance]

Thanks.

Why did you all of a sudden give up on CO?

by the way, my motherboard lets me set Chipset Core Voltage is that the same as vcore?

[sudden torrent]

No, it's the chipset

On the motherboard

[steady lance]

ahh rip

[steady lance]

Is there a way to install a ECLK onto a board that doesn't have one?

[sudden torrent]

Sort of

You need a board trace schematic and really good soldering skills

[steady lance]

oh nvm then

[steady lance]

all cores passed.

15m per core in core cycler

core 8 failed on Iteration 2.

Turned CO down on that.

going to let core-cycler run over night to make sure the -25 co is stable on the other 13 cores.

[steady lance]

I am trying a new stress test combo.

Running Prime95 (ffts L1-3), OCCT (AVX2, Large, Extreme, Variable), Aida64 (System Stabillity Test) all at the same time.

[steady lance]

think 20-30m of this is good for testing all-core?

just by adding aida64 max temps went up by about 10c.

[steady lance]

Now doing -35.

[modern walrus]

what are your temps for CPU & CPU Die (Average) at idle?

on hwinfo

my bad for ping

[modern walrus]

this doesn't seem to be helping lol

[sudden torrent]

My cooler is at minimum speed

[modern walrus]

heck I'm around 50-51C for those two oddly

my CPU IOD hotspot is spicy

maybe I should enable PBO just to see what happens

[sudden torrent]

Remember I'm on a LF2 420

[modern walrus]

I'm on 420+360 open air bench loop rn

with delid

direct die

[sudden torrent]

Your peak temps are better than mine

Offset mount?

[modern walrus]

on CPU/CPU Die (average)?

no I should probably look into that tho lol

I had to rig it up myself

[sudden torrent]

Offset mount alone can account for 4 degrees

[modern walrus]

not a block designed for direct die lol

[modern walrus]

do your CPU/CPU Die (average) peak that high? I stupidly didn't run R23 or anything else before I delidded

so I have no idea whether this is better or worse lmao

[sudden torrent]

I'll run r23 for 15 minutes

I am undervolted

[modern walrus]

I only ran a single render

so 1 is fine haha

[steady lance]

also only using 6 cores right?

[sudden torrent]

No? I haven't needed to do that yet, process lasso has been good for older, non-3D aware games and programs.

[steady lance]

Ahh thought you disabled the other cores because of latency.

[sudden torrent]

Oh well I peaked at 76-77

[modern walrus]

oh so maybe I'm not having contact issues then

[steady lance]

Currently my peak temps are 74 after days of core cycler + a couple hours of Aida64, P95, and OCCT at the same time.

[modern walrus]

wait jk

[sudden torrent]

I did get some major stuttering in certain games but I assign those to cores 0-5 and it goes away

[modern walrus]

just opened hwbot x265 & hit 84.2 on both values

[sudden torrent]

That's a harder test

[modern walrus]

damn & that was PPT only hitting 88.150w

so garbage

[modern walrus]

I don't think they run as hard if you run them all together cuz of scheduling or something I forget lol

[steady lance]

For all core if my CPU can survive P95 (ffts L1-3), OCCT (large, variable, extreme), and Aida64 (stress test CPU/Cache/RAM/something) at the same time is it safe to say it's probably very stable?

[steady lance]

temps shot up 10-12c within less then a minute.

and system felt a lot slower.

[sudden torrent]

Typically if you combine tests you do tests for different parts, like one cpu test, one ram test, and a gpu test

[steady lance]

So what would be a good combination?

P95, OCCT, and TM5?

[sudden torrent]

p95, tm5, and occt vram would be good. You'll see your max heat values for just about everything.

[steady lance]

aight.

1hr or 30m enough to check for stabillity?

[modern walrus]

OCCT power by itself tries to test PSU lmao

[sudden torrent]

depends on how stable you want it to be

[steady lance]

Very stable.

[sudden torrent]

Bench stable, 15 minutes.
Game stable, a couple hours.
Rock stable, 48 hours.

[steady lance]

rendering/3d design stable.

[sudden torrent]

A couple hours should be fine

[steady lance]

any concerns regarding CPU life with a two day test?

[sudden torrent]

no

As long as you don't mess with scalar

[steady lance]

okay, wanted to check cause of some reddits.

[sudden torrent]

They're designed to run at max temp 24/7

[steady lance]

Understood.

[sudden torrent]

Damage doesn't occur until around 130C

[steady lance]

I think once I've found the rough limits of every core I will try running a 12hr (overnight) stabillity test with P95, TM5, and Aida64 (because OCCT only lasts a hr)

or is there something else to try vram besides occt?

[sudden torrent]

You could do a full gpu test instead

[steady lance]

recommendations?

btw TM5 PCBdestroyer is the profile to use yes?

[sudden torrent]

If you want to heat it up yes

[steady lance]

Heat isn't really a concern more so stabillity.

(ran PCBdestroyer on this kit of ram nearly a hundred times by this point)

at least I don't think it is

[steady lance]

Okay, I am changing up the strategy. this is taking forever.

Do CO on one-two cores at a time at most and offsets of 10-15 till erroring then dropping by 5.

also set CoreCycler to 30m per core.

so far working much faster.

[steady lance]

disabling iGPU dropped my load temps by around 5c.

Also, I have yet to confirm this as this is a different test combo I am running right now but it appears the core frequencies are boosting higher and for longer.

Core 0 (v-cache) is now able to maintain 5Ghz under single core loads.

[steady lance]

cleaned out my filters. drops temps by 2-3c.

[steady lance]

After poking around it seems it is impossible to overclock CCD0 (v-cache) to boost above 5.25GHz using PBO. apparently it was by design according to a post I saw.

With the change in approach I am resetting my CO values. currently at -30 on Core 0 and trying -15 on Core 8.

[steady lance]

So far -40 CO on core 1 is doing pretty good.

[steady lance]

• Core 0 = 35
• Core 1 = 35
• Core 8 = 20

[steady lance]

Well f.

I think I may have corrupted my windows install.

THANK YOU SYSTEM RESTORE!

note to self, DO NOT ATTEMPT -40 on core 9.

[sudden torrent]

Ah yeah I borked my install with -50/-40 too

[steady lance]

Core 9 definitely couldn't hanlde -40. It couldn't even run -25.

well it was bench stable at 25. just cranked it down to 20

[steady lance]

makes me wonder how good of a sample do you need for -40

[sudden torrent]

I have -40 on CCD0

All 6 cores

CCD1 at -30

[steady lance]

wow.

You seem to have a really good sample then.

all the reddits I've seen cap out at -35 at best. with -15 to -25 being average.

Currently I am at:

• Core 0 (-35)
• Core 1 (-35)
• Core 8 (-20)
• Core 9 (-20)

doing the cores in alternating order to match Core Cycler.

[steady lance]

are you doing all core or per core CO?

[sudden torrent]

Per CCD

Cores 0-5 are my 3D cores at -40

[steady lance]

wow.

[steady lance]

rip. core 8 threw an error after two hours of stressing

[steady lance]

Decided I wanted more stabillity so I upped CoreCyler runtime to 1hr per core. Ran CoreCycler overnight and the four cores from yesterday passed two iterations at an hour each.

Do you guys have any ideas on how I can get my V-cache cores to boost beyond 5.25GHz?

[sudden torrent]

They're probably hitting the voltage limit so chances are they won't

[steady lance]

-40 on Core 2 passed 1hr of Core Cycler.

going to try -45.

[steady lance]

Decided to try -50 instead.

also wow, there does not appear to be any negative CO limits on Zen4?

welp -50 didn't work.

trying -45

[sudden torrent]

It used to have a -30 limit but AMD noticed that some chips benefitted from lower, 3D CCDs especially

I think the limit is -60 now

But I've not seen anyone do less than -50 stable

[steady lance]

mine went up to -200. I didin't try further.

or down to I suppose is more accurate

[sudden torrent]

Just because it lets you enter the value doesn't mean it actually applies it

[steady lance]

🤞 Core 10 will settle for -25

[steady lance]

so far so good

[steady lance]

trying -35 on core 3.

[steady lance]

-30 seems stable on Core 3. -35 was stable until the final like 2m.

[steady lance]

-25 on Core 11.

[frigid locust]

Wow you're REALLY going through a lot of hoops to get the most performance out of that chip

[steady lance]

Honestly at the end of the day performance gains are only about 5% but the reason I do it is because I push pushing the limits of computers.

Gonna do my ram next then maybe give my god forsaken 3070 one last chance at an OC.

[frigid locust]

Maybe you should instead let that poor 2070 retire in peace and bring in a 4090

[steady lance]

Was the original plan. Going to buy myself a new monitor first then upgrade GPU to nvidia's next flagship.

My current monitor does not provide enough screen real estate for my work as a developer so I am looking into either a WQHD or UWQHD monitor.

[steady lance]

Appears one of yesterday's four cores failed overnight.

going to try and see if switching to AVX512 will catch the error / strain the cpu more.

[frigid locust]

Oh you're going for overnight stress test-stable?

[steady lance]

yes.

also the issue almost 100% was core 11.

[frigid locust]

That's the coolest kind of stable overclock

But it takes so long to achieve

I only run a quick and dirty -20 offset in CO on all 6 cores

Never had an issue

I can probably try -25

but like

-20 already gives higher clocks at the same power use

so why bother?

[steady lance]

True.

[frigid locust]

pretty sure it's already slightly faster than default anyway

[steady lance]

as said, gains aren't that large. I do it more for fun personally.

[frigid locust]

it's a "Eh, I can so why not?" thing for me, but I'm not like super into optimizing every minute detail

there is no way you can get me to spend days dialing in individual cores

[steady lance]

about 5% from CPU, and if I can do same or better ram OC as on my former 7950x3D (R.I.P) I should squeeze out another 5% or so.

[steady lance]

xD, is that in reference to me?

[frigid locust]

Not particularly, but technically yes

Like I'm not trying to be mean to you

[steady lance]

No worries.

[frigid locust]

But it just doesn't make sense to me to spend days for 5% perf at most

15, 20 or 25%?
Now that is more what I'd consider worth it

[steady lance]

I wish there was a testbench I can run that really excels in making cores boost to high frequencies like when someone opens up a new chrome tab or opens some files.

[frigid locust]

5% you can get with a dirty undervolt

just from the VF curve being pushed back basically

[steady lance]

I've seen 5-6 of my P-cores boost past 5.8ghz and one of them boost up to 5.95Ghz.

[steady lance]

true, though all-core work benefits more from what I've seen.

sustained clocks have gone up by +100mhz up to +200mhz on low load all-core.

[frigid locust]

to be completely fair, I intend to eventually build a 7950X editing PC (once I have the means to properly start streaming/content creating)

That one I might actually take the effort to like dial in every core

But hey, if I ever actually get myself to bother I'll have someone experienced to ask for cues

so I think it's still a win

[steady lance]

aight.

[frigid locust]

That "someone experienced" is you obviously

[steady lance]

Also do not go 7950x3D unless you are prepared to likely deal with some weird setups. Otherwise it's a great cpu.

[frigid locust]

Yeah honestly I prefer the 7800X3D in one rig and 7950X in another setup

[steady lance]

No I am not talking about the scheduler just an fyi.

[frigid locust]

7950X3D is really just a neat "Woah we can now do this too" chip to me

[steady lance]

Yeah, it's a niche chip. Fantastic to those who have a use for it otherwise just save yourself the likely hassle and get AMD's normal zen4 chips.

[frigid locust]

I'm on the opinion that 2 separate rigs would be better optimized than one with a hybrid chip

I mean let's be real if you can afford a 7950X3D you can afford to have 2 PCs

And you're probably driving a high-end sedan because you're like a top earner

ANYWAY

if I ever need OC/undervolt advice I will seek you out @steady lance

[steady lance]

aight.

[steady lance]

yeah true.

[frigid locust]

Speaking from experience? :3

[steady lance]

Nah. I prefer my old Durango.

[frigid locust]

Yeah okay you ARE a top earner then

Sigma male with the old pickup

[steady lance]

Durangos aren't that expensive were they? got one used and has served me quite well.

[frigid locust]

Ones before 2005 are dirt cheap by now, later ones get more and more expensive as they get younger

though admittedly I don't like Durangos

it's one of those half-truck/half-offroader that is exclusively used on asphalt kinda vehicle that just rubs me the wrong way

[steady lance]

Ahh I got one from 2000, it was the 5.8 Magnum one from a dealer for dirt cheap because it had been in a accident from the prior owner. they restored it and looks fine now, works far better then it looks.

[frigid locust]

at least that has a V8

IDK it just looks like a weird fish to me

[steady lance]

Yeah it was odd getting used to it at first.

Once I got used to it I actually started liking it. Has served my family well. Been driven from California down to Cabo San Luca (MX) up to Cali then up into Canada then Alaska.

[frigid locust]

Yeah reliability is big

to be fair

I can't really chastise you for it since my dream car is a stupid powerful modified R35

with carbon fiber GT3 body panels

[steady lance]

Yeah, we all have some kind of dream goal that's way overkill for all of us haha.

[frigid locust]

I can DM you what to expect

if you don't mind talking cars there

[steady lance]

We want it but definitely don't need it, xD.

Sure, though just a heads up in that I gotta work in a short bit.

[frigid locust]

Take your time

[steady lance]

switched Core Cycler to use AVX512 instead of SSE. Hitting the same boost clocks while stressing the cpu a lot more it seems.

[steady lance]

Prime95 set to AVX512 is definitely better then SSE for testing in CoreCycler.

Hitting very similar/same boost clocks but it's detecting errors for what managed to pass multiple 1hr runs of P95 SSE.

[steady lance]

experiment was a hit or miss.

Saw clocks bounce all over the place but did not go higher then it would otherwise hit in AVX512

[steady lance]

R.I.P seems core 2 wasn't 100% stable at -40. bench stable at -45, game stable at -40 however not error free.

trying -35 on it.

Yikes, seems I have taken over the channel. Just looked at the chat history.

[sudden torrent]

Not much goes on here between hardware releases anyway

[steady lance]

Should hire me to keep channels alive, xD.

[steady lance]

this is annoying.

1/8 cores I've undervolted so far keeps being a problem child and crashing while I am asleep but when I go look at logs in Core Cycler they are useless.

[sudden torrent]

Do you still have it set to 2 core mode? If so turn that off and instead do 2 thread mode so it uses both logical threads on a single core. That way you can have it be the heavier load you want while also still being able to tell which core is the problem.

[steady lance]

gotcha. so enable both threads

now set to use both threads, 30m (from 1h), and P95 in AVX512

[steady lance]

@sudden torrent, I realized the issue. 🤦‍♂️

The reason the logs didn't show an error was because another power outage had happened. This whole week my area has been experiencing power outages that only last a few seconds at best but would trigger my pc to shut down.

So my best bet is to just test for stabillity whenever I am at the pc rather then relying on overnight tests.

[sudden torrent]

Ah, that could be an issue indeed.

[steady lance]

yeah it's been very frustrating. during the day when it happens even for like a split second, I can see my overhead light flicker so it's been my way of figuring out if my PC crashed from instability or power outage.

going to put the 8 cores (4V, 4P) through 30m runs of YCRUNCHER mode in CoreCycler using 22-ZN4 ~ Kizuna.

double thread

Found a problem child.

Core 0 which was at -35.

[sudden torrent]

Alright you've inspired me to do per core CO.
I've got corecycler set to run through 60 different FFT sizes and do a full cycle of each in AVX-512 mode, single core single thread. I'm adjusting 2 cores at a time and testing just those cores, plus any that errored in previous tests after increasing my offset.

[steady lance]

😄

60 FFT sizes? is that P95 mode in Huge setting?

[sudden torrent]

It's the runtime setting that sets that. If you tell it "auto" then it does a full loop of every FFT size

[steady lance]

Could you send me a screenshot?

[sudden torrent]

I'll do you one better

[steady lance]

thanks!

ahh gotcha.

So this way time won't be a concern. just how fast the cpu can do things?

[sudden torrent]

It takes about 9 hours for me to run all cores with those settings

[steady lance]

Gotcha.

I am not to worried about time since I am running CoreCycler 24/7 xD

[sudden torrent]

Given you have 16 cores and I have 12 I'd guess it would take 12 hours for you to run all cores

[steady lance]

which is better? All or Huge for fft size.

[steady lance]

probably.

[sudden torrent]

Each will detect different kinds of errors

[steady lance]

Though I am only testing 8 right now.

I guess I will try all since all my previous runs were at Huge.

@sudden torrent, why am I seeing so much more information when starting CoreCycler?

[sudden torrent]

Setting it to auto? It records the FFT sizes that have run previously

[steady lance]

oh

on huge i only saw those in log files

not huge

i mean 1hr

[sudden torrent]

It also has limits on number of ticks it can run, but I've effectively got that set to infinite

[steady lance]

gotcha. ty fpr the config

[sudden torrent]

If you use the config as is then make sure to set coreTestOrder to random or alternate if you want to test all cores

[steady lance]

i configured it to a alternating order on the cores i undervolted

By the way, max temp after a week of core cycler is 72c.

average is 60c.

Nvm, 55c thought it was higher

[steady lance]

Dear g0d. @sudden torrent, I think it's only half finished with one core.....

[sudden torrent]

Here's where you can see

Yours was on 116 of 180, I guess it does more for higher core counts

[steady lance]

I had changed mine from Huge to ALL which was described as a larger set.

Lower minimum, same maximum

Ohhh okay. looked at a few and at size ALL there are 180 FFT sizes.

I am at 142 currently for core 8.

[sudden torrent]

So far I have 1 core that does -50 and 3 that do -45 on CCD0

[steady lance]

dang

changed the log settings because on ALL fft sizes it was so large that if a core errored you can't scroll up high enough to see which did it.

[steady lance]

far easier to read the terminal now for sure.

Just a simple, core pass or fail

roughly 2h30m per core on ALL setting.

[steady lance]

All 8 undervolted cores passed the test

[steady lance]

gonna start testing the other 8 cores today

[steady lance]

started later then expected.

testing -40 on Core 4.

[steady lance]

attempting -30 now

[steady lance]

-25 is definitely stable, ran the test overnight and it passed all 4 iterations.

[steady lance]

-35 on Core 5.

[steady lance]

Fal, do you think it's safe to call a OC stable if each core passes at least 2 runs back to back of Core Cycler in my current config? (Your config except in ALL setting)

it takes about 2h30m-3h per run.

[steady lance]

testing -40 on Core 7 now.

-30 on core 6 stable

[steady lance]

Is Aida64 SHA3 a reliable stress test?

[steady lance]

tried SHA3 crashed my system, luckily I figured out which core's CO was to aggressive.

it was core 7 at -35. trying -30.

[steady lance]

Core 7 couldn't even pass Aida64 SHA3 with -5 CO but everything else passed. Moved onto Core 12 which passed and booted up till -30 (no-post at -35).

Going to run Core Cycler on it overnight.

[steady lance]

3/8 cores on CCD1 are doing 5.925-5.95Ghz and 3 others doing +5.825Ghz

Combination SHA3 first then Core Cycler seems best for checking errors.

SHA3 seems to be very sensitive to negative CO values when boosting to high numbers and Core Cycler seems best for testing endurance.

[steady lance]

4/8 cores on CCD1 hit +5.9Ghz!

[steady lance]

5/8 cores hit +5.9Ghz.

[steady lance]

going to start doing some OCing again.

6/8

[proven canopy]

That's old vs new ycruncher though

[sudden torrent]

True but still a funny anecdote

I could probably beat it now, I'm almost done with per core CO

I also got my memory up to 6200 and tighter timings

[proven canopy]

If curious, you could run the old yc from benchmate

[modern walrus]

ever since I put the board/cpu/etc in the case, 6400 won't get thru VT3 so I guess I'm going with this

still gonna try to up FCLK tho

62.4ns AIDA at 2k...

61.0 at 2067

[modern walrus]

wait... maybe it has something to do with never downloading AMD igpu driver?

[steady lance]

Any luck?

Currently I have paused my OC until I figure out the cause for Warframe crashing.

[modern walrus]

My best guess: corecycler is useless. I passed hours & hours of corecycler only to fail R23 on the first run. Maybe if it tested multiple cores at once it would be helpful, but I think if you do that it can't tell you which core failed.

Also what's strange is I sometimes get "IRQ_NOT_LESS..." BSODs when core offsets are too ambitious. That makes very little sense to me since that's a memory-related BSOD & I'm like 10 hours VT3 certified lol

[modern walrus]

Presently I'm +200 with magnitudes between 25-35. I'm gonna roll with that to see if it borks out then reinstall

I thought it was good to go before when I was testing on a throwaway win11 stock install

I had corecycler running every fft iteration on SSE per core then same with AVX512

Here I was thinking only Intel is unstable lol I blame ddr5

And skill issue(s)

[dull ginkgo]

maybe something to do with power draw or temp affecting IO die stuff too?

[modern walrus]

Temps r so strange on am5

Like I'll have max core temp in the 60s only to have tdie or whatever at 85C

I disabled igpu but not much impact there

I still can't figure out why 6400 passed VT3 before the motherboard was installed in the case but not after

[steady lance]

@modern walrus, if you are testing for CPU CO and Override stability I've found a combination of Aida64 SHA3 (10 runs) a good way to get a idea for the limit of the CO offset then proceed to run CoreCycler (P95, AVX512, ALL FFT sizes, 4 iterations per core) to be a good combination.

[steady lance]

Agreed.

[modern walrus]

no ycruncher? lol

[mental turret]

ycruncher >

[steady lance]

No, is Y-Cruncher better for stability testing?

[modern walrus]

it might be cuz ycruncher is realllly stressful

but I would've said the same about Prime95

[steady lance]

I mean I can try again later, though the CO seems to of been very stable. No crashes or system stutters.

(aprox 2hrs per iteration, ran 4 times per core)

[modern walrus]

oh well good job then

I've had issues lol

[sudden torrent]

I passed ycruncher, posted a few results on hwbot yesterday. It's just you and floppa above me, and if I'm not mistaken you're delidded.

[modern walrus]

Yes. Yes I am lol

[steady lance]

Has delidding offered you any sort of benefit? So far it seems like it's been downs only.

@modern walrus

[sudden torrent]

From what I've seen you get all the same benefits with lapping the IHS and washer modding. That's probably what I'll end up doing.

A lot less risky than delid

[short blade]

ive gotten great results from delidding older intel chips but nothing modern

i7-8700k was the newest

[mental turret]

delid all ur soldered cpus

like this

[steady lance]

why?

[sudden torrent]

The solder is pretty poor at conducting heat; swapping to liquid metal gets excellent results

If you're feeling spicy you can swap to an all copper IHS

[steady lance]

Actually, on that topic. I was curious so I looked it up and apparently Silver is more thermally conductive then copper by like 9% so I was wondering if silver heatsinks are a thing?

[sudden torrent]

Not that I've seen. I think the reason for that is how reactive it is with other metals and how quickly it oxidizes.

Slight correction, silver oxidizes a little slower than copper but silver tarnish is not a good conductor, so it would lose effectiveness without regular polishing over time.

[steady lance]

So it's a high maintenance.

What causes the silver to tarnish? I had silver jewelry before and they'd tarnish even without touching water very fast.

[sudden torrent]

General recommendation I see is to polish it every 3 months, so about as often as you should flush a custom liquid loop

[sudden torrent]

Chemicals in the air bond to the silver molecules

You end up with silver sulfide iirc

[steady lance]

Gotcha. So it works better but don't use it unless you are prepared for maintenace every 3 or so months. Is that the gist?

[mental turret]

no it was a joke

that solder is fine

[humble jay]

the ihs already is all copper tho

[sudden torrent]

I mean without the nickel plating

[humble jay]

to me it seems like the only way delid is worth it is with the special direct die waterblocks, someone sent me a delidded 14900KS and it is completely uncoolable with floating custom IHS + liquid metal

[dull ginkgo]

At that point I wonder if it'd be easier to mod the socket to seat the CPU higher

[steady lance]

May I ask what the purpose of the nickel plating is? I assume it's to do with preventing tarnishing/oxidizing of the copper?

[modern walrus]

eh lmk what benchmark or stress test to run & I'll send hwinfo temps. It's strange because core temps are very good but other parts of CPU hit 85C but I think it's designed that way? Also, I kinda gave up on core magnitude offsets for now but might be some room left for improvement there eventually

[modern walrus]

nickel plating helps with liquid metal a lot. Apparently liquid metal + copper forms an alloy so you have to reapply the liquid metal cuz it pretty much ends up absorbed into the copper (I'm sure there's a much better scientific explanation but practically speaking)

[steady lance]

gotcha. what CO did you achieve?

[modern walrus]

lol I think I left it on -25 for everything

like to heck with it

I don't feel like reinstalling windows again for now

[steady lance]

I completely forgot about this. Not sure on the exact process but I remember hearing about it from someone who had used LM on copper directly.

[modern walrus]

eventually I'll connect a SATA drive or something to put a test windows on & bork that one

[steady lance]

His solution was to keep applying LM overtime till the copper couldn't absorb anymore.

[modern walrus]

ya that's how my son's 10900K & LF2 are now lol

well I hope so

I've repplied like 4x in the last 2 years or so

every now & then temps will just suck

then I'll reapply & they'll max out at like 75C lol

[steady lance]

Another option is what I did. Save a backup of your windows on a secondary drive so if you mess up your install, you can use the backup from the second drive in the system.

[modern walrus]

iirc it's OCed at 51x all-core & 47x ring

ring sucks on that CPU

[steady lance]

ring?

[modern walrus]

ya I should make a macrium image or something ig

[modern walrus]

Intel thing lol

[steady lance]

Never had a overclockable Intel cpu so not sure what that's like.

[modern walrus]

can be fun

[steady lance]

I should probably finish my OC now that I've found the cause for Warframe crashing.

[modern walrus]

my six-year-old's 10900K memory settings (DDR4) lol

[steady lance]

It was a game setting (it was trying to override GPU performance preference)

[modern walrus]

dude I keep having this issue with both Fallout & CyberPunk with this nvidia file

[steady lance]

Ouch. Luckily for me, Wolfgang pointed me to the launcher settings which made it a easy toggle.

In anycase, I've finished CO (in increments of 5) on 15/16 cores.

Unfortunately one of my cores refused to pass SHA3 at even -5.

[modern walrus]

what does it say in event viewer after crash??

[steady lance]

Only the game crashes. Not system.

[modern walrus]

that's what was happening with me in CyberPunk & Fallout

also btw

[steady lance]

Yikes. sounds rough. In my case it was the frustration of during 2-3hr survival runs for loot then randomly crashing and losing it all.

[modern walrus]

do you have 4 NVMe slots?

[steady lance]

Yeah, it was one of my requirements in picking this board. MSI X670-P.

Why do you ask?

[modern walrus]

do you ever lose an NVMe?

like it doesn't show a drive?

btw the .sys problem

you don't see any of these in Event Viewer?

it's .sys not .dll my bad

[steady lance]

In bios or?

[modern walrus]

both actually

but after I turn off PC then turn off PSU then restart it'll come back

apparently it's a problem on a few Asus & Aorus boards too from my brief investigation lol

[steady lance]

There was a time that I couldn't find my drive in Bios but I haven't ran into that issue in windows.

[modern walrus]

there's a new beta BIOS for my motherboard but I don't wanna redo the OC crap lol

[steady lance]

The problem I think was due to the Windows Installer being on a different drive from the drive with windows

which I solved by doing a clean windows install with just one drive in the system

[modern walrus]

installer? you install windows from the drive you're installing windows on? lol

I use external always

[steady lance]

I use a USB stick.

then install windows onto the drive.

[modern walrus]

oh okay that's what I do

well ventoy on a nvme thing but basically same

[steady lance]

I can't for the life of me find the cause for these errors, they have been around forever. not causing me any harm though

[modern walrus]

heck why do I have so many of these?

rhetorical question lol

[steady lance]

You as well huh.

[modern walrus]

it says metadata or something

sometimes Event Viewer is super useful

other times it's like uh what?

[steady lance]

Agreed

[modern walrus]

the EK direct die is kinda decent once you lap all the nickel off & get it flat--finally cools better than Heatkiller IV Pro with IHS on lol

[modern walrus]

temps after a couple hours of MC lol

[steady lance]

Minecraft?

[obsidian cobalt]

Had anyone here did vbios flash on a Gigabyte aero 16 up to 130w?

[late kraken]

Flashing is although risky,
I don't think anyone would have done here.

[alpine mantle]

Flashing vbios on laptop, risky as heck

[sudden torrent]

Yeah. On desktop you can just plug in a backup GPU to recover a bad flash. No such luck for laptops.

[alpine mantle]

^^^

[obsidian cobalt]

But had anyone tried on that laptop

[celest notch]

https://www.reddit.com/r/GamingLaptops/comments/ww6c0v/gigabyte_aero_16_3080ti_laptop_gpu_flash_vbios/

Google seems to be working

[barren ridge]

On a laptop you're pretty much risking it

Tho somehow I did a vbios flash just fine lol

Asus delivered it on my lappy

But it was an update, if anything

Also I don't recc flashing for higher wattage, as the laptop's cooling system might not be designed for that spec

[sudden torrent]

And you'll probably need a new power brick too

[steady lance]

Working on OCing again.

Just one core left to test for stability, the rest are 2-3 weeks no crashes stable.

[steady lance]

3/4 passes so far.

Gonna try -25 CO once it passes the 4th round.

[steady lance]

What is a good benchmark I can use to check my CPU scores? Also Is there a benchmark to test my CPU's flops?

[sudden torrent]

Cinebench, ycruncher, time spy extreme cpu test.
Flops aren't really something that matters, it's only one type of operation.

If you really want to test flops you can use aida64

[steady lance]

My aida64 trial ran out. Is there another app? Also thanks for the recommendations.

[sudden torrent]

Passmark?

[steady lance]

Will look into it.

[sudden torrent]

https://www.passmark.com/products/performancetest/

[steady lance]

Also @sudden torrent -30co on my final core tested stable.

Haven't tried going higher yet.

Unfortunately I can't use Aida64 SHA3 anymore to test for those micro instabilities.

[sudden torrent]

I got 2 cores down to -50 and 3 to -45. The 3D cores tend to be better as far as I can tell.

-50 seems to be a hard limit

[steady lance]

The -30 CO on core 15 did pass 5 full runs in core cycler though.

[steady lance]

Nice.

Wow.

You tested those cores with SHA3 test?

[sudden torrent]

I don't use aida for that

But it did pass AVX-512 corecycler

[steady lance]

That test seems to be extremely good for finding unstable CO values. Caused my cores to all dip by 10-15.

Aida64 SHA3 I mean.

Not good for finding if the CO value is stable in long durations though like under load ai mean.

[steady lance]

May I ask why? I want to learn as much as I can about overclocking and the whys and why nots of different things for it.

Meant to react to aida64 msg.

[sudden torrent]

Ah

I just don't have a license for it

[steady lance]

Oh we're in the same boat then.

How good is the Aida64 extreme tools? Wondering if it's worth the $60 for a license.

I only tried the SHA3 test and the all core one which raised my temps faster then even p95.

[sudden torrent]

Good if you use it a lot but not worth it for occasional users

[steady lance]

Would you say the tools give more accurate results?

[sudden torrent]

Eh, accuracy isn't a problem in most tools

It's background tasks that interfere

[steady lance]

Hmm gotcha.

NGL part of me wishes I didn't use SHA3 so I could claim to have -35-45 stable on my CPU xD.

Instead I am at around -20-25 average.

[sudden torrent]

How much testing you need is based on how stable you need it to be

[steady lance]

Fair.

[sudden torrent]

I mostly need it for gaming and occasional compiling so I'm happy where it is. It hasn't crashed on me yet.

[steady lance]

I need it to be rock stable since I don't want to randomly lose a large amount of work when I am coding and also I am going to be learning blender again soon and so don't want it crashing during a render.

[steady lance]

I am amazed you managed to get two golden samples back to back as far as I can tell.

[sudden torrent]

That 5800X was insane

I would have loved to get that under LN2

[steady lance]

The bios ruined it right?

[sudden torrent]

Yeah

[steady lance]

Did they ever refund you or smth?

[sudden torrent]

That big "you must update because security flaws blah blah" update somehow made it force 1.4V into it under load

It was just barely out of warranty

[steady lance]

Ouch.

That genuinely is annoying. At least you have an upgrade now.

[sudden torrent]

Yeah lucklily I have a couple connections where I was able to manage that

I get to join the ranks of 7900X3D daily users

All dozen of them

[steady lance]

I wonder which is more common.

7950x3D or 7900x3D users.

[sudden torrent]

7950

[steady lance]

Interesting.

[sudden torrent]

Most people that need that many threads can get the better one

[steady lance]

Fair.

The 7900x3D seems to be a very niche group of users then.

[sudden torrent]

It is extremely niche

[steady lance]

I wonder if my current 7950x3D has more ram oc potential.

[sudden torrent]

I seem to top out at 6200

6400 boots but it's not stable

[steady lance]

I remember being able to do 6400 on my current sample. Haven't tried 6600 though I doubt that will be stable.

[sudden torrent]

2133 fclk is good too

Any higher and performance doesn't change so it's probably got micro instability

2200 boots but no improvement

2233 no post

[steady lance]

Fclk is ram or CPU limited?

[sudden torrent]

On 7000 it's CPU

[steady lance]

Gotcha.

In order for desync to beat sync I need my ram to do 7200+ right?

[sudden torrent]

It's permanently desynced from mclk but there's a slight advantage to running 3:2.
The issue going higher is desync from uclk, and yeah about there is where it needs to be

[steady lance]

Back.

So to make sure I understand and recall

On 7000 series AM5:

• MCLK = UCLK
• FCLK is desynced so higher is better

[sudden torrent]

Ideally yes

[steady lance]

And by running the ram in I think it was called G2 the ram is then desynced from the MCLK right?

[sudden torrent]

DIV2 mode yeah

It's similar to Intel Gear 4

[steady lance]

Noted.

And to overcome the latency penalty I'd need my ram to get to 7200+ mhz range?

[sudden torrent]

Give or take a bit

[steady lance]

Gotcha.

Wondering because I am going to start ram oc.

Debating if I wanna try deyncing first or try with synced first.

What are your suggestions? I think I have Hynix A.

[sudden torrent]

If you want to see what your max clocks are then you want to desync

It's more likely that you'll want to run synced so that's a good way to end it

[steady lance]

While I do want to see what the max clock is, my goal is to maximize performance.

[sudden torrent]

If you can get it to 7600 then you might as well leave it there

When AMD can get high like that it's more stable than Intel at the same speeds too

[steady lance]

Gotcha. What should my max voltages be on Hynix A die? I think it's A because I previously ran 1.6v through it without it becoming unstable or overheating.

[sudden torrent]

Yeah that's fine

Max will vary by stick and cooling

[steady lance]

Other then DRAM I am trying to remember the other 2 voltages. I don't want to fry or damage something.

[sudden torrent]

VDD, VDDIO, and VDDQ

[steady lance]

VDDq should be equal or less then VDD if I recall right?

[sudden torrent]

Yes

[steady lance]

What should I max my voltage at for VDDIO and VDDQ?

[sudden torrent]

Hard to say, maybe 1.5V but if you go too high it'll get unstable

[steady lance]

Would turning any of those up to high hurt the cou?

[sudden torrent]

VDD and VDDQ are on the PMIC located on the memory sticks, VDDIO goes to the IO die on the CPU.

[steady lance]

So VDD and VDDQ will only harm the ram whereas vddio will harm the cpu if I shove to much voltage then?

[sudden torrent]

Yes

[steady lance]

.... I am wondering if I may've fried my previous 7950x3D.

My motherboard by default sets VDDQ and VDDIO = to VDD.

Which was previously at 1.6v.

[steady lance]

Alright I think I am gonna start doing ram PC this weekend.

I am pretty confident in my cpu oc now.

[manic stump]

Anybody know about qfan tuning?

[steady lance]

What did you guys recommend for testing FCLK stability?

[dull ginkgo]

Y cruncher might be good

[steady lance]

Gotcha.

I am gonna try testing my ram desynced when I get a chance.

If it can't hit like 7400mhz then I will probably switch back to synced

[proven canopy]

3dm11 physics

[steady lance]

HOLY!

@dull ginkgo, @sudden torrent, @modern walrus, I just successfully booted into windows at 6600mhz! synced.

No artifacting or anything of that sort so far.

What do you think the likelyhood of me booting in at 6800mhz is?

[sudden torrent]

Near zero

And you are desynced

Which on Ryzen 7000 refers to mclk:uclk being 2:1

So if you have A die you could possibly boot higher while desynced like that

[steady lance]

Oh wait I was desynced?

Well darn... I thought I was synced.

[sudden torrent]

Afraid not

[steady lance]

Wish I saw that.

7000+ is where desync is better then synced?

Cause I tried 7200 desycn before this run which did not boot.

[sudden torrent]

That's where the latency penalty matters less at least

[sterile flame]

hey overclockers

[steady lance]

At what point does desync beat sync?

[sterile flame]

what voltage can b die do